Model predictive control of air pollution control processes

[0001] The invention relates generally to process control. The invention particularly relates to techniques for increasing control of methods such as those used in controlling air pollution. Examples of such methods include, but are not limited to, wet and dry flue gas desulfurization (WFGD / DFGD), nitrogen oxides removal by selective catalytic reduction (SCR). selective catalytic reduction) and removal of solid particles by electrostatic precipitation (ESP). US 2004/0102890 A1 discloses a method of designing the operation and control of an aircraft engine in the form of a gas turbine. EP 1 382 905 A1 discloses a method for determining a kiln feed combination.

Background

Wet Flue Gas Desulphurization:

[0002] As mentioned, there are several ways to control air pollution that form the basis for discussion; the focus will be on the WFGD process. The WFGD process is the most commonly used method for SO2 removal from flue gases in the energy sector. FIG. 1 is a block diagram depicting a wet flue gas desulphurization (WFGD) subsystem for removing SO2 from contaminated flue gas such as those produced by coal-fired power plant systems and producing commercial-grade by-product, for example having features that will allow it to be removed at a minimized disposal cost or having features that enable it to be sold for commercial purposes.

[0003] In the United States of America, the current preferred by-product of WFGD is commercial grade gypsum having a relatively high quality (95 +% purity) suitable for use in wall panels, which in turn are used for the construction of apartments and offices. High-quality commercial grade gypsum (~ 92%) is also currently the preferred by-product of WFGD in the European Union and Asia, but is more often produced for use in cement and as a fertilizer. However, if there is a reduction in the market for high-quality gypsum, the quality of commercial grade gypsum produced as a by-product of WFGD should be lowered to meet lower quality requirements for disposal at minimal cost. Therefore, disposal costs can be minimized if, for example, the quality of the plaster is suitable for landfills at human settlements or for backfilling areas from which coal used for power generation has been selected.

[0004] As shown in Fig. 1, SO2 containing contaminated exhaust gas is emitted by a boiler or economizer (not shown) of coal-fired power plant system 110 to the air pollution control system (APC) (120).

Typically, contaminated flue gas 112 introduced into APC 120 is not only contaminated with SO2 but also contains other so-called contaminants such as NOx and particulates. Prior to processing by the WFGD subsystem, contaminated exhaust 112 entering the APC 120 is first directed to other APC 122 subsystems to remove NOx and particulate matter from contaminated exhaust 112. For example, contaminated exhaust gas can be treated with the Selective Catalytic Reduction (SCR) subsystem (not shown) for NOx removal and with an electrostatic filter (EPS) system (not shown) or a filter (not shown) for removal of solid particles.

[0005] Exhaust gas containing SO2 114 separated from the APC 122 subsystems is directed to the WFGD 130 subsystem. Exhaust gas containing SO2 114 is processed by the absorption tower 132. As will be understood by those skilled in the art, SO2 in exhaust gas 114 has a high acid concentration. Accordingly, the absorption tower 132 operates to contact SO2 containing exhaust gas 114 with liquid suspension 148 having a higher pH than exhaust gas 114.

[0006] It is understood that most conventional WFGD subsystems contain a WFGD processing unit of the type shown in Fig. 1. This is true for many reasons. For example, as is well known in the art, WFGD processing units having spray absorbers have some desirable process properties for the WFGD process. However, WFGD processing units having other absorption / oxidation equipment configurations may, if desired, be used instead of the unit shown in Fig. 1 and also provide similar flue gas desulphurization functionality and achieve similar benefits from the advanced process control improvement shown in the application. For clarity and brevity, the discussion will refer to the commonly used spray tower shown in Fig. 1, but it should be noted that the concepts presented can be applied to other WFGD configurations.

[0007] During processing in the countercurrent absorber tower 132, SO2 in flue gas 114 will react with a slurry rich in calcium carbonate (limestone and water) 148 to form calcium sulfite, which is essentially salt, and thus can remove SO2 from flue gas 114. Flue gas purified from SO2 116 are secreted from the absorber tower 132, into the exhaust chimney 117 or into a downstream processing device (not shown). The resulting transformed suspension 144 is directed to crystallizer 134, where the salt crystallizes. Crystallizer 134 and absorber 132 are usually located in one tower without physical separation between them - because there are different activities (absorption in the gas phase and crystallization in the warm phase), these two activities occur in the same process tank. Hence, the gypsum slurry 146 that contains the crystallized salt is directed from the crystallizer 134 to the drainage device 136. In addition, the slurry for recycling 148, which may or may not contain the same concentration of crystallized salts as the gypsum slurry 146 is directed from the crystallizer 134 through pumps 133 and back to absorption tower 132 to continue the absorption cycle.

[0008] A blower 150 compresses the surrounding air 152 to form oxidizing air 154 for the crystallizer 134. The oxidizing air 154 is mixed with the slurry in the crystallizer

-3134 oxidizing calcium sulfite to calcium sulfate. Each calcium sulfate molecule binds to two water molecules to form a compound that is commonly referred to as gypsum 160. As shown, gypsum 160 is removed from the WFGD 130 processing unit and sold, for example, to building grade wall manufacturers.

[0009] The recovered water 167 from the drainage device 136 is directed to the mixer / pump 140, where it is combined with freshly ground limestone 174 from mill 170 to form a limestone slurry. Since part of the process water is lost in both the gypsum 160 and waste stream 169, additional fresh water 162 is added from the fresh water source 164 to maintain the density of the lime slurry. In addition, waste such as ash is removed from the processing unit 130 by means of waste stream 169. This waste may, for example, be directed to an ash container or otherwise disposed of.

[0010] In summary, SO2 in the exhaust gas containing SO2 114 is absorbed by the suspension 148 in the contact area with the suspension in the absorption tower 132, and then crystallized and oxidized in the crystallizer 134 and dehydrated in a dewatering device 136 to form the desired by-product of the process, which is example of commercial grade 160 gypsum. Exhaust gas containing SO2 114 passes through absorption tower 132 in a few seconds. The total crystallization of the salt in the sludge 144 transformed by the crystallizer 134 may require from 8 hours to over 20 hours. Thus, the crystallizer 134 has a large volume that serves as a reservoir for the crystallization slurry. The recycling slurry 148 is pumped back to the top of the absorber to recover additional SO2.

[0011] As shown, the slurry 148 is fed to the top of the absorption tower 132. The tower 132 typically includes multiple levels of spray nozzles for introducing the slurry 148 into the tower 132. The absorber 132 operates in a countercurrent configuration: a stream of sprayed slurry flows down the absorber and contacts with SO2 114 flue gas flowing upwards that was introduced into the bottom of the absorption tower.

[0012] Fresh limestone 172 from a limestone source 176 is first ground in a mill 170 (usually in a ball mill) and then mixed with recovered water 167 and fresh water / make-up 162 in a mixer 140, forming a lime slurry 141 The flow of ground limestone 174 and water 162 through valve 163 to mixer / tank 140 is controlled to maintain sufficient supply of fresh lime slurry 141 in mixer / tank 140. The flow of the lime slurry 141 to the crystallizer 134 is adjusted to maintain the proper pH of the slurry 148, which in turn controls the amount of SO2 removed from the exhaust gas 114. WFGD processing typically achieves 92-97% SO2 removal from the exhaust gas, however, for a specialist in this it will be apparent to the art that the use of certain techniques and the addition of organic acids to the suspension can increase SO2 removal by more than 97%.

[0013] As described above, conventional WFGD subsystems transform a suspension. Although in the production of gypsum, a certain amount of sewage and other waste usually accumulates

-4 waste, water is recovered whenever possible and used to replenish fresh limestone slurry, thus minimizing waste and costs that would be incurred in purifying process water.

[0014] It will be understood that since limestone is readily available in large quantities in most places, it is widely used as a reagent in the desulfurization of carbon gas. Instead of limestone, however, other reagents such as calcium oxide or sodium compounds may alternatively be used. These other reagents are usually more expensive and are not currently competitive with the limestone reagent. However, with very minor modifications to the mixer 140 and top reactant source, existing WFGD for limestone can be utilized using calcium oxide or a sodium compound. In fact, most WFGD systems contain a limestone stock subsystem so that WFGD can operate if limestone supply problems occur and / or problems with extended mill 170 maintenance.

[0015] Fig. 2 shows further details of the WFGD subsystem shown in Fig. 1. As shown, the drainage device 136 may include both a primary drainage device 136A and a secondary drainage device 136B. The primary drainage device 136A preferably includes hydrocyclones for separating gypsum and water. The secondary drainage device 136B preferably includes a belt dryer for drying the plaster. As previously discussed, the exhaust gas 114 enters the absorber 132, usually from the side and upwardly through the mist of a lime slurry that is atomized in the upper part of the absorption tower. Before leaving the absorber, the exhaust gas is passed through a mist eliminator (MEz) (not shown), which is located at the top of the absorber 132; the mist eliminator removes entrained liquid and solid particles from the exhaust stream. To eliminate solid particles from the mist eliminator, the mist eliminator is washed with ME 200 wash water. As will be seen, the ME 200 wash maintains the cleanliness of the ME in the absorption tower 132 with water from a fresh water source 164. ME 200 wash water is the purest water entering the WFGD 130 subsystem.

[0016] As noted above, the limestone suspension absorbs a large percentage of SO2 (e.g., 92-97%) from the exhaust gas that flows through the absorption tower 132. After SO2 absorption, the semi-fluid suspension falls into the crystallizer 134. In practical implementation, the absorption tower 132 and the crystallizer 134 is often in a single uniform structure, the absorption tower being located directly above the crystallizer in the structure. In such embodiments, the semi-liquid suspension simply falls into a uniform structure for crystallization.

[0017] The limestone slurry reacts with SO2 to form gypsum (dehydrated calcium sulfate) in the crystallizer 134. As mentioned earlier, the forced circulation of compressed oxidation air 154 is used to promote oxidation, which occurs in the following reaction:

SO2 + CaCO3 +% O2 + 2 H2O CaSO4-2H2O + CO2 (1)

The oxidizing air 154 is forced into the crystallizer 134 via a blower 150. The oxidizing air provides the additional amount of oxygen needed to convert calcium sulfite to calcium sulfate.

[0018] The absorption tower 132 is used to achieve close exhaust / liquid suspension contact needed to achieve the high removal efficiency required by environmental specifications. Countercurrent spray absorption towers provide especially desirable properties for WFGD limestone-gypsum processing: they are inherently reliable, have a lower propensity to clog than other components of WFGD tower-based processing units, cause low pressure drop, and are cost-effective in terms of both capital and cost operating.

[0019] As shown in Fig. 2, the water source 164 typically includes a water tank 164A for storing enough fresh water. Typically, one or more pumps 164B are also included for pumping ME 200 fluid into the absorption tower 132 and one or more pumps 164C for pumping fresh water stream 162 into mixer 140. The mixer 140 includes a mixing tank 140A and another slurry pump 140B to transfer fresh lime slurry 141 to the crystallizer 134. One or more additional very large slurry pumps 133 (see Fig. 1) is required to lift the slurry 148 from the crystallizer 134 to many spray levels in the absorption tower 132.

[0020] As will be described below, usually the lime slurry 148 enters the absorption tower 132, through spray nozzles (not shown), located at different levels of the absorption tower 132. At full load, most WFGD subsystems operate with at least one spare slurry pump 133 At reduced loads, it is often possible to achieve the required SO2 removal efficiency with a reduced number of slurry pumps 133. There is a significant economic incentive to reduce the pumping load of slurry pumps 133. These pumps are one of the largest pumps in the world and are powered by electricity, which could otherwise be sold directly to the power grid (parasitic supply).

[0021] The gypsum 160 is separated from the liquid in the gypsum slurry 146 in the basic drainage device 136A, usually using a hydrocyclone. The top product in the hydrocyclone and / or one or more other components of the 136A primary drainage device contains low levels of solids. As shown in Fig. 2, this upper slurry 146A is recycled to the crystallizer 134. The recovered water 167 is sent back to the mixer 140 to produce fresh lime slurry. Other waste 168 is usually directed from primary drainage device 136A to ash container 210. Lower slurry 202 is directed to secondary drainage device 136B, which often takes the form of a band filter, where it is dried to produce a byproduct in the form of plaster 160. Recovered water 167 secondary drainage 136B is again directed to mixer / pump 140. As shown in Fig. 1, material samples or other gypsum samples 161 are taken and analyzed, usually every few hours, to determine gypsum purity 160. No direct method of measuring gypsum purity is widely available.

[0022] As shown in Fig. 1, a proportional-integral-differential (PID) controller is commonly used to control the operation of the WFDG subsystem

-6integral derivative) 180 in conjunction with the feedforward controller (FF) 190. In the past, PID controllers controlled pneumatic analog control functions. Today, PID controllers manage digital control functions using mathematical formulas. The purpose of the FF 190 / PID 180 regulator is to control the pH of the suspension based on established relationships. For example, there may be a fixed relationship between the position of the valve 199 shown in Fig. 1 and the measured pH of the suspension 148 flowing from the crystallizer 134 to the absorption tower 132. In this case, the valve 199 is controlled such that the pH of the suspension 148 corresponds to the desired value 186, often referred to as the setpoint (SP).

[0023] The FF 190 / PID 180 controller will adjust the flow of lime slurry 141 through valve 199, based on the setpoint pH value, so as to increase or decrease the slurry pH 148 value, as measured by the pH sensor 182. As will be understood, this will be achieved by an FF / PID controller that sends the appropriate control signals 181 and 191 that result in valve matching instructions, shown as SP 196 flow control, to a flow controller, which is preferably part of valve 199. In response to the flow control SP 196, the flow controller in turn directs the adjustment of the valve 199 to modify the flow of lime slurry 141 from the mixer / pump 140 to the crystallizer 134.

[0024] The example shows pH control using a combination of FF 190 controller and PID 180 controller. Some installations will not include the FF 190 controller.

[0025] In the example, the PID controller 180 generates the PID control signal 181 by processing the measured pH value of the suspension 183 received from the pH sensor 182, according to a limestone flow control algorithm representing the established relationship between the measured pH value 183 of the suspension 148 flowing from the crystallizer 134 to the absorption tower 132. The algorithm is usually stored in the PID 180 controller, but it is not mandatory. Control signal 181 may reflect, for example, the valve setpoint (VSP for valve setpoint) for valve 199 or for the measured value setpoint (MVSP) for flow of ground limestone slurry coming out of valve 199.

[0026] As is well known in the art, the algorithm used by the PID 180 controller has a proportional element, an integral element and a derivative element. The PID 180 controller first calculates the difference between the desired SP and the measured value to determine the error. The PID controller then applies the error to the proportional element of the algorithm, which is the regulated constant for the PID controller or for each of the PID controllers if multiple PID controllers are used in the WFGD subsystem. The PID controller usually multiplies the matching factor or process gain by an error to obtain a proportional function to adjust valve 199.

[0027] However, if the PID controller 180 does not have the correct value for the process fit or gain factor, or if the process conditions are variable, the proportional function will be inaccurate. Because of this inaccuracy, VSP or MVSP

7 generated by the PID 180 controller will in fact be offset from those corresponding to the desired SP. Therefore, the PID controller applies a cumulative error over time using an integration element. The integration element is a factor of time. Here again, the PID 180 controller multiplies the fit factor or the process gain by the cumulative error to eliminate the shift.

[0028] Let us now proceed to the derivative element. The derivative element is an acceleration factor associated with a constant change. In practice, the derivative element is rarely used in PID controllers used to control WFGD processes. This is because the use of a derivative element is not particularly advantageous for this type of control application. Therefore, most controllers used in WFGD subsystems are in fact PI controllers. However, those skilled in the art know that, if desired, the PID 180 controller can easily be configured with the required logic to use the derivative element in a conventional manner.

In summary, there are three constant adjustments that can be used by standard PID controllers to control process values, such as the pH of the recycling slurry 148, fed into the absorption tower 132 to a set value and such as the flow of fresh limestone slurry 141 to the crystallizer 134. Regardless of what setpoint is used, such as the SO2 value remaining in the exhaust gas 116 coming out of the absorption tower 132, it is always determined in terms of process values ​​and not in terms of the desired result. In other words, the setpoint is defined in terms of the process and it is necessary that the controlled value for the process is directly measurable so that it can be controlled by the PID controller. Although the exact form of the algorithm may vary depending on the equipment supplier, the basic PID control algorithm has been used in the processing industry for over 75 years.

[0030] Referring again to Figs. 1 and 2, based on the instructions received from PID controller 180 and FF controller 190, the flow controller generates a signal that causes the valve 199 to open or close, thereby increasing or reducing the flow of ground limestone slurry. 141. The flow controller continues to control the valve control until valve 199 opens or closes to match the VSP or the measured amount of lime slurry 141 flowing from the 1992 valve reaches MVSP.

[0031] In the example standard WFGD control described above, the pH of the slurry 148 is controlled based on the desired setpoint pH 186. To perform the PID control 180, it receives the process value, i.e. the measured pH value 183 of the slurry 148 from sensor 182. The PID controller 180 processes the process value for generating instruction 181 for valve 199 to adjust the flow of fresh limestone slurry 141, which has a higher pH than the slurry in crystallizer 144 from mixer / tank 140, and thus adjust the pH of the slurry 148. If instructions 181 cause further opening the valve 199, more slurry 141 will flow from the mixer 140 and into the crystallizer 134, which will increase the pH of the slurry 148. On the other hand, if instructions 181 close the valve 199, less lime slurry 141 will be

Flow from the mixer 140 and thus to the crystallizer 134, which will cause the pH of the suspension 148 to drop.

[0032] In addition, the WFGD subsystem may include a forward feedback loop that is performed using a forward feedback device 190 to ensure stable operation. As shown in Fig. 1, the SO2 concentration value 189 in the exhaust gas 114 introduced into the absorption tower 132 is measured by sensor 188 and fed into the forward coupling device 190. Many WFGD systems that include an FF control element can combine the SO2 concentration in the input 189 with the load measurement of the generator from the Power Generation System 110 to determine the amount of SO2 introduced, not just the concentration, and then use the amount of SO2 introduced as the value entered into FF 190. The forward coupling device 190 serves as a proportional element with time delay.

[0033] In this exemplary embodiment discussed, the forward coupling device 190 receives the SO2 measurement sequence 189 from sensor 188. The forward coupling device 190 compares the currently obtained concentration value with the concentration value obtained immediately before the currently obtained value. If the forward feedback device 190 determines that there has been a change in measured SO2 concentrations, e.g. from 1000 to 1200 parts per million, there is a control system configuration for smoothing the step function, thus avoiding a sudden change in operations.

[0034] The forward loop significantly improves the stability of normal operations because the relationship between the pH of the slurry 148 and the amount of limestone slurry flowing to the crystallizer 134 is strongly non-linear, and the PID controller 180 is effectively a linear regulator. Thus, without a forward loop, it is very difficult to provide PID 180 with adequate control over a wide pH range with the same matching constants.

[0035] By controlling the pH of the slurry 148, the PID controller 180 controls the effects of both SO2 removal from the SO2 containing flue gas 114 and the quality of the by-product gypsum 160 produced by the WFGD subsystem. Increasing the pH of the suspension by increasing the flow of fresh limestone suspension 141 increases the amount of SO2 removed from flue gas containing SO2 114. On the other hand, increasing the flow of limestone slurry 141, and therefore the pH of the slurry 148, slows down the oxidation of SO2 after absorption, and thus the conversion of calcium sulfite into sulfate, which in turn results in lower quality of the resulting gypsum 160.

[0036] There are therefore conflicting objectives for controlling SO2 removal from SO2-containing exhaust gas 114 and maintaining the required quality of gypsum by-product 160. Thus, there may be a conflict between meeting SO2 emission requirements and gypsum quality requirements.

[0037] Fig. 3 gives details of further embodiments of the WFGD subsystem described in relation to fFig. 1 and 2. As shown, exhaust gas containing SO2 114 enters the lower portion of the absorption tower 132 through the opening 310, and exhaust gas without SO2 116 exits through the upper portion of the absorption tower 132 through the opening 312. In this exemplary standard embodiment, a countercurrent tower is shown

-Absorbent with multiple levels of suspension spraying. As shown, the ME 200 fluid is introduced into the absorption tower 132 and is dispersed by fluid atomizers (not shown).

[0038] A number of nozzles for the suspension tower 306A, 306B and 306C are also provided, each having a 308A, 308B or 308C slurry atomizer that atomizes the slurry into the exhaust gas for SO2 absorption. The slurry 148 is pumped from the crystallizer 134 shown in Fig. 1 by multiple pumps 133A, 133B and 133C, each of which pumps the slurry up to a different level of the slurry nozzles 306A, 306B or 306C. It should be understood that although 3 different levels of slurry nozzles and sprayers are shown, the number of nozzles and sprayers will vary depending on the specific embodiment.

[0039] The ratio of the flow rate of the liquid suspension 148 entering the absorber 132 to the flow rate of the exhaust gas 116 leaving the absorber 132 is usually denoted as L / G. L / G is one of the key design parameters in WFGD subsystems.

[0040] The exhaust gas flow rate 116 (saturated with steam), designated as G, is a function of the exhaust gas introduced 112 from the power generating system 110 before the WFGD 130 processing unit. Therefore, G is not and cannot be controlled, but must be included in the treatment WFGD. Therefore, to affect L / G, you must adjust "L". Adjusting the number of slurry pumps and "ordering" these slurry pumps controls the flow rate of liquid slurry 148 to the absorption tower WFGD 132, designated as L. For example, if only two pumps are started, starting the pumps at the top two spray levels compared to the pumps at the top and bottom of the spray levels will cause a different "L".

[0041] "L" can be adjusted by controlling the operation of the slurry pumps 133A, 133B and 133C. Individual pumps can be turned on and off to adjust the flow rate of liquid suspension 148 to the absorption tower 132 and the effective height at which the liquid suspension 148 is introduced into the absorption tower. The higher the suspension is introduced into the tower, the more time it has to contact the exhaust gas, which results in greater SO2 removal, but this additional SO2 removal occurs at the expense of increasing the energy consumption for pumping the suspension to a higher spray level. It will be understood that the larger the number of pumps, the greater the granularity of such control.

[0042] The 133A-133C pumps, which are very large rotary devices, can be turned on and off automatically or manually. In the USA, these pumps are most often controlled manually by the subsystem operator. In Europe, the automation of starting / stopping rotary devices such as 133A-133C pumps is more common.

[0043] If the exhaust gas flow rate 114 entering the WFGD processing unit 130 is modified due to a change in the operation of the power generating system 110, the operator of the WFGD subsystem may control the operation of one or more 133A-133C pumps. For example, if the exhaust gas flow falls to 50% of the assumed load, the operator or special controller in the control system can stop one or

-10 more pumps that pump the slurry to the nozzle at spray levels at one or more spray levels.

[0044] Although not shown in Fig. 3, it is known that there are often additional spray levels with associated slurry pumps and nozzles for use during maintenance of another pump or other slurry nozzles and / or slurry sprayers associated with basal spray levels. The addition of this additional spray level adds the capital cost to the absorption tower and thus to the subsystem. Therefore, some WFGD owners choose to eliminate additional spray levels and avoid additional capital costs and instead add organic acids to the slurry to increase its ability to absorb and remove SO2 from flue gas during maintenance periods. However, these add-ons are often costly and, therefore, their use will increase operating costs, which may, over time, outweigh capital cost savings.

[0045] As indicated in Equation 1 above, for SO2 absorption, a chemical reaction must occur between SO2 in the exhaust gas and limestone in the suspension. The result of a chemical reaction in the absorber is the formation of calcium sulfite. In crystallizer 134, calcium sulfite is oxidized to form calcium sulfate (gypsum). Oxygen is consumed during this chemical reaction. To provide sufficient oxygen and increase the reaction rate, additional O2 is added by blowing compressed air 154 into the liquid suspension in the crystallizer 134.

[0046] In particular, as shown in Fig. 1, the surrounding air 152 is compressed to produce compressed air 154 and forced into the crystallizer 134 by means of a blower, for example a fan 150 for oxidizing calcium sulfite in the recycling suspension 148 which is recycled from the crystallizer 134 to absorber 132 and the gypsum slurry 146 is transferred to drainage system 136 for further processing. To facilitate the control of the oxidative air flow 154, the blower 150 may have a speed or load control mechanism.

[0047] Preferably, the suspension in the crystallizer 134 has excess oxygen. However, there is an upper limit to the amount of oxygen that can be absorbed or suspended. If the O2 level in the suspension becomes too low, the chemical oxidation of CaSO3 to CaSO4 in the suspension will stop. When this happens, it is commonly referred to as limestone masking. When limestone masking occurs, limestone ceases to dissolve in the suspension solution, and SO2 removal can be significantly reduced. The presence of trace amounts of certain minerals can significantly slow down the oxidation of calcium sulfite and / or the dissolution of limestone causing masking of limestone.

[0048] Because the amount of O2 that dissolves in the suspension is not a measurable parameter, the suspension in standard WFGD subsystems may have an O2 deficiency if proper precautions are not taken. This is especially important in the summer months, when the higher temperature of the ambient air reduces the density of the ambient air 152 and reduces the amount of oxidizing air 154 that can be injected into the

- crystallizer 134 by blower 150 at maximum speed or load. In addition, if the amount of SO2 removed from the exhaust stream increases significantly, a sufficient amount of additional O2 is required to oxidize the SO2. Thus, a lack of O2 may effectively occur in the slurry due to the increase in the SO2 flow in the processing unit

WFGD.

[0049] It is necessary to introduce compressed air 154 in sufficient quantity, within the assumed ratios, for the oxidation of absorbed SO2. It is possible to adjust the speed or load of the blower 150 and it is desirable to screw the blower 150 at lower SO2 loads and / or during periods of colder ambient temperature due to energy saving. When the blower 150 reaches the maximum load or all of the O2 in the unregulated blower is used, the incremental increase in SO2 cannot be oxidized. At maximum load or without a blower speed control 150 that closely monitors SO2 removal, O2 deficiency can be caused in the crystallizer 134.

[0050] However, since the measurement of O2 in suspension is not possible, the concentration of O2 in suspension is not used as a limitation of the standard activities of the WFGD subsystem. Thus, it is not possible to accurately monitor the formation of O2 deficiency in crystallizer 134. Therefore, in the best case, operators will assume that an O2 deficiency is formed in the suspension if a decrease in the quality of the by-product 160 gypsum is noticeable and will use their best judgment to control blower speed or charge 150 and / or reduce SO2 absorption efficiency to balance O2 injected into the suspension, with absorbed SO2, which must be oxidized. Thus, in standard WFGD subsystems the O2 balance injected into the suspension with SO2, which should be absorbed from the exhaust gas, most preferably depends on the operator's assessment.

[0051] In summary, standard control of large WFGD subsystems for utility applications is usually performed without a distributed control system (DCS) and generally consists of a logic two-state controller as well as an FF / PID feedback control loop. Controlled parameters are limited to the level of pH in the suspension, L / G ratio and forced flow of oxidizing air.

[0052] The pH value must be kept within a certain range to ensure high solubility of SO2 (i.e. SO2 removal efficiency), high quality (purity) gypsum and to prevent scale build-up. The operating pH range depends on the equipment and operating conditions. The PH is controlled by adjusting the flow of fresh limestone suspension 141 to the crystallizer 134. The limestone suspension flow control is based on the measured suspension pH detected by the sensor. In a typical embodiment, the PID controller and, optionally, the FF controller located in the DCS are arranged sequentially towards the limestone suspension flow controller. The standard / default PID algorithm is used for use in pH control.

The liquid to gas ratio (L / G) is the ratio of the liquid suspension 148 flowing to the absorption tower 132 to the exhaust gas flow 114. For a given set of subsystem variables to achieve the desired SO2 absorption, based on the solubility of SO2 in the liquid suspension 148, a minimum L / G ratio is required. The L / G ratio changes when the exhaust gas flow 114 changes or when the flow of liquid suspension 148 changes, which usually occurs when the suspension pumps 133 are turned on or off.

[0054] Oxidation of sodium sulfite to form sodium sulfate, i.e. gypsum, is increased by forced oxidation, with additional oxygen in the crystallizer reaction tank 134. Additional oxygen is introduced by blowing air into the suspension solution in the crystallizer 134. Insufficient oxidation may occur sulfite-limestone masking, which results in poor plaster quality and potentially lower SO2 removal efficiency and high chemical oxygen demand (COD) chemical oxygen demand) in wastewater.

[0055] The standard WFGD process control scheme consists of standard blocks with independent instead of integrated targets. Currently, the operator, in consultation with the engineering staff, must strive to ensure optimal overall process control. To ensure this control, the operator must consider various goals and restrictions.

[0056] Minimized Operating Costs for WFGD - power plants only work to ensure profit for their owners. Thus, it is beneficial to operate the WFGD subsystem at the lowest applicable costs, while respecting the limitations of the process, control and quality of the by-product and the business environment.

[0057] Maximizing SO2 Removal Efficiency - Air purity regulations determine the requirements for SO2 removal. WFGD subsystems should work to remove SO2 as efficiently as is desirable from the point of view of process, control and quality by-product and business environment restrictions.

[0058] Compliance with Specification of Gypsum Quality - the sale of gypsum as a by-product reduces operating costs and depends largely on the purity of the by-product that meets the desired specification. WFGD subsystems should be operated to produce by-product in the form of gypsum of appropriate quality from the point of view of restrictions on the process, control and quality of the by-product and the business environment.

[0059] Limestone Masking Prevention - fluctuations and changes in the sulfur content of the fuel can cause deviations of the SO2 content in the exhaust gas 114. Without proper correction for compensation, this can lead to high sulfite concentration in the suspension, which in turn causes masking of limestone, lower SO2 removal efficiency in the absorption tower 132, low gypsum quality and high chemical oxygen demand (COD) in wastewater. WFGD subsystems should operate to prevent limestone masking in the light of process limitations.

[0060] In a typical work sequence, the WFGD subsystem operator determines the set values ​​for the WFGD process to balance these competing goals and constraints, based on standard working procedures and knowledge of the WFGD process. Settings often include the pH and operating condition of the slurry pumps 133 and the oxidizing air blower 150.

[0061] There are complex interactions and dynamics in the WFGD process; as a result, the operator selects conservative operating parameters so that the WFGD subsystem can meet / exceed the stringent restrictions on SO2 removal and gypsum purity. In making these conservative choices, the operator often, if not always, gives up the operation at minimal cost.

[0062] For example, Fig. 4 shows SO2 removal efficiency and gypsum purity as a function of pH. As the pH increases, the SO2 removal efficiency increases, but the gypsum purity decreases. Because the operator is interested in both improving SO2 removal efficiency and gypsum purity, the operator must determine the pH setting, which is a compromise between these two competing goals.

[0063] In addition, in most cases, the operator must meet a guaranteed level of plaster purity such as 95% purity. Due to the complexity of the relationships shown in Fig. 4, the lack of direct measurement of gypsum purity in line, the long-term dynamics of gypsum crystallization and random changes in operations, the operator often decides to introduce a pH setting that will guarantee that the level of gypsum purity is higher than specified limitation in all circumstances. However, guaranteeing the purity of the plaster, the operator often gives up SO2 removal efficiency. For example, based on the graph of Fig. 4, the operator can select a pH of 5.4 to guarantee a 1% margin over a 95% reduction in plaster purity. However, selecting this setting for pH, the operator gives up 3% SO2 removal efficiency.

[0064] The operator faces similar trade-offs when the SO2 load, i.e. exhaust gas flow 114 drops from full to half. At some point during this transition, it may be beneficial to turn off one or more slurry pumps 133 for energy saving, because continuous pump operation can only provide slightly better SO2 removal efficiency. However, because the relationship between energy costs and SO2 removal efficiency is not well understood by most operators, operators typically take a conservative approach. Using this approach, operators may not regulate a number of slurry pumps 133, although it would be more beneficial to turn off one or more slurry pumps 133.

[0065] It is also known that many emission regulations introduce both temporary emission limits and certain types of emission limits in the form of moving average. The moving average of emission limits is the average of instantaneous values ​​of emissions in a moving or rolling time window. The time window can be as short as 1 hour and as long as 1 year. Some typical time windows are 1 hour, 3 hours, 8 hours, 24 hours, 1 month and 1 year. To allow dynamic process deviations, instantaneous emission limit values ​​are usually higher than the average limit

-14kroczącej. However, continuous operation at the momentary emission limit value will violate the moving average limit.

[0066] By default, PID 180 controls emissions to an instantaneous allowable amount, which is relatively simple. To this end, the operational limitation for the process, i.e. the instantaneous value, is set within the limits of the actual legal emission limitation, thus providing a safety margin.

[0067] On the other hand, emission control to reduce moving average is more complex. The time window for the moving average is constantly advancing. Therefore, at any time several time windows are active, covering one window from a given period from a given moment backwards and another window covering a period from a given moment for a certain period forward. By default, the operator attempts to regulate emissions to a moving average limit by simply maintaining a sufficient margin between the operating limit specified in PID 180 to limit the instantaneous and actual legal emission limit, or by using the operator's assessment to set limits due to the moving average limit. Under no circumstances is there a clear regulation of moving average emissions, and therefore there is no way to ensure compliance with a moving average limit or to prevent costly excessive compliance.

Selective Catalytic Reduction System:

[0068] Returning to another exemplary method of controlling air pollution, a selective catalytic reduction (SCR) system for NOx removal, similar operational challenges can be distinguished. An overview of the SCR method is shown in Fig. 20.

[0069] The following discussion of the method is from the document "Control of Nitrogen Oxide Emissions: Selective Catalytic Reduction ISCR)", Topical Report Number 9, Clean Coal Technology, US Dept. of Energy, 1997:

Discussion of the method [0070] NOx, which consists mainly of NO with smaller amounts of NO2, is converted to nitrogen by reaction with NH3 in the presence of a catalyst and in the presence of oxygen. The small SO2 fraction produced in the boiler by oxidation of sulfur in coal is oxidized to sulfur trioxide (SO3) in the presence of an SCR catalyst. In addition, side reactions may cause the formation of undesirable by-products: ammonium sulfate, (NH4) 2SO4 and NH4HSO4 ammonium hydrogen sulfate. There are complex relationships that determine the formation of these by-products, but they can only be minimized by proper control of process conditions.

Ammonia residue [0071] The unreacted ammonia in the flue gas downstream of the SCR reactor is referred to as the NH3 residue. It is important to keep the ammonia residue below 5 ppm, preferably 2-3 ppm, to minimize the formation of (NH4) 2SO4 and NH4HSO4, which may

-15 cause clogging and corrosion of equipment behind the reactor. This is a larger problem for high sulfur carbons, caused by higher SO3 levels, resulting from both higher initial SO3 concentrations due to the sulfur content of the fuel and SO2 oxidation in the SCR reactor.

Operating temperature [0072] The cost of the catalyst is 15-20% of the capital cost of the SCR device; therefore, it is important to work at the highest temperature possible to maximize the volumetric speed and thus minimize the volume of the catalyst. At the same time, it is necessary to minimize the oxidation rate of SO2 to SO3, which is more temperature sensitive than the SCR reaction. The optimum operating temperature for the SCR process using titanium catalyst and vanadium oxide is about 650-750 ° F. Most installations use an economizer bypass to deliver flue gas to the reactor at the desired temperature during periods when flue gas temperatures as well as the load are low.

Catalysts [0073] SCR catalysts are made of a ceramic material that is a mixture of a support (titanium oxide) and active ingredients (vanadium oxides and in some cases tungsten). The two most important shapes of SCR catalysts are currently honeycomb and slab. The honeycomb form is usually an extruded ceramic with a catalyst incorporated in a (homogeneous) structure or applied to a substrate. In plate geometry, the substrate material is usually coated with a catalyst. When treating flue gases containing dust, the reactors are usually vertical with flue gas descending. The catalyst is usually placed in a series of two or four beds or layers. For better use of the catalyst, it is common to use three or four layers, provided an additional layer is present that is not initially installed.

[0074] As the catalyst activity decreases, an additional catalyst is installed in the available places in the reactor. As the deactivation progresses, the catalyst is replaced continuously, one layer at a time, starting from the top. This strategy leads to the maximum utilization of the catalyst. The catalyst is periodically blown away with soot to remove deposits, using steam as a cleaning agent.

Chemistry:

[0075] The chemistry of the SCR method is as follows:

4NO + 4NH + O2 4N2 + 6H2O 2NO2 + 4NH3 + O2 3N2 + 6H2O [0076] Side reactions are as follows:

SO2 + / O2 SO3

2NH3 + SO3 + H2O (NH4) 2SO4

-16NH3 + SO3 + H2O NH4HSO4

Description of the method [0077] As shown in Fig. 20, contaminated exhaust 112 exits the power generation system 110. This exhaust can be purified by other air pollution control subsystems 122, before entering the selective catalytic reduction (SCR) 2170 subsystem. The flue gas can also be purified by other APC subsystems (not shown) after leaving the SCR and before leaving the chimney 117. NOx in the exhaust gas introduced is measured by one or more analyzers 2003. The exhaust gas from NOx 2008 is passed through an ammonia (NH3) 2050 injection grid. Ammonia 2061 is mixed with dilution air 2081 using an ammonia / dilution air mixer 2070. Mixture 2071 is dosed to the exhaust through the 2050 injection mesh. The dilution air blower 2080 supplies ambient air 152 to the 2070 mixer, and the ammonia storage and supply system 2060 supplies ammonia to the 2070 mixer. The exhaust gas containing NOx, ammonia and dilution air 2055 passes into the SCR reactor 2002 and through the SCR catalyst. The SCR catalyst promotes NOx reduction with ammonia to nitrogen and water. The "NOx" free exhaust gas leaves the SCR 2002 reactor and leaves the plant through potentially other APC subsystems (not shown) and chimney 117.

[0078] There are additional NOx 2004 analyzers in the "free" NOx exhaust gas stream coming out of the SCR 2002 reactor or in chimney 117. The measured NOx value at outlet 2111 is combined with the measured NOx value at inlet 2112 to calculate the NOx 2110 removal efficiency. NOx is defined as the percentage of NOx at the inlet removed from the exhaust gas.

[0079] The calculated NOx removal capacity 2022 is a contribution to a regulatory control system that restores the 2021A ammonia flow rate setpoint to the ammonia / dilution air mixer 2070 and finally to the 2050 ammonia injection grid.

SCR method controllers [0080] The standard SCR control system is based on the cascade control system shown in Fig. 20. The internal PID controller loop 2010 is used to control the flow of ammonia 2014 to the 2070 mixer. The external PID controller loop 2020 is used to control NOx emissions. The operator is responsible for entering the NOx emission removal settings 2031 into the external loop 2020. As shown in Fig. 21, a 2030 selector may be used to place the upper limit 2032 in the operator 2031 setting. Additionally, forward feed signal 2221 for load (not shown in Fig. 21) is often used so that the controller can handle load transitions appropriately. For such embodiments, the load sensor 2009 generates a measurable load 2809 of the power generation system 110. This measured load 2809 is sent to controller 2220, which produces a 2221 signal . Signal 2221 is combined with the 2021A ammonia flow setpoint to form the corrected 2021B ammonia flow setpoint that is sent to

- PID controller 2010. PID 2010 combines the 2021B setpoint with the measured ammonia flow 2012, creating the VP 2011 ammonia flow that controls the amount of ammonia fed to the 2070 mixer.

[0081] The advantages of this controller are as follows:

1. Standard controller: it is a simple standard controller design that is used to meet the requirements specified by the SCR manufacturer and the catalyst seller.

2. DCS based controller: The design is relatively simple, it can be implemented in a DCS device and is the least expensive control option that will meet the operational requirements of the equipment and catalyst.

SCR working challenges:

[0082] The operation of the SCR is influenced by a number of operating parameters:

• NOx inlet content, • Local NOx molar ratio: ammonia, • Exhaust gas temperature and • Catalyst quality, availability and activity.

[0083] The operational challenges associated with the control scheme of Fig. 20 are as follows:

1. Measurement of Ammonia Residues: Keeping ammonia residues below a certain limit is crucial for SCR operation. However, calculations and direct measurements of ammonia residues are often not carried out. Although ammonia residue measurement is possible, it is often not included in the control loop. Therefore, one of the most key variables for SCR operation is not measured.

The SCR working goal is to achieve the desired level of NOx removal with minimal ammonia 'residue'. Ammonia "residues" is defined as the amount of unreacted ammonia in the "NOx" free exhaust stream. Although low economic costs are associated with the actual amount of ammonia in ammonia residues, there are significant negative effects of ammonia residues:

• Ammonia can react with SO3 in the exhaust gas to form salt, which deposits on the heat exchanger surfaces of the air heater. This salt not only reduces heat exchange in the air heater, but also attracts ash, which further reduces heat exchange. At some point, the heat exchange in the air preheater decreases to the point where the preheater needs to be serviced for cleaning. In the best case, washing the air heater lowers the machine's rating.

-18 • Ammonia is also absorbed in the catalyst (the catalyst may be considered a sponge for ammonia). A sudden drop in exhaust / NOx content may cause large, rapidly arising ammonia residues. This is only a transient condition - outside the scope of a typical control system. Despite their transitional form, ammonia residues still combine with SO3 and the salt deposited in the air preheater - despite instability, this dynamic transient form can cause significant build-up of a salt layer on the air preheater (and promote the attraction of fly ash).

• Ammonia is also referred to as air pollution. Although ammonia residue is very small, ammonia has a very strong smell, so even relatively trace amounts can create a bad smell problem for the local community.

• Ammonia is absorbed on fly ash. If the concentration of ammonia in fly ash is too high, there may be a significant cost associated with the removal of fly ash.

2. NOx Removal Performance Setting: Without measuring ammonia residue, the 20x NOx removal performance setting is often conservatively set by the operator / engineering staff to keep ammonia residues well below the residue limit. By conservatively selecting the NOx setting, the operator / engineer reduces the overall SCR removal efficiency. A conservative setting for NOx removal efficiency can guarantee that the ammonia residue limit will not be exceeded, but also results in lower efficiency than would be possible if the system were operated around an ammonia residue limit.

3. Effect of Temperature on the SCR: With a standard control system, there are no visible attempts to control the exhaust gas temperature in the SCR. Usually, certain methods are used to provide gas temperature within acceptable limits that usually prevent ammonia injection if the temperature is below the minimum value. In most cases, no attempt is made to actually control or optimize the temperature. In addition, no changes are made to the NOx setting based on temperature or based on the temperature profile.

4. NOx Profile and Speeds: The operation of boilers and pipes contributes to the formation of an uneven profile on the SCR face. For minimal ammonia residues, the NOx: ammonia ratio should be controlled, and without even mixing, this control must be local and avoid places with large ammonia residues. Unfortunately, the NOx distribution profile is a function not only of the ducts but also of the boilers. Thus, changes in the operation of boilers affect the distribution of NOx. Standard controllers do not take into account the fact that NOx profiles at inlet and SCR speeds are rarely uniform or static. This results in excessive injection of the reagent in some parts of the conductor cross section to ensure the correct amount of reagent

-19 in other areas. This increases the ammonia residue for a given NOx removal efficiency. Again, the operator / engineering team often respond to poor distribution by lowering the NOx setting.

It should be understood that the NOx analyzers at the inlet and outlet of 2003 and 2004 may be single analyzers or certain forms of the analytical system. In addition to the average NOx concentration, information on the distribution / profile of NOx could provide many analytical values. Using additional information on NOx distribution would require many 2010 ammonia flow controllers with some intelligence to dynamically distribute the total ammonia flow between different regions of the injection grid so that the ammonia flow is closer to local NOx concentrations.

5. Dynamic control: The standard controller also does not provide effective dynamic control. That is, when the SCR inlet conditions change, thus requiring modulation of the ammonia injection rate, it is unlikely that closed-loop NOx reduction control can prevent significant variations in this process variable. Rapid load changes and process delays are dynamic events that can cause significant process deviations.

6. Catalyst Decomposition: The catalyst decomposes over time, reducing the SCR removal efficiency and increasing residual ammonia. The control system must consider this distribution to maximize the NOx removal rate.

7. Moving average emissions: Many emission regulations introduce restrictions for both instantaneous emissions and some form of average moving emissions. To enable dynamic process deviations, the limit for instantaneous emissions is higher than for the moving average; continuous operation at the instantaneous emission limit value would violate the moving average limit. The moving average emission limit is the average value of instantaneous emissions in a moving or rolling time window. The time window can be as short as 1 hour and as long as 1 year. Some typical time windows are 1 hour, 3 hours, 24 hours, 1 month and 1 year. The standard controller does not take into account automatic control of moving averages. Most NOx emission limits relate to regional 8-hour limits of moving average NOx concentration in ambient air.

[0084] Operators usually set the desired NOx removal efficiency setting for SCR and make minor adjustments based on rare information from fly ash. Little effort is made to improve dynamic SCR control during load changes or to optimize SCR performance. Choosing the right instantaneous NOx removal performance and, if possible, moving average is also an unreliable and changing problem due to business, regulatory / credit issues and process issues that are similar to those associated with optimal WFGD performance.

[0085] Other APC processes have problems associated with:

-20 • Control / optimization of dynamic process flow, • Control of by-products / co-products quality, • Control of moving average emissions and • Optimization of APC measures.

[0086] These problems in other methods are similar to those detailed in the above discussions of WFGD and SCR.

<a name="caption1"></a>BRIEF SUMMARY OF THE INVENTION [0087] The invention discloses a controller for controlling a method carried out essentially for controlling the emission of a particular pollutant into air according to claim 1 and an article for controlling the method carried out essentially for controlling the emission of a particular pollutant into the air according to claim 9.

BRIEF DESCRIPTION OF THE FIGURES [0088]

Fig. 1 is a block diagram showing the outline of the standard wet flue gas desulphurization (WFGD) subsystem.

Fig. 2 shows further details of some embodiments of the WFGD subsystem shown in Fig. 1.

Fig. 3 shows further details of other forms of the WFGD subsystem shown in Fig. 1.

Fig. 4 is a graph of SO2 removal efficiency versus gypsum purity as a function of pH.

Fig. 5A shows the WFGD restriction field with the efficiency of the WFGD process within the comfort zone.

Fig. 5B shows the WFGD restriction field of Fig. 5A with optimized WFGD process efficiency.

Fig. 6 is a functional block diagram of an example MPC control architecture.

Fig. 7 shows components of an exemplary MPC controller and estimator suitable for use in the architecture of Fig. 6.

Fig. 8 shows further details of the processing unit and memory disk of the MPC controller shown in Fig. 7.

Fig. 9 is a functional block diagram of the estimator included in the MPC controller described in detail in Fig. 8.

Fig. 10 shows the MPCC multi-level architecture.

-21Fig. 11A represents the screen interface shown to the user by the multi-level MPC controller.

Fig. 11B illustrates another screen interface shown by the multi-level MPC controller for reviewing, modifying and / or adding planned downtime. Fig. 12 is an enlarged view of the multi-level architecture of MPCC in Fig.

10.

Fig. 13 shows a functional block diagram of an MPCC interface with an estimator and DCS for the WFGD process.

Fig. 14A shows a DCS screen for monitoring MPCC control.

Fig. 14B shows another DCS screen for entering laboratory and / or other values.

Fig. 15A shows the WFGD subsystem with all subsystem operation controlled by MPCC.

Fig. 15B shows the MPCC which controls the WFGD subsystem shown in Fig. 15A.

Fig. 16 shows further details of some embodiments of the WFGD subsystem shown in Fig. 15A, which correspond to those shown in Fig. 2.

Fig. 17 shows further details of the WFGD subsystem shown in Fig. 15A, which correspond to those shown in Fig. 3.

Fig. 18 shows further details and yet other forms of the WFGD subsystem shown in Fig. 15A.

Fig. 19 shows further details of the MPCC form shown in Fig. 15B.

Fig. 20 is a block diagram illustrating the outline of a typical selective catalytic reduction (SCR) device.

Fig. 21 shows the control diagram in a standard method for the SCR subsystem.

Fig. 22 shows details of a processing unit and a controller memory disk

MPC.

Fig. 23A shows the SCR subsystem with all subsystem operation being controlled by MPCC.

Fig. 23B shows further details of the form of MPCC shown in Fig. 23A.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS [0089] The efficient and effective operation of WFGD and similar subsystems has now been shown to be more complex than ever before. In addition, it is likely that this complexity will increase in the coming years under the influence of additional competitive pressure and additional pollution regulations.

-22 Standard strategies and techniques for process control are not able to cope with this complexity, and therefore are not able to provide optimal control of such activities.

[0090] In a business environment that changes dynamically during the useful life of the subsystem, it is desirable to maximize the commercial value of the subsystem operation at a given time. This asset optimization can be based on factors that are not even considered in the standard process control strategy. For example, in a business environment where there is a market for regulatory credit trading, the effective operation of the subsystem may impose the possibility of creating and selling additional regulatory loans to maximize the value of the subsystem, regardless of the additional operating costs that may be incurred to produce such loans.

[0091] Thus, instead of a simple strategy for maximizing SO2 absorption, minimizing operational costs, and meeting specifications for by-product quality, a more complex strategy can be used to optimize subsystem operations, regardless of whether SO2 absorption is maximized, operational costs are minimized, and the specification the by-product quality is met. In addition, you can not only provide tools to significantly improve subsystem control, but you can fully automate subsystem control. Thus, operations can be automated and optimized not only for operating parameters and constraints, but also for the business environment. The subsystem can be automatically controlled to operate very close or even exactly with legally permitted limits when the market value of regulatory loans created is less than the additional operating costs of producing such loans in the subsystem. However, the subsystem can also be automatically controlled to adjust such activities to operate below legally permitted limits, and thus can produce regulatory loans when the market value of regulatory loans created is greater than the additional operating costs of producing such loans in the subsystem. In fact, automatic control can direct the subsystem to operate to remove as much SO2 as possible up to a marginal dollar value, i.e. when the value of the issue credit is equal to the processing costs for producing the loan.

[0092] In summary, the optimized operation of WFGD and similar subsystems requires consideration of not only complex process and regulatory factors, but also complex economic factors and dynamic changes in these different types of factors. Optimization may require consideration of economic factors that are local, e.g., exclusion of one of the many WFGD processing units, and / or regional, e.g. disable the operation of another entity's WFGD processing unit in the region or even global. The optimization of operations can also take into account the widely and dynamically changing market prices of, for example, long-term and short-term SO2 regulatory loans.

[0093] Therefore, controllers should preferably be able to tailor the operation to minimize SO2 removal, subject to legal limit, or to maximize SO2 removal. The possibility of making such adjustments will allow the subsystem owner to take advantage of the dynamic change in the value of the regulatory loan and generate loans from one subsystem to offset operations beyond the limit of other subsystems or to use the need of another subsystem owner to purchase regulatory loans to offset operations beyond the limit of that subsystem. In addition, controllers should also be able to re-adjust operations when making further regulatory credits is no longer beneficial. In other words, the control system should continuously optimize the operation of APC assets in terms of hardware, process, legal and economic constraints.

[0094] Since there are no incentives to exceed the required purity of the gypsum by-product, the controllers should advantageously facilitate work optimization to match the gypsum by-product quality to the gypsum quality specification or other commercial restrictions. Optimized control should facilitate the avoidance of limestone masking by anticipating and directing actions to adjust the O2 level in terms of the desired SO2 absorption level and gypsum production requirements.

[0095] As mentioned above, controlling emissions to moving average is a complex problem. This is because, at least in part, the time window for the moving average always advances and many time windows are active at any time. Usually active windows stretch from time to time in the past, and other active windows stretch from time to time in the future.

[0096] Management of moving average emissions requires the integration of all emissions in the moving average time window. Therefore, optimizing emissions relative to the moving average target requires that the temporary emission target be selected to take into account actual past emissions and predicted future emissions or work plans for all "active" time windows.

[0097] For example, the optimization of the moving average of four hours requires checking multiple time windows, the first of which begins 3 hours and 59 minutes earlier and ends at the present, and the last starts at the present and ends in 4 hours in the future . It should be noted that with a one-minute "resolution" of each time window, optimization of this relatively short four-hour moving average would involve the selection of a temporary target that meets the limits of 479 time windows.

[0098] Setting a moving average emission target for a single integrated time window involves first calculating the total past emissions in the integrated time window and then, for example, predicting the pace of future emissions to remind you of that single integrated time window that will result in average emissions during a single integrated emission window at or below the moving average limit. Future emissions start at present

-24 point in time. However, to be accurate, future emissions must also include predicting emissions from activities during the reminder period of a single integrated time window.

[0099] It will be understood that the longer the time window is, the more difficult it is to predict future emissions. For example, emissions from activities over the next few hours can be predicted quite accurately, but emissions from activities for the next 11 months are more difficult to predict because factors such as seasonal changes and planned downtime should be taken into account. In addition, you may need to add a safety margin for unplanned downtime or capacity constraints in the subsystem.

[0100] Therefore, for optimizing the WFGD process, e.g., minimizing operational costs and / or maximizing SO2 removal while keeping the process within operational constraints, optimal settings for the WFGD process should be automatically determined.

[0101] In the embodiments described in detail below, the multivariable predictive control (MPC) method has been used to provide optimal control of the WFGD process. Generally, MPC technology provides multi-input and multi-output dynamic process control. As experts know in this field, MPC technology was originally developed in the second half of the 1970s. Technical innovations in this field continue to this day. MPC includes a number of model-based control techniques or methods. These methods allow the control engineer to deal with complex, interacting, dynamic processes more effectively than is possible with standard closed-loop PID control systems. MPC techniques can control both linear and nonlinear processes.

[0102] All MPC systems explicitly use dynamic models to predict future process behavior. Additionally, the specific control action is calculated to minimize the target's performance. Finally, a shifting horizon is introduced, with the horizon shifting one value towards the future for each increase in time. In addition, for each increment, the first control signal corresponding to the operation of the control having the sequence calculated in this step is applied. There are many commercial programs available for control engineers such as Generic Predictive Control (GPC), Dynamic Matrix Control (DMC) and Pegasus' Power Perfected^TM. Comancho and Bordons provide an excellent overview of the MPC topic in Model Predictive Control, Springer-Verlag London, Ltd. 1999, and Lennart Ljund System Identification, Theory for the User, Prentice-Hall, Inc. Second Edition, 1999 is a classic work on dynamic process modeling that is necessary for the actual implementation of MPC.

[0103] MPC technology is most commonly used in supervisory mode to perform operations normally performed by the operator, rather than replacing the basic regulatory control introduced by DCS. MPC technology is capable of automatically

- balancing competitive goals and process constraints using analytical techniques to ensure optimal settings for the process.

[0104] MPC usually has the following features:

Dynamic Models: A dynamic model for prediction, e.g. a non-linear dynamic model. The model is easily developed using parametric and gradual testing of the plant. The high quality of the dynamic model is the key to excellent optimization and control performance.

Dynamic Identification: The dynamics of the process or how the process changes over time is determined using stage tests of the installation. Based on these stage tests, an optimization algorithm is used to determine the dynamics of the installation.

Stable Optimization: The Stable Optimizer is used to find the optimal working point for the process.

Dynamic Control: The dynamic controller is used to calculate the optimal control movements around the solution in a stable state. Control motions are calculated using the optimizer. The optimizer is used to minimize the user-assigned cost function, which is subject to a number of restrictions. The cost function is calculated using a dynamic process model. Based on the model, cost function and constraints, you can calculate the optimal control movements for the process.

Dynamic Feedback: The MPC driver uses dynamic feedback for model updates. By using feedback, you can greatly reduce the effects of interference, model mismatch, and sensor noise.

Advanced Matching Features: The MPC driver provides a complete set of matching options. For manipulated variables, the user can set the desired value and factor; coefficient of motion for changing parameters; lower and upper border; speed of changing restrictions; and upper and lower hard limits. The user can also use the result of the stable state optimizer to set the desired value of the manipulated variable. For controlled variables, the user can set the desired value and factor; error weights; limits, priority hard and track limits.

Simulation Environment: An off-line simulation environment is provided for initial testing and adjustment of the controller. The simulation environment allows testing of model mismatch and interference rejection capability.

Network system: The MPC control algorithm is preferably implemented in a standard server with software that can be run in a standard commercial operating system. The server communicates with DCS via a standard interface. Engineers and operators can

-26 advantageously view the predictions of the MPC output algorithm using a graphical user interface (GUI).

Reliable Error Handling: The user defines how the MPC algorithm should respond to errors in input and output signals. The driver can be turned off if an error appears in critical variables or the last known good value can be used for non-critical variables. By properly dealing with errors, you can maximize the operational efficiency of the controller.

Virtual Network Analyzers: In cases where direct measurements of process variables are not available, the environment provides the infrastructure for implementing a virtual on-line analyzer (VOA) based on software. With this MPC tool, you can develop a model for the desired process variable using historical data from the installation, including laboratory data, where appropriate. Then, you can enter process variables in real time into the model and predict an immeasurable process variable in real time. This prediction can then be used in a driver with predictive modeling.

Optimization of the WFGD Process [0105] As described below in more detail, the efficiency of SO2 removal can be improved. This means that the SO2 removal rate from the device can be maximized and / or optimized, meeting required or desired limits such as limiting plaster purity, instantaneous emission limit, and rolling emission limit. Additionally, you can also minimize or optimize operating costs. For example, slurry pumps can be automatically turned off when the exhaust gas flow to WFGD is reduced. In addition, the oxidizing air stream and SO2 removal can also or alternatively be dynamically adjusted to prevent limestone masking conditions. By using the described MPC controller, you can manage the WFGD process closer to the limits and achieve better performance compared to standard controlled WFGD processes.

[0106] Figs. 5A and 5B show the WFGD 500 and 550 "constraints" fields. As shown, by determining process and hardware constraints 505-520 and using process-based stable relationships between multiple independent variables (MVs) multiple variables) and specific constraints, ie dependent / controlled variables, it is possible to map constraints to a common "space" in MV conditions. This space is in fact an n-dimensional space where n is equal to the number of degrees of freedom MV manipulated in the problem. However, if for the purposes of illustration we assume that we have two degrees of freedom, i.e. two MVs, it is therefore possible to present the limitations and dependencies of the system using a two-dimensional (XY) chart.

[0107] Preferably, the process and hardware constraints limit the non-empty solution space, which is shown as areas of action that can be performed 525. Any solution in this space will meet the limitations of the WFGD subsystem.

[0108] All WFGD subsystems exhibit some degree of variation. Referring to Fig. 5A, a typical standard operating strategy is to conveniently place the normal variability of the WFGD subsystem in the comfort zone 530 of the space of a feasible solution 525 - this will generally ensure safe operation. Maintaining activities in the 530 comfort zone allows activities to be kept away from areas of unenforceable / undesirable activities, i.e. away from areas outside the 525 feasibility region. Typically, distributed control system (DCS) alarms are set at or near the limits of measurable restrictions to alert operators of an expected problem.

[0109] Although it is true that any point in the feasibility space 525 meets the limitations of the 505-520 system, different points in the feasibility space 525 do not have the same operating costs, SO2 absorption efficiency, or gypsum by-product capacity. Maximizing profit, SO2 absorption efficiency or the production / quality of the by-product in the form of plaster or minimizing the cost requires determining an economically optimal point to operate within feasibility area 525.

[0110] Process variables and the cost or benefit of maintaining or changing the value of these variables may, for example, be used to create a goal function that reflects profit, which in some cases can be considered a negative cost. As shown in Fig. 5B, using linear, square or non-linear programming solution techniques as will be described below, it is possible to determine the optimal feasible point of solution 555, such as the least cost or maximum profit solution point within the feasible area of ​​measure 525. Because restrictions and / or costs may change at any time, it is beneficial to redefine the optimal feasible point of 555 in real time, e.g. every time the MPC driver works.

[0111] Thus, the automatic re-orientation of the process operation from a standard operating point within comfort zone 530 to an optimal operating point 555 or from an optimal operating point 555 to another optimal working point when a change in cost constraints can be facilitated. After determining the optimal point, the changes required for the MV value are calculated to move the process to the optimal working point. These new MV values ​​become target values. These target values ​​are stable and do not determine the dynamics of the process. However, to move the process safely, you also need to control and manage the dynamics of the process, which leads us to the next point.

[0112] To move the process from the old operating point to the new optimal operating point, prediction process models, feedback and high frequency execution are used. Using MPC techniques, a dynamic path or track of controlled variables (CV) is predicted. By using this prediction and managing MV matching not only in the current time but also in the future, e.g. in the near future, you can manage the dynamic CV path. one can

-28 calculate new CV target values. In addition, you can also calculate the dynamic error within the desired time horizon as the difference between the predicted path for the CV and the new CV target values. Once again, using the optimization theory, you can calculate the optimal path that minimizes error. It should be understood that in practice an engineer can weigh errors so that some CVs are more strongly controlled than others. Predictive process models also allow you to control a path or track from one work point to the next - this way you can avoid dynamic problems when moving to a new optimal work point.

[0113] In summary, operations can be conducted at virtually any point in the area of ​​feasible operations 525, which may be required to optimize the process to obtain any desired result. That is, the method can be optimized regardless of whether the goal is to achieve the lowest possible emissions, the highest quality or quantity of by-product, the lowest operating costs or any other results.

[0114] To approach the optimal operating point 555, MPC preferably reduces the process variation so that small deviations do not violate the restrictions. For example, by using a predictive process model, feedback and high-frequency execution, MPC can significantly reduce the process variation of a controlled process.

Stable and Dynamic Model [0115] As described in previous paragraphs, stable and dynamic models are used in MPC controllers. These models are described in this chapter.

[0116] Steady State Models: The process steady state for a certain set of input signals is a state that has been described by the associated process values ​​that the process would have obtained if all the input signals had been kept constant for a long period, such that earlier signal values input would no longer affect the status. For WFGD, due to the large volume and relatively slow reaction in the processing unit crystallizer, the time to steady state is usually about 48 hours. The stable state model is used to predict process values ​​associated with a stable state for a set of process input signals.

[0117] Stable State Model of Basic Principles: One approach to developing a model of stable state is to use a set of equations that were derived from engineering knowledge of the process. These equations can reflect known basic relationships between process input and output signals. To derive this set of equations, known physical, chemical, electrical and technical equations can be used. Because these models are based on known principles, they are referred to as basic principles models.

[0118] Many processes are originally developed using basic principles techniques and models. These models are generally accurate enough to ensure safe operation in the comfort zone as described above with reference to Fig. 5A. However, providing very accurate models of basic principles is often time consuming

-29i expensive. In addition, unknown influences often have a significant impact on the accuracy of basic principle models. Therefore, alternative approaches to create very accurate stable models are often used.

[0119] Experimental models: Experimental models are based on actual data collected from the process. The experimental model is built using the data regression technique to establish the relationship between the input and output signals of the model. Very often, data is collected in a series of installation tests in which individual input signals are carried to record their effect on the output signals. These plant tests can last from several days to weeks to gather sufficient data for experimental models.

[0120] Linear Experimental Models: Linear experimental models are created by fitting lines or planes in higher dimensions to a series of input and output data. Algorithms for fitting such models are widely available, for example, Excel provides a regression algorithm for fitting lines to a series of experimental data. Neural network models: Neural network models are another form of experimental models. Neural networks allow more complex curves than a line to be matched to a series of experimental data. Architecture and training algorithms for the neural network model are inspired by biology. The neural network consists of nodes that model the basic functions of a neuron. The nodes are connected by masses that form the basic interactions between neurons in the brain. Weights are determined using a training algorithm that mimics learning in the brain. Using models based on neural networks, a much richer and more complex model can be developed than can be achieved with linear experimental models. Process relationships between input (X) and output (Y) signals can be reflected using neural network models. Future references are neural networks or neural network models in this document should be interpreted as process models based on a neural network.

[0121] Hybrid models: Hybrid models include a combination of elements from basic principles or known relationships and experimental relationships. For example, the relationship between X and Y (element of the basic principle) may be known. Dependencies or equations include a series of constants. Some of these constants can be determined using knowledge of basic principles. Other constants would be very difficult and / or expensive to set based on basic principles. However, it is relatively easy and cheap to use actual process data for X and Y and knowledge of basic principles. These unknown constants reflect experimental / regressive elements in the hybrid model. Regression is much smaller than in the experienced model and the experienced character of the hybrid model is much smaller because the model's form and some constants are determined based on the basic principles that govern physical dependencies.

[122] Dynamic Models: Dynamic models reflect the effects of changes in input signals on output signals over time. While steady state models are only used to predict the final resting state of the process, the models

Dynamic are used to predict a path that leads from one stable state to another. Dynamic models can be developed using knowledge of basic principles, empirical data or a combination thereof. In most cases, however, models are created using experimental data collected from a series of step tests of important variables that affect the state of the process.

[0123] Pegasus Power Perfecter Model: Most MPC controllers only allow the use of linear experimental models, i.e. the model consists of linear experimental stable state models and linear experimental dynamic models. Model Pegasus Power Perfecter^TM enables the combination of non-linear, experimental models of basic principles to create the final model that is used in the controller and is therefore preferably used to perform MPC. One algorithm for combining different types of models to create the final model for Pegasus Power Perfecter is described in US Patent No. 5,933,345.

WFGD Subsystem Architecture [0124] Fig. 6 shows a functional block diagram of the WFGD subsystem architecture with model predictive control. The 610 controller contains the logic necessary to calculate the real-time settings for manipulated MV 615, such as the pH or oxidizing air of the WFGD 620 process. The 610 controller bases these calculations on observed process variables (OPV 625 such as state MV, interference variables (DV disturbance variables) and controlled variables (CVs). In addition, a set of reference values ​​(RV) 640, which usually have one or more tuning parameters, will also be used to calculate the MV 615 manipulated settings.

[0125] Estimator 630, which is preferably a virtual on-line analyzer (VOA), contains the logic necessary to generate estimated process variables (EPVs) 635. EPVs are usually process variables whose cannot be measured accurately. Estimator 630 uses logic to generate real-time estimates of the EPV work status of the WFGD process based on current and previous OPV values. It should be understood that OPV may include both DCS process measurements and / or laboratory measurements. For example, as discussed above, gypsum purity can be determined based on laboratory measurements. The estimator 630 can advantageously provide alarms for various types of problems in the WFGD process.

[0126] Controller logic 610 and estimator logic 630 may be used in software or other means. It should be understood that if desired, the controller and estimator can easily be used in a single computer process, which will also be understood by those skilled in the art.

Model Predictive Control (MPCC) controller [0127] The controller 610 of Fig. 6 is preferably used using the model predictive controller (MPCC). MPCC provides multi-input and multi-output

- Dynamic real-time control of the WFGD process. MPCC calculates the settings for the MV set based on the observed and estimated PV 625 and 635 values. MPCC in WFGD can use any of these values ​​or their combinations, measured by:

• pH probes • Suspension Density Sensors • Temperature Sensors • Oxidated Reduction Potential (ORP) Sensors oxidation-Reduction Potential) • Absorber Level Sensors • SO2 Sensors at the Input and Output / in the Chimney • Exhaust Gas Sensors at the Entrance • Laboratory Analysis of Absorber Chemistry (Cl, Mg, FI) • Laboratory Analysis of Plaster Purity • Laboratory Analysis of Fragmentation and Limestone Purity [0128] MPCC WFGD may also use any calculated setpoint or any combination of calculated setpoints to control the following:

• Limestone feeder • Limestone crushers • Limestone suspension stream • Chemical additives / reagent feeders / valves • Oxidative air flow control valves or shock absorbers or blowers • Valve or pH setting • Recycling pumps • Valves / pumps for adding and removing make-up water • Absorber chemistry (Cl, Mg, FI) [0129] MPCC WFGD can therefore control any of the following CVs or combinations thereof:

• SO2 Removal Performance • Gypsum Purity • pH • Suspension Density • Absorber Level

• Limestone Fragmentation and Purity • Operating Costs [0130] The MPC method provides the flexibility to optimally calculate all aspects of the WFGD process in one connected controller. The main challenge in operating WFGD is to maximize operating profit and minimize operational losses by balancing the following competing goals:

• Maintaining the SO2 removal rate at an appropriate level, in relation to the desired limiting limits, eg admissibility limits or limits, which in justified cases maximize SO2 removal credits.

• Maintaining an appropriate value of gypsum purity, in relation to the desired limiting limits, eg the limit of gypsum purity specification.

• Maintaining operating costs at an appropriate level in relation to the desired limit, eg minimum costs of electricity consumption.

[0131] Fig. 7 shows an example MPCC 700 that includes both a controller and an estimator similar to those described with reference to Fig. 6. As will be described below, MPCC 700 can balance the competitive goals described above. In a preferred embodiment, the MPCC 700 includes the MPC Pegasus Power Perfecter control system^TM and models based on a neural network, however, other control systems and models not based on neurons may be used instead, as discussed above and as will be understood by those skilled in the art.

[0132] As shown in Fig. 7, MPCC 700 includes a processing unit 705, with multiple I / O input / output ports 715 and a memory disk unit 710. Memory disk 710 may be one or more devices of any suitable kind or types and may use electronic, magnetic, optical or other forms or forms of storage media. It should also be understood that although a relatively small number of I / O ports have been described, the processing unit may contain any number of I / O ports suitable for a particular embodiment. It should also be understood that DCS process data and settings sent back to DCS can be packed together and sent as a single message using standard communication protocols between computers - and the basic functionality of data communication is necessary for MPCC to operate, details of implementation are well known to those skilled in the art in this field and are not relevant to the control problem described herein. Processing unit 705 communicates with a memory disk 710 for storing and retrieving data via communication link 712.

[0133] MPCC 700 also includes one or more input devices for receiving user input, e.g., operator input signals. As shown in Fig. 7, the 720 keyboard and 725 mouse facilitate manual input of commands or data to the processing unit 705 via communication links 722 and 727 and 715 I / O ports. MPCC 700 also includes a 730 display

- presenting information to the user. The processing unit 705 transmits information so that it is presented to the user on the display 730 via a communication link 733. In addition to facilitating the transmission of user input data, the I / O ports also facilitate the transmission of non-user data to the processing unit 705 via communication links 732 and 734 and forwarding directives, e.g. generated control directives from processing unit 715 via communication links 734 and 736.

Processing Unit, Logic and Dynamic Models [0134] As shown in Fig. 8, the processing unit 705 includes a processor 810, memory 820 and interface 830 to facilitate the reception and transmission of I / O signals 805 via communication links 732-736 of Fig. 7 820 memory is usually a type of random access memory (RAM). The 830 interface facilitates the interaction between the 810 processor and the user via a 720 keyboard and / or 725 mouse, as well as between the 810 processor and other devices, as will be described in more detail below.

[0135] As also shown in Fig. 8, the memory disk unit 710 stores the logic for estimation 840, the logic for prediction 850, the logic for generating control 860, the dynamic control model 870 and the dynamic estimation model 880. Stored logic are executed in accordance with the stored models for controlling the WFGD subsystem to optimize operations, as will be described in more detail below. The storage disk device 710 also includes a data storage 885 for storing received or calculated data and a database 890 for storing SO2 emission history.

[0136] The control matrix detailing the input and output signals that are used in the MPCC 700 to balance the three above objectives is shown in Table 1 below.

Table 1: Control Matrix

<td></td><td>SO removal₂</td><td>Plaster cleanliness</td><td>Operating cost</td>

<td>Manipulated Variables</td><td></td><td></td><td></td>

<td>PH</td><td>X</td><td>X</td><td></td>

<td>Air blower amplifiers</td><td></td><td>X</td><td>X</td>

<td>Pump amplifiers for recycling</td><td>X</td><td></td><td>X</td>

<td></td><td></td><td></td><td></td>

<td>Disturbing variables</td><td></td><td></td><td></td>

<td>SO₂ at the inlet</td><td></td><td></td><td>X</td>

<td>Exhaust speed</td><td></td><td></td><td>X</td>

<td>chloride</td><td>X</td><td>X</td><td></td>

<td>Magnesium</td><td>X</td><td>X</td><td></td>

<td>Fluoride</td><td>X</td><td>X</td><td></td>

<td>Limestone fineness and purity</td><td></td><td>X</td><td>X</td>

<td>Internal Energy Cost</td><td></td><td></td><td>X</td>

<td>The cost of limestone</td><td></td><td></td><td>X</td>

<td>Plaster price</td><td></td><td></td><td>X</td>

[0137] In the embodiment described here, MPCC 700 is used to control CVs including SO2 removal rate, gypsum purity and operating costs. Settings for MV including pH, oxidative air blower load and recycling pump load are manipulated for CV control. MPCC 700 also includes a number of DVs.

[0138] MPCC 700 must balance the three competitive goals associated with the CV, taking into account a number of restrictions. Competitive goals are formulated in a goal function that is minimized using the optimization technique of non-linear programming coded in the MPCC control system. By entering weighting factors for each of these purposes, for example using a 720 keyboard or 725 mouse, a WFGD system operator or other user can determine the relative importance of each of the goals depending on the specific circumstances.

[0139] For example, under certain circumstances, the SO2 removal rate may have more weight than gypsum purity and operating costs, and operating costs may have more weight than gypsum purity. In other circumstances, operating costs may have more weight than gypsum purity and SO2 removal rate, and gypsum purity may have more weight than SO2 removal rate. In yet other circumstances, the cleanliness of the plaster may be more important than the SO2 removal rate and operating costs. You can specify any number of weighing combinations.

[0140] MPCC 700 will control the operation of the WFGD subsystem based on individual weights so that the subsystem will operate at the optimal point, e.g. the optimal point 555 shown in Fig. 5B, while still noting the corresponding set of restrictions, e.g. the 505-520 restrictions shown in Fig. 5B.

[0141] For this particular example, the restrictions are set out in Table 2 below. These restrictions are typical CV and MV related restrictions described above.

<td>Table 2:</td><td colspan="3">Controlled and Restricted Manipulated Variables.</td>

<td>Variables manipulated:</td><td>Limit Minimal</td><td>Limit maximum</td><td>The value requested</td>

<td>SO removal₂</td><td> 90%</td><td> 100%</td><td>Maximal</td>

<td>Plaster cleanliness</td><td> 95%</td><td> 100%</td><td>Minimal</td>

<td>Operation costs</td><td>Lack</td><td>Lack</td><td>Minimal</td>

<td></td><td></td><td></td><td></td>

<td>Variables manipulated:</td><td>Limit Minimal</td><td>Limit maximum</td><td></td>

<td>PH</td><td> 5,0</td><td> 6,0</td><td>calculated</td>

<td>Blower air</td><td> 0%</td><td> 100%</td><td>calculated</td>

<td>Recycling Pump No. 1</td><td>disabled</td><td>enabled</td><td>calculated</td>

<td>Recycling Pump No. 2</td><td>disabled</td><td>enabled</td><td>calculated</td>

<td>Recycling Pump No. 3</td><td>disabled</td><td>enabled</td><td>calculated</td>

<td>Recycling Pump No. 4</td><td>disabled</td><td>enabled</td><td>calculated</td>

Dynamic Control Model [0142] As noted above, MPCC 700 requires an 870 dynamic control model with an input / output structure as shown in the control matrix in Table 1. To develop such a dynamic model, a basic principle model and / or experimental model is initially developed based on WFGD process tests at the plant. The basic principles model and / or experimental model can be developed using the techniques discussed above.

[0143] In the example of the WFGD subsystem being exemplified, a stable (basic or experimental) model of the WFGD process is preferably developed for the SO2 removal rate and gypsum purity. Using the basic principles method, the stable state model is developed based on the known basic relationships between the input and output signals of the WFGD process. Using the neural network method, the model of the stable state of SO2 removal rate and gypsum purity is developed by collecting experimental data from real processes in various operating states. A model based on neural networks that can capture process non-linearity is set up using experimental data. It can again be seen that although a neural network based model may be beneficial in some embodiments, the use of such a model is not mandatory. On the contrary, a non-neural network model can be used if desired, and is even preferred in some embodiments.

[0144] In addition, a stable state model is developed for operating costs from basic principles. Cost factors are sufficient to develop a total cost model. In this embodiment, the cost of different

-36 raw materials such as limestone and the cost of electricity are multiplied by the right amount of consumption to develop a total cost model. The income model is determined by multiplying the SO2 removal loan price by the tonnage of SO2 removed and multiplying the plaster price by the plaster tonnage. Operating profit (or losses) can be determined by subtracting the cost from income. Depending on the pump controller (fixed or variable speed), optimization of pump preparation may include binary OFF-ON decisions; this may require a secondary optimization step to fully evaluate the various pump preparation options.

[0145] Although accurate stable state models can be developed and could be suitable for a solution based on stable state optimization, such models do not include process dynamics and are therefore not particularly suitable for use in MPCC 700. Therefore, step-by-step tests in the WFGD subsystem are performed to collect real dynamic process data. The step test response data is then used to create the experimental dynamic control model 870 for the WFGD subsystem, which is stored by the processor 810 in the memory disk unit 710, as shown in Fig. 8.

Dynamic Estimation Model and Virtual Network Analyzer [0146] Fig. 6 illustrates how an estimator such as that contained in MPCC 700 is used for total advanced control of the WFGD process. In MPCC 700, the estimator is preferably in the form of a virtual network analyzer (VOA). Fig. 9 shows further details of the estimator included in MPCC 700.

[0147] As shown in Fig. 9, the observed MV and DV are introduced into the experimental dynamic estimation model 880 for the WFGD subsystem, which is used to perform control systems 840 on the processor 810. Therefore, the processor 810 performs control 840 in accordance with the dynamic model 880 estimation. In this case, the 840 control system calculates the actual CV values, e.g. SO2 removal efficiency, plaster purity, and operating costs.

[0148] Table 3 shows the structure of the dynamic estimation model 880. It should be noted that the control matrix and dynamic estimation model 880 used in MPCC 700 have the same structure.

Table 3: Process model for the estimator.

<td></td><td>SO removal₂</td><td>Plaster cleanliness</td>

<td>Manipulated Variables</td><td></td><td></td>

<td>PH</td><td>X</td><td>X</td>

<td>Air blower amplifiers</td><td></td><td>X</td>

<td>Pump amplifiers for recycling</td><td>X</td><td></td>

<td>Disturbing variables</td><td></td><td></td>

<td>SO₂ at the inlet</td><td></td><td></td>

<td>Exhaust speed</td><td></td><td></td>

<td>chloride</td><td>X</td><td>X</td>

<td>Magnesium</td><td>X</td><td>X</td>

<td>Fluoride</td><td>X</td><td>X</td>

<td>Limestone fineness and purity</td><td></td><td>X</td>

[0149] At the output of the estimated logic 840 there is an open value loop for SO2 removal and gypsum purity. The dynamic estimation model 880 for VOA was developed using the same method described above for the development of the dynamic control model 870. It should be noted that although the dynamic estimation model 880 and the dynamic control model 870 are basically the same, the models are used for very different purposes. The dynamic estimation model 880 is used by the 810 processor to perform logic for estimating 840 to generate accurate estimations of current process constant (PV) values, e.g. estimated CV 940. The dynamic control model 870 is used by the 810 processor to perform the estimation logic 850, to optimally calculate the manipulated MV 615 settings shown in Fig. 6.

[0150] As shown in Fig. 9, a feedback loop 930 from the estimation block 920 is provided that reflects the estimated CV generated by the processor 810 as a result of performing the logic to estimate 840. Thus, the best CV estimate is sent back to the dynamic estimation model 880 through the 930 feedback loop. The best CV estimate from the previous iteration of the estimator is used as a starting point for adapting the dynamic 880 estimation model to the current iteration.

[0151] The validation block 910 reflects the validation of the observed values ​​of CV 950, for example from sensor measurements and laboratory analyzes, by the processor 810 using the results of the logic system for estimation 840, according to the dynamic estimation model 880 and the observed MV and DV 960. Validation presented in block 910 is also used to determine potential conditions for masking limestone. For example, if the observed MV is the pH value measured by one of the pH sensors, validation of 910 of the measured pH based on the pH value estimated according to the dynamic estimation model 880 may indicate that the pH sensor is defective. If the observed SO2 removal, gypsum purity or pH value is found to be incorrect, the processor 810 will not use this value to estimate 920. Instead, a substitute value, preferably the resulting output value will be used

-38 from estimation based on a dynamic estimation model. In addition, an alarm will be sent to DCS.

[0152] To calculate the estimation 920, the processor 810 combines the result of the execution of the logic for the estimation 840 based on the dynamic estimation model 880, with the observed and validated CV. The Kalman filter method is preferably used to combine the estimation result with the observed, validated data. In this case, the validated SO2 removal rate calculated from the input and output of the SO2 sensors is combined with the generated removal rate value to obtain an estimate of true SO2 removal. Due to the accuracy of the SO2 sensors, the logic for estimating 840 preferably places great emphasis on the filtered version of the observed data to the detriment of the generated value. Plaster purity is measured only at most every few hours. The 810 processor will also combine new observations regarding gypsum purity with the generated gypsum purity estimate. During the interval between measurements of the plaster sample, the processor 810, according to the dynamic estimation model 880, will perform updated estimates of plaster purity in an open loop based on changes in the observed MV and DV 960. Thus, the processor 810 also performs estimates of plaster purity in real time.

[0153] Finally, the processor 810 performs logic for estimating 840, according to the dynamic estimation model 880, to calculate the operating costs for WFGD. Because there is no direct cost measurement in line, the 810 processor must perform real-time operational cost estimation.

Emissions Management [0154] As mentioned above, US operating permits generally set limits for both instantaneous emissions and average rolling emissions. There are two types of moving average emission problems that the MPCC 700 solves in controlling the WFGD subsystem. The first type of problems occurs when the moving average time window is less than or equal to the time horizon of the logic for prediction 850 performed by the 810 MPCC 700 processor. The second type of problems occurs when the moving average time window is greater than the time horizon of the logic for estimating 850 .

Single Level MPCC Architecture [0155] The first type of problem, the short time window problem is solved by adapting normal MPCC 700 constructs to integrate moving average emissions as an additional CV with MPCC 700 control. More specifically, the predictive logic 850 and control generating logic 860 will treat steady state conditions as a process constraint that should be maintained at or below an allowed level rather than an economic constraint, and will impose a dynamic control path that maintains the current and future moving average values ​​in the appropriate time window at or below the allowed level

-39niego. In this way, an MPCC 700 with tuning configuration for average moving emissions is provided.

Considerations of Interference Variables [0156] In addition DV for factors such as planned operating events, e.g. load change, which will affect emissions within the current horizon are included in the predictive logic 850, and thus in the MPF 700 control of the WFGD process. In practice, the actual DVs that are stored as part of the 885 data on the data disk unit 710 will be different depending on the type of WFGD subsystem and the specific working philosophy adopted for the subsystem, e.g. base load and swing load. DVs can be adjusted, from time to time, by the operator via input pulses input via the 720 keyboard and 725 mouse, or by the logic itself to generate 860 control or by an external scheduling system (not shown) via the 830 interface.

[0157] However, DVs are usually not in a form that can be easily customized by operators or other users. Accordingly, a work plan interface tool is preferably provided as part of the predictive logic 850 to assist the operator or other user in setting up and maintaining the DV.

[0158] Figs. 11A and 11B show the interface presented on the display 730 for entering scheduled downtimes. As shown in Fig. 11A, a screen 1100 is presented that displays the designed factor of the power generating system and the designed factor of the WFGD subsystem to the operator or other user. Additionally, buttons are displayed that allow the user to enter one or more planned outages and display previously entered planned outages for review or modification.

[0159] If the button allowing the user to enter downtime is selected using the mouse 725, the screen 1110 shown in Fig. 11B is shown to the user. The user can then enter, via the 720 keyboard, various details about the new planned downtime as shown. By clicking the provided button add downtime, the new planned downtime is added as DV and taken into account by the 850 prediction logic. The logic containing this interface sets the appropriate DV so that the future work plan is passed to the MPCC 705 processing unit.

[0160] Regardless of the actual DV, the DV function will be the same, and it is embedding the effect of planned work events in the predictive logic 850, which can be performed using the MPCC 810 processor to predict future dynamic conditions and the steady state moving average CV emission. In this way, the MPCC 700 performs 850 prediction logic to calculate the predicted moving average of emissions. The predicted moving average of emissions is in turn used as an input signal for the logic to generate the 860 control that is made by the MPCC 810 processor to take into account planned operational events

-40 in the control plan. In this way, the MPCC 700 is equipped with a tuning configuration for moving average emissions due to planned operational events, and therefore with the ability to control WFGD operation within the limits of allowable moving average emissions, despite planned operational events.

MPCC Two Level Architecture [0161] The second type of problem, the long time window problem, is preferably solved using the MPCC two level method. In this method, MPCC 700 contains many, preferably two, cascaded controller processors.

[0162] Referring to Fig. 10, level 1 705A controller processor (CPU) works to solve the problem of a short or short time window, as described above with respect to single-level architecture. As shown in Fig. 10, CPU 705A includes an 810A processor. The 810A processor performs 850A prediction logic stored in the 710A memory disk unit to provide dynamic management of the rolling average of emissions within a time window equal to the short period of time in force. A CV that reflects the short-term or applicable goal of controlling the moving average emission horizon is stored as part of the 885A data in the 710A CPU 705A data carrier.

[0163] CPU 705A also includes memory 820A and interface 830A similar to memory 820 and interface 830 described above with reference to Fig. 8. The interface receives a subset of MPCC 700 I / O signals, i.e. 805A I / O signals. The 710A storage disk unit also stores 840A estimation logic and 880A dynamic estimation model, 860A control logic generation and 870A dynamic control model and 890A SO2 emission history database, all of which are described above with reference to Fig. 8. CPU 705A also includes a 1010 timer, usually a processor clock. The function of timer 1010 will be described in more detail below.

[0164] The two-level CPU 705B works to solve the problem of long-term or long time window. As shown in Fig. 10, CPU 705B includes an 810B processor. The 810B processor performs 850B prediction logic to also provide dynamic management of moving average emissions. However, the 850B predictive logic is implemented to manage dynamic moving average emissions due to the reduction of the entire future moving average emission time window and to determine the optimal short-term or appropriate time horizon, the target of moving average emissions, i.e. the maximum limit for level 1 CPU 705A. Therefore, the CPU 705B serves as an optimizer for long-term moving average emissions and allows predicting the moving average of emissions over a suitable time horizon for controlling the moving average of emissions over the entire future time window.

[0165] A CV reflecting the limitation of the long-term moving average emission over a time horizon is stored as part of 885B data in a memory disk unit 710B. The CPU 705B also contains 820B memory and an 830B interface, similar to

- memory 820 and interface 830 described above. The 830B interface receives a subset of the MPCC 700 I / O signals, i.e. 805B I / O signals.

[0166] Although the two-level architecture in Fig. 10 includes a plurality of CPUs, it is recognized that multi-level logic circuits may be made if desired for other predictions. For example, in Figure 10, MPCC 700 level 1 is reflected by the 705A CPU, and MPCC 700 level 2 is reflected by the 705B CPU. However, a single processor such as CPU 705 of Fig. 8, can be used to carry out both the 850A prediction program and the 850B prediction program, and thus determine the optimal short-term or appropriate target of the moving average emission time horizon, due to the fact that the predicted optimal long-term moving average emission solves the problem of long-term or long window emissions and optimized the short-term or relevant moving average of emissions for a specific purpose.

[0167] As noted above, CPU 705B is looking for a long-term time horizon, sometimes referred to as a control horizon, corresponding to a moving average time window. Preferably, the CPU 705B manages dynamic moving average emissions over the entire future window of moving average emissions and sets the optimal limit for short-term moving average emissions. The CPU 705B operates at a frequency fast enough to allow it to capture changes in the work plan over relatively short periods.

[0168] CPU 705B uses short-term or corresponding moving average emission targets, which is considered CV by COU 750A, as MV, and includes long-term moving average CV emissions. The long-term moving average is therefore stored as part of the 885B data on the 710B memory disk unit. The 850B predictive logic will treat steady state conditions as a process constraint that should be maintained at or below an allowed level rather than an economic constraint, and will also impose a dynamic control path that maintains current and future moving average values ​​in the appropriate time window at or below the permitted level. In this way, an MPCC 700 with tuning configuration for average moving emissions is provided.

[0169] In addition, DV for factors such as planned operational events, e.g. load change, which will affect emissions within the applicable horizon, are included in the 850B predictive logic, and thus in the WFGD MPCC 700 control. As noted above, in practice, the actual DVs that are stored as part of the 885B data on the 710B data disk unit will be different depending on the type of WFGD subsystem and the specific working philosophy adopted for the subsystem can be adjusted by the operator or by the 705B logic processor for generating an 860B control or by an external scheduling system (not shown) via the 830B interface. However, as discussed above, DVs are usually not in a form that can be easily customized by operators or other users, and therefore a work plan interface tool such as that shown in Fig. 11A is preferably provided

-42i 11B, as part of the logic to predict 850A and / or 850B to assist the operator or other user in setting up and maintaining the DV.

[0170] However, also here, regardless of the actual DV, the DV function will be the same, and it is embedding the effect of planned work events in the 850B predictive logic, which can be performed using the MPCC 810B processor to predict future dynamic conditions and steady state moving average CV emission.

[0171] In this way, the CPU 705B performs a predictive logic 850B to determine the optimal short-term or corresponding reduction of moving average emissions due to planned operational events in the control plan. Optimal short-term or appropriate reduction of average moving emissions is transmitted to the CPU 705A via communication link 1000. In this way, the MPCC 700 is equipped with a tuning configuration to optimize the average moving emission due to planned operational events, and therefore the ability to optimize the control of WFGD operation within the limits of the allowable average moving emission, despite planned operational events.

[0172] Fig. 12 is an enlarged view of the multi-level MPCC architecture. As shown, the operator or other user uses the 1220 remote control terminal to communicate with both the process history database 1210 and MPCC 700 via communication links 1225 and 1215. MPCC 700 includes CPU 705A and CPU 705B of Fig. 10, which are interconnected by links communication 1000. Data related to the WFGD process is sent, via communication link 1230, to the process history database 1210, which stores this data as historical process data. As described below, the necessary stored data is downloaded from database 1210 via communication link 1215 and processed by CPU 705B. The necessary data related to the WFGD process is also transmitted over a 1235 communication link and processed by the 705A CPU.

[0173] As described earlier, CPU 705A receives CV targets corresponding to the current desired long-term moving average goals from CPU 705B via communication link 1000. The passed moving average goal is the optimized goal for long-term moving average generated by CPU 705B, performing logic 850B . Communication between the CPU 705A and CPU 705B is handled in the same way as the communication between the MPC controller and the optimizer in real time.

[0174] CPU 705A and CPU 705B preferably have a handshake protocol that ensures that if the CPU 705B stops sending optimized targets for the long-term moving average to the CPU 705A, the CPU 705A will be in replacement mode or adopt an intelligent and conservative working strategy for long-term average constraints moving. The 850A prediction logic may include such a tool for establishing such a protocol, thereby providing the necessary reconciliation and offloading. However, if the 850A prediction logic does not contain such

The tools, typical features and functions of DCS can be customized in a manner well known to those skilled in the art to carry out the required reconciliation and relief.

[0175] It is critical to ensure that the CPU 705A consistently uses timely, i.e. fresh - not outdated, long-term moving average goals. Each time the CPU 705B performs 850B predictive logic, it will calculate a fresh, new long-term moving average target. The CPU 705A receives a new target from the CPU 705B via communication link 1000. Based on the receipt of a new target, the CPU 705A performs 850A prediction logic to reset the 1010 timer. If the CPU 705A does not receive the new target from the 705B via communication link 1000 on time, the 1010 timer is out of time or expires. Based on the 1010 timer expiration, the CPU 750A, according to the logic to predict, considers the current long-term moving average goal to be outdated and proceeds to a secure working strategy until it receives a fresh new long-term time average goal from the CPU 705B.

[0176] Preferably, the minimum timer setting is slightly longer than the frequency the CPU 705B carries out the computer load room / planning issue. Due to the irregular operation of many optimizers in real time, it is common standard practice to set the communication time controllers in half to twice the time to stable state of the controller. However, since the execution of the logic for CPU 705B prediction is regular, the recommended guideline for setting the 1010 timer is not that of the optimization link in the stable state, but it should be no more than twice the operating frequency of the controller on the CPU 705B plus from about 3 to 5 minutes.

[0177] If the CPU 705A determines that the current long-term moving average target is deprecated, the long-term moving average limit should be reset. Without the CPU 705B providing a fresh new long-term moving average goal, the CPU 705A has no long-term guidance or goal. Accordingly, in this case, the CPU 705A increases the safety margin of process operations.

[0178] For example, if the rolling average period is relatively short, e.g., 4 to 8 hours, and the subsystem is operating under low load conditions, the CPU 705A may increase the outdated moving average removal target by 3 to 5 percent by weight, according to logic for predicting 850A. Such an increase should, under such conditions, establish a sufficient safety margin for continuing operations. With regard to the operator impact necessary to cause growth, only a single value, e.g. 3 weight percent, is required in the prediction logic.

[0179] On the other hand, if the time average period is relatively long, e.g. 24 or more hours and / or the subsystem is operating under varying load, the CPU 705A may return to the conservative purpose according to the logic to control 850A. One way is for the CPU 705A to use the assumed constant operation at the planned load

-44 subsystem or above over the whole period of the moving average time window. The CPU 705A can then calculate based on such permanent activities, a fixed emission target and add a small safety margin or comfort factor that can be determined by facility management. To implement this solution in the CPU 705A, the 850A prediction logic must include the functionality. However, it should be noted that, if necessary, the functionality of setting this conservative goal can be done in DCS, not in CPU 705A. It would also be possible to introduce this conservative goal as a secondary CV in the 705A single-level controller and to enable this CV only if the short-term moving average 1000 goal is out of date.

[0180] Thus, regardless of whether the moving average period is relatively short or long and / or the subsystem operates at constant or variable load, preferably the 850A predictive logic includes load limits so that operator action is not required. However, other techniques can also be used to set the load limit as long as the technique establishes safe / conservative operation against moving average limits during periods when the CPU 705B does not provide fresh, new, long-term moving average goals.

[0181] It should be noted that actual SO2 emissions are tracked by MPCC 700 in the process history database 1210, regardless of whether the CPU 705B is functioning properly or providing fresh, new long-term goals for moving average CPU 705A. Stored emissions can therefore be used by the CPU 705B to track and account for SO2 emissions that occur when the CPU 705B is down or not communicating correctly with the CPU 705A. However, when the CPU 705B is running again and is able to communicate properly, it will, according to the 850B predictive logic, again optimize the moving average emissions and increase or decrease the current moving average emission targets used by the 705A CPU to match actual emissions that have occurred during downtime, and provided fresh, new, long-term goals for the moving average CPU 705A via communication link 1000.

Implementation in line [0182] Fig. 13 shows a functional block diagram of the interaction of MPCC 1300 with DCS 1320 in the WFGD 620 process. MPCC 1300 includes both a 1305 controller that may be similar to the 610 controller in Fig. 6 and an estimator 1310 that may be similar to the estimator 630 of Fig. 6. MPCC 1300 can be, if desired, the MPCC shown in Figs. 7 and 8. MPCC 1300 can also be configured using a multi-level architecture such as that shown in Figs. 10 and 12.

[0183] As shown, controller 1305 and estimator 1310 are connected to DCS 1320 via a data interface 1315, which may be part of the interface 830 of Fig. 8. In a preferred embodiment, the data interface 1315 is made using the Pegasus Data Interface software module (^TM) (PDI). However, this is not necessary and the 1315 data interface can be implemented using other interface logic circuits.

-45 Data interface 1315 transmits settings for manipulated MV and read PV. Settings can be transmitted as I / O signals 805 in Fig. 8.

[0184] In this preferred embodiment, the controller 1305 is made using Pegasus (^TM) Power Perfecter (PPP), which consists of three software components: a data server element, a controller element and a graphical user interface (GUI). The data server element is used to communicate with PDI and collect local data related to the application of the control. The controller element performs the logic for prediction 850 for performing algorithmic calculations of the model predictive control in terms of the dynamic control model 870. The GUI element displays, e.g. on the display 730, the results of these calculations and provides an interface for tuning the controller. Here also the use of Pegasus (^TM) Power Perfecter is not mandatory and the 1305 driver can be implemented using some of the other driver programs.

[0185] In this preferred embodiment, the estimator 1310 is implemented using the Pegasus software module (^TM) Run-time Application Engine (RAE). RAE communicates directly with PDI and PPP. It is believed that RAE provides a number of features that make it a very economical environment for VOA. The function of the error checking program, activity monitoring, ability to supervise communication and computer processes and the alarming device are preferably implemented in RAE. However, again, the use of Pegasus (^TM) Run-time Application Engine is not mandatory and you can perform estimator 1315 using another estimator program. If desired, it is also possible and is recognized by specialists in this field to implement functionality equivalent to VOA in DCS for WFGD 620.

[0186] Controller 1305, estimator 1310 and PDI 1315 preferably operate in one processor, e.g. processor 810 of Fig. 8 or 810A of Fig. 10, which is connected to the controller network, including DCS 1320 for the WFGD 620 process, by means of a connection Ethernet. Currently, the processor operating system is usually based on Microsoft Windows^TM, although it is not mandatory. The processor can also be part of a high power workstation computer module or other type of computer, as shown in Fig. 7, for example. In any case, the processor and its associated memory must have sufficient processing power and mass storage to execute the programs necessary for performing advanced WFGD control as described herein.

DCS Modifications [0187] As described above with reference to Fig. 13, the control processor performing the prediction logic 850 interacts with DCS 1320 for the purpose of the WFGD 620 process through interface 1315. To facilitate proper interaction of the 1305 and DCS 1320 controller, standard DCS will usually required modification. Accordingly, DCS 1320 is preferably standard DCS that has been modified in a manner well known in the art to include the elements described below.

[0188] DCS 1320 is preferably adapted, i.e. programmed with the required control systems typically employing software, to enable the operator or other user to perform the following functions from the DCS interface screen:

• Change of PPP CONTROL MODE between automatic and manual.

• CONTROLLER STATUS.

• Viewing the ALARM COUNT ("ACTIVITY") status.

• Viewing MV attributes for STATUS, MIN, MAX, CURRENT.

• ENABLING each MV switching off each MV.

• Viewing CV attributes for MIN, MAX and CURRENT values.

• Entering laboratory values ​​for gypsum purity, absorber chemistry and limestone properties.

[0189] To assist the user in accessing this function, DCS 1320 is adapted to display two new screens as shown in Figs. 14A and 14B. Screen 1400 in Fig. 14A is used by the operator or other user to monitor MPCC control, and screen 1450 in Fig. 14B is used by the operator or other user to enter laboratory and / or other values ​​as needed.

[0190] For convenience and to avoid unnecessary complexity in understanding the invention, elements such as operating costs have been excluded from the control matrix for the purposes of the following description. However, it should be understood that operating costs can easily be incorporated into the control matrix and this can be beneficial in many cases. Additionally, for convenience and to simplify the discussion, recycling pumps are treated as DV and not MV. Here again, experts in the field will know that in many cases, treatment of pumps for recycling as MVs may be beneficial. Finally, it should be noted that the following discussion assumes that the WFGD subsystem has two absorption towers and two associated MPCCs (one case of MPCC for each absorber in the WFGD subsystem).

DCS Advanced Control Screens [0191] Referring now to Fig. 14A, as shown, screen 1400 includes CONTROLLER MODE, which is an operator / user selectable identifier that can be automatic or manual. In AUTOMATIC mode, the 1305 controller that performs logic for prediction 850, e.g. Pegasus (^TM) Power Perfecter, calculates MV motions and causes the logic to generate 860 control to direct control signals, sending these motions to DCS 1320. The 1305 controller performing the predictive logic 850 will not calculate MV motions unless the variable is enabled, i.e. marked as AUTO.

[0192] Controller 1305 performing logic prediction 850 such as Pegasus (^TM) Power Perfecter includes an alarm or "activity" counter function

- monitors the integrity of the 1315 communication interface with DCS 1320. If the 1315 communication interface fails, an alarm indicator (not shown) appears on the screen. The 1305 controller performing the predictive logic 850 will recognize the alarm state and, based on the alarm state, begin to offload all enabled, i.e. active, lower level DCS configuration selections.

[0193] Screen 1400 also includes a COMPLETE STATUS, which indicates whether prediction logic 850 has been successfully performed by controller 1305. For controller 1305 to continue to operate, its GOOD status (as shown) is required. The 1305 controller performing the predictive logic 850 will recognize the BAD state and, responding to the BAD state recognition, will break all active connections, i.e. it will return DCS 1320 control.

[0194] As shown, MVs are displayed with the following information headers:

ENABLED - This field can be set by an operator or other user input to a 1305 controller performing 850 prediction logic to enable or disable any MV. Turning off the MV is responsible for switching the MV to the off state.

SP - means the logic setting for prediction 850.

MODE - Indicates whether the predictive logic 850 recognizes the appropriate MV as enabled, suspended, or completely disabled.

MIN LMT - Displays the minimum boundary used by the logic to predict 850 for MV. It should be noted that preferably these values ​​cannot be changed by the operator or other user.

MAX LMT - Displays the maximum limit used by the 850 predictive logic for MV. Here, too, it is preferable not to change these quantities.

PV - Shows the latest and current value of each MV that has been recognized by the 850 predictive logic.

[0195] Screen 1400 further includes the following details of the MV status field indicators:

The 1305 controller performing the predictive logic 850 will set the specific MV value only in MODE in ON. Four conditions must be met for this to happen. First, the operator or other user must select the enable field. DCS 1320 must be in automatic mode. The unloading conditions must be false, according to the calculations of the 1305 controller performing 850 prediction logic. Finally, the suspension conditions must be false, according to the calculations of the 1305 controller performing the 850 prediction logic.

[0196] The controller 1305 performing the prediction logic 850 will change and display the MV HOLD mode status if conditions are present that will not allow the controller 1305 to match that particular MV. In HOLD mode, controller 1305, according to

-48 logic to predict 850, will maintain the current MV value until it can remove the suspension conditions. For MV to be HOLD, four conditions must be met. First, the operator or other user must select the launch field. DCS 1320 must be in automatic mode. The unloading conditions must be false, according to the calculations of the 1305 controller performing 850 prediction logic. Finally, the suspension conditions must be true, according to the calculations of the 1305 controller that performs the 850 prediction logic.

[0197] The controller 1305 performing the prediction logic 850 will change the MV mode state to off and display the mode state off if there are conditions that will not allow the controller 1305 to adapt that particular MV based on any of the following conditions. First, the start box for the control mode is deselected by the operator or another user. DCS mode is not automatic, e.g. it is manual. All unloading conditions are true, as calculated by the 1305 controller, performing 850 logic prediction.

[0198] The controller 1305 performing the prediction logic 850 will recognize various unloading conditions, including estimator failure and error in entering laboratory values ​​during a predetermined period, e.g., the last 12 hours. If the 1305 controller performing 850 prediction logic determines that any of the above unloading conditions are true, it will return control of the MV to DCS 1320.

[0199] As also shown in Fig. 14A, the CVs are displayed with the following information headers:

PV - Indicates the last measured CV value received by the 1305 controller.

LAB - Indicates the last value of the laboratory tests along with the sample time received by the 1305 controller.

ESTIMATE - Indicates the current or latest CV estimate generated by the 1310 estimator that performs logic to estimate 840 based on a dynamic estimation model.

MIN - displays the minimum limit for the CV.

MAX - displays the maximum limit for the CV.

In addition, the 1400 screen displays trend charts over a certain pre-set period, e.g. over the last 24 hours of operation, for estimated CV values.

Laboratory Sample Application Form [0200] Referring now to Fig. 14B, an operator or other user is displayed a DCS screen with a prototype of a Laboratory Sample Application Form. This screen can be used by the operator or other user to enter the test values ​​of the laboratory samples that will be processed by the estimator 1310 of Fig. 13, according to the logic to estimate 840 and the dynamic estimation model 880, as previously described with reference to Fig. 8.

[0201] As shown in Fig. 14B, the following values ​​are entered along with the associated time stamp generated by estimator 1310:

Part 1 Values ​​for Laboratory Samples:

• Plaster purity • Chlorides • Magnesium • Fluorides

Part 2 Values ​​for Laboratory Samples:

• Plaster purity • Chlorides • Magnesium • Fluorides

Combined Values ​​for Laboratory Samples from Part 1 and Part 2:

• Plaster Purity • Limestone Purity • Limestone Crush [0202] The operator or other users enter the laboratory test values ​​along with associated sample execution times, for example using the keyboard 720 shown in Fig. 7. After entering these values, the operator activates the update button, for example using the 725 mouse shown in Fig. 7. Activating the update button will cause the 1310 estimator to update the values ​​for these parameters the next time the logic is performed for estimating 840. It should be noted that, if necessary, these laboratory test values ​​can alternatively be automatically entered into MPCC 1300 from the appropriate laboratory in digital form using an MPCC processing unit interface such as the interface 830 shown in Fig. 8. In addition, the MPCC driver layout can be easily customized, e.g. by programming to automatically activate the update function represented by the update button responding to receiving test values ​​in digital form from the appropriate laboratory or laboratories.

[0203] To ensure proper control of the WFGD process, laboratory test values ​​for gypsum purity should be updated every 8 to 12 hours. Accordingly, if cleanliness is not updated during this period, MPCC 1300 is preferably adapted, e.g. programmed with the necessary software, to relieve control and generate an alarm.

[0204] In addition, the absorber chemistry and limestone properties should be updated at least once a week. Here again, if these values ​​are not updated in a timely manner, MPCC is preferably configured to issue an alarm.

[0205] The validation logic is included in the estimation logic 840 performed by the estimator 1310 for validating the values ​​entered by the operator. If the values ​​are entered incorrectly, the estimator 1310, according to the logic for estimating 840, will return to the previous values, and the previous values ​​will still be displayed in Fig. 14B and the dynamic estimation model will not be updated.

Control of all activities of the WFGD [0206] Control of all activities of the WFGD subsystem by MPCC of any type discussed above will now be described with reference to Figs. 15A, 15B, 16, 17, 18 and 19.

[0207] Fig. 15A shows a power generation system (PGS) 110 and an air pollution control system (APS) 120 similar to those described with reference to Fig. 1, with similar references denoting similar elements in systems, some of which will not already described below to avoid unnecessary duplication.

[0208] As shown, the WFGD 130 'subsystem includes multi-variable control, which in this embodiment is carried out by MPCC 1500, which may be similar to the MPCC 700 or 1300 described above, and which, if desired, may incorporate the multi-level architecture of the type described with reference to Figs. 10-12.

[0209] Exhaust gas 114 from SO2 is routed from other APC subsystems 122 to the absorption tower 132. Ambient air 152 is compressed by a blower 150 and directed as oxidized compressed air 154 'to crystallizer 134. Sensor 1518 detects a measure of ambient conditions 1520. Measured ambient conditions 1520 may, for example, include temperature, humidity and atmospheric pressure. Blower 150 includes blower load control 1501 that can provide the current blower load value 1502 and modify the current blower load based on the blower load 1503 obtained.

[0210] As also shown, the limestone slurry 148 'is pumped with slurry pumps 133 from the crystallizer 134 to the absorption tower 132. Each of the slurry pumps 133 includes pump condition control 1511 and pump load control 1514. Pump condition control 1511 is able to provide the current value pump status 1512, e.g. indicating pump on / off status and change the current pump status based on the received pump status SP 1513. Pump load control 1514 can provide the current pump load value 1515 and change the current pump load based on the SP 1516 pump load. The flow of fresh limestone slurry 141 'from the mixer and tank 140 to the crystallizer 134 is controlled by the flow control valve 199 based on the slurry flow SP 196 '. The suspension flow SP 196 'is based on the PID control signal 181' determined based on the pH SP 186 ', as will be discussed in more detail below. Fresh suspension 141 'flowing into the crystallizer 134 is used to adjust the pH of the suspension

Used in the WFGD method, and thus to control SO2 removal from SO2 containing flue gas 114 entering the absorption tower 132.

[0211] As previously discussed, exhaust gas containing SO2 114 is introduced into the bottom of the absorption tower 132. SO2 is removed from the exhaust gas 114 in the absorption tower 132. Pure exhaust gas 116 ', which preferably does not contain SO2, is directed from the absorption tower 132 to, for example chimney 117. The SO2 analyzer 1504, which is shown at the outlet of the absorption tower 132, but can be located in the chimney 117 or in another position behind the absorption tower 132, detects the SO2 measure at the outlet 1505.

[0212] On the control side of subsystem 130 ', the multi-process process controller for the WFGD process, i.e. MPCC 1500 shown in Fig. 15B, receives a plurality of input signals. Input signals to MPCC 1500 include measured suspension pH 183, measured SO2 at inlet 189, blower load value 1502, measured SO2 at outlet 1505, gypsum purity value tested in laboratory 1506, measured PGS 1509 load, slurry pump condition values ​​1512, pump load value suspension 1515 and measured values ​​of ambient conditions 1520. As will be described below, these process parameter input values, in combination with other input signals, including non-process 1550 input signals and 1555 input limits, and 1560 computer estimated input parameters are used by MPCC 1500 to generate controlled parameter settings (SP) 1530 .

[0213] In operation, the SO2 analyzer 188 located in front of the WFGD 132 absorption tower, detects the SO2 measure at the inlet in the exhaust gas 114. The measured value of 189 SO2 at the inlet is fed into the forward-coupled device 190 and MPCC 1500. Power system load ( PGS) 110 is also detected by the PGS 1508 load sensor and sent as measured loads PGS 1509 to MPCC 1500. In addition, the SO2 1504 analyzer detects the SO2 measure at the outlet in the exhaust gas leaving the absorption tower 132. The measured value of 1505 SO2 at the outlet is also sent to MPCC 1500.

Gypsum Quality Estimation [0214] Also now referring to Fig. 19, the parameters entered into MPCC 1500 include parameters reflecting the current conditions in the absorption tower 132. Such parameters can be used by MPCC 1500 to generate and update a dynamic estimation model for gypsum. The dynamic estimation model for gypsum can, for example, be part of the dynamic 880 estimation model.

[0215] Since there is no practical way to directly measure gypsum purity in line, a dynamic gypsum estimation model can be used in conjunction with the estimation logic performed by the 1500B MPCC 1500 estimator such as the 840 estimation logic to calculate quality estimation plaster, shown as calculated plaster purity 1932. Estimator 1500B is preferably a virtual network analyzer (VOA). Although the 1500A driver and the 1500B estimator are

As depicted in one device, it should be noted that, if desired, the 1500A controller and the 1500B estimator can be placed separately and form separate components, as long as the 1500A controller units and the 1500B estimator are properly connected to allow the required communication. The calculated estimation of plaster quality 1932 may also reflect the control by the logic system for estimation based on laboratory measurements of plaster quality, presented as the plaster purity value 1506, entered into MPCC 1500.

[0216] The estimated 1932 gypsum quality is then passed by the 1500B estimator to the 1500A MPCC 1500 controller. The 1500A controller uses the estimated 1932 gypsum quality to update a dynamic control model such as the dynamic 870 control model. Predictive logic, such as predictive logic 850, is implemented by controller 1500A, according to the dynamic control model 870, to compare the adapted plaster quality 1932 with the plaster quality limitation reflecting the desired plaster quality. The desired plaster quality is usually assumed in the plaster sales contract specification. As shown, the plaster quality restriction is introduced into MPCC 1500 as a requirement for plaster quality 1924 and is stored as 885 data.

[0217] The controller 1500A, performing the prediction logic, determines whether, based on the results of the comparison, an adjustment is required for the operation of the WFGD subsystem 130 '. In this case, the calculated difference between the estimated 1932 gypsum quality and the 1924 gypsum quality limitation is used to make the logic to predict by the 1500A controller to determine the required adjustments in the operations of the WFGD subsystem to improve the 160 'gypsum quality within the 1924 gypsum quality limitation.

Maintaining Compliance With Gypsum Quality Requirements [0218] To bring the 160 'gypsum quality into compliance with the 1924 gypsum quality restriction, the required adjustments to WFGD operations, determined by the logic for prediction, are sent to the logic to generate control such as the system logic to generate the 860 control, which is also performed by the 1500A controller. The 1500A controller performs a logic to generate control to generate control signals corresponding to the required increase or decrease in plaster quality 160 '.

[0219] These control signals may, for example, correct the operation of one or more valves 199, slurry pumps 133 and blower 150, shown in Fig. 15A, so that the WFGD subsystem process parameters, e.g. measured slurry pH 148 ', flowing from the crystallizer 134 to the absorption tower 132, which is reflected by the measured pH value of the suspension 183 detected by the pH sensor 182 in Fig. 15A reflect the desired set point (SP), e.g. the desired pH value. This adjustment of the pH 183 of the 148 'suspension in turn causes a change in the quality of the 160' by-product gypsum actually produced by the WFGD 130 'subsystem and the estimated 1932 gypsum quality, calculated by the estimator 1500B, to better match the desired 1924 gypsum quality.

[0220] We will now also refer to Fig. 16, which contains further details regarding the structure and operation of the fresh water source 164, mixer / tank 140 and drainage device 136. As shown, the fresh water source 164 includes water tank 164A, with whose ME 200 fluid is pumped via the pump 164B to the absorption tower 132 and the fresh water source 162 is pumped via the pump 164C to the mixing tank 140A.

[0221] Operation and control of the drainage device 136 is not changed by the addition of MPCC 1500.

[0222] Operation and control of the limestone slurry preparation area, including mill 170 and Mixer / Tank 140, remain unchanged after the addition of MPCC 1500.

[0223] Referring now to Figs. 15A, 15B and 16, the controller 1500A may, for example, perform a logic to generate control to direct the flow change of the lime slurry 141 'to the crystallizer 134. The volume of the slurry 141' that flows to the crystallizer 134, is controlled by opening and closing valve 199. Opening and closing valve 199 is controlled by PID 180. The operation of PID 180 to control the operation of valve 199 is based on the slurry pH setting entered.

[0224] In order to properly control the flow of the slurry 141 'to the crystallizer 134, the controller 1500A determines the slurry pH setting that will bring the quality of gypsum 160' into compliance with the 1924 gypsum quality restriction. As shown in Figs. 15A and 16, determined the slurry pH setting, shown as pH 186 ', is sent to PID 180. Then, PID 180 controls the operation of valve 199 to modify the flow of the slurry 141' to match the obtained SP SP 186 '.

[0225] To control the operation of the valve 199, PID 180 generates the PID control signal 181 'based on the obtained pH of the suspension SP 186' and the obtained pH value 183 of the suspension 141 ', measured with a pH 182 sensor. The PID control signal 181' combines with the control signal Forward (FF) 191, which is generated by the FF 190. As is well known in the art, the FF 191 control signal is generated based on the measured SO2 at the inlet 189 of the flue gas 114 obtained from the SO2 analyzer 188 located in front of the absorption tower 132. The PID control signal 181 'and (FF) control signal 191 are combined in summation block 192, which is usually included, as a built-in function in the DCS output block that communicates with valve 199. The combined control signals leaving the summation block 192 are reflected by the mixture flow settings 196 '.

[0226] The slurry flow setting 196 'is sent to the valve 199. By default, the valve 199 includes another PID (not shown) that directs the actual opening or closing of the valve 199 based on the received slurry flow setting 196' to modify the slurry flow 141 'through the valve. In any case, based on the received slurry flow setting 196 ', the valve 199 is opened or closed to increase or decrease the volume of the slurry 141' and thus the volume of the slurry 240 'flowing

To crystallizer 134, which in turn modifies the pH of the suspension in the crystallizer 134 and the quality of the 160 'gypsum produced by the WFGD 130' subsystem.

[0227] Factors to be considered in determining when and whether MPCC 1500 is to reset / update the pH setpoint in PID 180 and / or PID 180 is to reset / update the slurry flow setting in valve 199 can be programmed using well known techniques in MPCC 1500 and / or if necessary in PID 180. As understood by those skilled in the art, factors such as PID 180 performance and pH 182 sensor accuracy are typically included in such determinations.

[0228] The controller 1500A generates the pH SP 186 'by processing the measured pH value of the suspension 148' flowing from the crystallizer 134 to the absorption tower 132, obtained from the pH sensor 182, reflected by the suspension pH 183, according to the plaster quality control algorithm or reference table, in dynamic control model 870. The algorithm or reference table reflect the established relationship between the quality of plaster 160 'and the measured pH value 183.

[0229] PID 180 generates the PID control signal 181 'by processing the difference between pH SP 186' obtained from controller 1500A and the measured pH value of the suspension 148 'obtained from the sensor pH 182, reflected by the pH of the suspension 183, according to a limestone flow control algorithm or reference table. This algorithm or reference table reflects the established relationship between the value of the change in volume of the slurry 141 'flowing from the mixer / tank 140 and the value of the change in the measured pH value 183 of the slurry 148' flowing from the crystallizer 134 to the absorption tower. It may be worth noting that although in the embodiment shown in Fig. 16, the amount of crushed limestone 174 flowing from mill 170 to mixing tank 140A is managed by a separate controller (not shown), in case of benefits it can also be controlled by MPCC 1500. In addition, although not shown, MPCC 1500 can, if necessary , also control the dosing of additions to the mixture in the 140A mixing tank. Accordingly, based on the pH SP 186 'obtained from the 1500A MPCC 1500 controller, PID 180 generates a signal that opens or closes valve 199, thereby increasing or reducing the flow of fresh lime slurry to the crystallizer 134. PID still controls the valve control until the volume of lime slurry 141 'flowing through valve 199 matches the MVSP reflected by the lime slurry flow setting 196'. It should be understood that the adjustment is preferably carried out with the PID (not shown) included as part of valve 199. However, alternatively, the adjustment can be made with PID 180 based on given flow volumes measured and sent back from the valve.

Maintaining Compliance with SO2 Removal Requirements [0230] By controlling the pH of suspension 148 ', MPCC 1500 can control SO2 removal from exhaust containing SO2 114 along with the quality of the by-product gypsum 160' produced by the WFGD subsystem. Increasing the pH of the 148 'suspension by increasing

The flow of fresh lime slurry 141 'through valve 199 will increase the amount of SO2 removed by the absorption tower 132 from exhaust gas with SO2 content 114. On the other hand, reducing the flow of lime slurry 141' through valve 199 will reduce the pH of slurry 148 '. Reducing the amount of SO2 absorbed (now in the form of calcium sulfite) flowing into the crystallizer 134 will also cause more calcium sulfite to be oxidized in the crystallizer 134 to calcium sulfate, thereby giving higher quality gypsum.

[0231] Thus, there is a tension between the two primary control objectives, the first of which is the removal of SO2 from the SO2-containing exhaust gas 114, and the second, to produce a by-product in the form of gypsum 160 'having the required quality. This means that there may be a conflict between meeting SO2 emission requirements and the plaster specification.

[0232] We now refer to Fig. 17, which provides further details regarding the construction and operation of the slurry pumps 133 and the absorption tower 132. As shown, the slurry pumps 133 include a plurality of separate pumps, shown as slurry pumps 133A, 133B and 133C in in this embodiment, which pumps the slurry 148 'from the crystallizer 134 to the absorption tower 132. As previously described with reference to Fig. 3, each of the 133A-133C pumps direct the slurry to different from a plurality of levels of slurry nozzles in the absorption tower 306A, 306B and 306C. Each level of 306A-306C suspension directs the suspension to a different of many levels of the 308A, 308B, and 308C suspension sprayers. The 308A-308C slurry atomizers spray the slurry, in this case the slurry 148 ', in the exhaust gas containing SO2 114, which is introduced into the absorption tower 132 through a gas inlet, for SO2 absorption. The clean exhaust 116 'is then discharged from the absorption tower 132 to the outlet opening of the absorption tower 312. As also described earlier, the ME fluid in the spray 200 is directed to the absorption tower 132. It will be understood that although 3 different levels of slurry nozzles and sprayers and three different pumps, the number of nozzle and atomizer levels and the number of pumps will most likely depend on the particular embodiment.

[0233] As shown in Fig. 15A, pump status values ​​1512 are transferred back from pump status controllers 1511 such as on and off switches and pump load values ​​1515 are transferred back from pump load controllers 1514, such as a motor, to MPCC 1500 to introduce to the dynamic control model. As also shown, pump 1513 status settings such as on or off instructions are sent to 1511 pump status controllers, and 1516 pump load settings are sent to 1514 pump load controllers by MPCC 1500 to control the status, e.g. on or off and load each pump 133A-133C, and therefore control to which nozzle levels the slurry 148 'is pumped and the amount of slurry 148' that is pumped to each nozzle level. It should be noted that in most current WFGD applications, slurry pumps 133 do not include variable load capability (on and off only), and therefore 1516 pump load settings and 1514 pump loads would not be available for use or control by MPCC

1500.

[0234] As detailed in the embodiment shown in Fig. 17, the pump state controllers 1511 include a separate pump state controller for each pump, designated by reference numerals 1511A, 1511B and 1511C. Similarly, the 1514 pump load controllers include a separate pump status controller for each pump, identified by reference numerals 1514A, 1514B and 1514C. Individual pump status values ​​1512A, 1512B and 1512C are sent to MPCC 1500 from the pump status controllers 1511A, 1511B and 1511C, respectively, to indicate the current state of the respective slurry pump. Similarly, individual pump load values ​​1515A, 1515B and 1515C are sent to MPCC 1500 from the pump load controllers 1514A, 1514B and 1514C, respectively, to indicate the current state of the respective slurry pump. Based on the state values ​​of the 1512A, 1512B and 1512C pumps, MPCC 1500 performs 850 prediction logic to determine the current state of each pump 133A, 133B and 133C, and as a result what is commonly called pump scheduling at any time.

[0235] As discussed earlier, the ratio of the flow rate of lime slurry 148 'entering the absorption tower 132 to the flow rate of exhaust gas 114 entering the absorption tower is commonly characterized as L / G. L / G is one of the key design parameters in WFGD subsystems. Because the exhaust gas flow rate 114, designated as G, is set upstream of the WFGD processing unit 130 ', usually by the operation of the power generating system 110, it is not and cannot be controlled. However, the flow rate of the liquid suspension 148 ', designated as L, can be controlled by MPCC 1500, based on the G value.

[0236] One way this is done is by controlling the operation of the slurry pumps 133A, 133B and 133C. Individual pumps are controlled by MPCC 1500, by sending pump state settings 1513A, 1513B and 1513C to pump state controllers, 1511A pump 133A, 1511B pump 133B and 1511C pump 133C, respectively, to obtain the desired ordering of pumps, and thus the levels at which slurry 148 'will be introduced into the absorption tower 132. If available in the WFGD subsystem, MPCC 1500 can also send the settings of the 1516A, 1516B and 1516C pump load controllers to the pump load controllers 1514A pump 133A, 1514B pump 133B and 1514C pump 133C, respectively, to obtain the desired suspension flow volume 148 'to the absorption tower 132 at each active nozzle level. Accordingly, MPCC 1500 controls the flow rate, L, of liquid slurry 148 'to the absorption tower 132 by controlling to which nozzle levels 306A-306C are pumped slurry 148' and the amount of suspension 148 'that is pumped to each nozzle level. It is understood that the greater the number of pumps and nozzle levels, the greater the fineness of such control will be.

[0237] Pumping slurry 148 'into the nozzles at higher levels, such as 306A nozzles, will cause the slurry that is sprayed from the slurry sprayers 308A to have a relatively long contact time with SO2 114 containing exhaust gas. This in turn will cause absorption or more SO2 from flue gas 114 through the slurry than for the slurry introduced into the absorber at lower spray levels. On the other hand, pumping slurry into lower nozzle levels such as 306C nozzles,

Will cause the suspension 148 'that is sprayed from the 308C slurry sprayers to have a relatively shorter contact time with the SO2 containing exhaust gas 114. This will lead to the absorption of a relatively small amount of SO2 from the exhaust gas 114 by the suspension. Thus, with the same amount and composition of the slurry 148 ', more or less SO2 will be removed from the exhaust gas 114, depending on the level of the nozzles to which the slurry is pumped.

[0238] However, pumping liquid suspension 148 'onto higher nozzle levels, such as 306A nozzles, requires relatively more power and therefore higher operating costs than is required when pumping liquid suspension 148' onto lower nozzle levels such as 306C nozzles . Accordingly, by pumping larger amounts of liquid slurry into higher nozzle levels to increase absorption, and thus removal of sulfur from exhaust gases 114, the operating costs of the WFGD subsystem increase.

[0239] The 133A-133C pumps are very large rotary devices. These pumps can be started and stopped automatically with the MPCC 1500 by sending the pump status SP or manually by the operator or other user of the subsystem. If the exhaust gas flow rate 114 entering the absorption tower 132 is modified due to a change in the operation of the power generation system 110, MPCC 1500, by performing a logic to predict 850, according to dynamic control model 870 and the logic to generate control will match the operation of one or more slurry pumps 133A-133C. For example, if the exhaust gas flow rate falls to 50% of the designed load, the MPCC may send one or more SP pump status to close, i.e. turn off one or more pumps currently pumping 148 'slurry to the absorption nozzle at one or more atomizer levels and / or one or more pump load control SPs to reduce the pump load on one or more pumps currently pumping slurry into the nozzles absorption tower at one or more spray levels.

[0240] In addition, if an organic acid or similar substance dispenser (not shown) is included as part of a mixer / pump 140 or is a separate subsystem that dispenses organic acid directly into the process, MPCC 1500 may also or alternatively send SP control signals (not shown) to reduce the amount of organic acid or other similar additives dosed to the suspension, to reduce the ability of the suspension to absorb and thus remove SO2 from the exhaust gas. It is understood that these additives can be quite expensive, and therefore their use has been relatively limited, at least in the United States of America. Once again, there is a conflict between SO2 removal and operating costs: additions are expensive, but additions can significantly increase SO2 removal with little or no impact on gypsum purity. If the WFGD subsystem includes the additive injection subsystem, it would therefore be appropriate to allow MPCC 1500 to control the injection of the additive along with the control of other WFGD method variables so that MPCC 1500 runs the WFGD method with the lowest possible operational costs regarding equipment, process and legal restrictions. By entering the cost of such additions to the MPCC 1500, this cost factor can be incorporated into the dynamic control model and taken into account by

- performing logic for prediction in controlling WFGD control.

Avoiding Limestone Masking [0241] As previously mentioned, for the oxidation of absorbed SO2 in gypsum formation, a chemical reaction must occur between SO2 and the limestone in suspension in the absorption tower 132. During this chemical reaction, oxygen is consumed to form calcium sulfate. The exhaust gas 114 introduced into the absorption tower 132 is low in O2, so usually additional oxygen is added to the liquid suspension flowing to the absorption tower 132.

[0242] Also referring to Fig. 18, the blower 150, which is most often described as a fan, compresses the surrounding air 152. The resulting oxidized compressed air 154 'is directed to the crystallizer 134 and introduced into the slurry in the crystallizer 134, and then pumped to absorber 132, as discussed earlier in relation to Fig. 17. The addition of compressed oxidation air 154 'to the slurry in the crystallizer 134 causes the re-circulated slurry 148' that flows from the crystallizer 134 to the absorber 132 to have an increased oxygen content that will facilitate oxidation and thus calcium sulfate formation.

[0243] Preferably, there is an excess of oxygen in the suspension 148 ', although it should be noted that there is an upper limit to the amount of oxygen that can be absorbed or retained by the suspension. To facilitate oxidation, it is preferred to conduct WFGD with a significant amount of excess O2 in suspension.

[0244] It should also be noted that if the O2 concentration in the suspension becomes too low, the chemical reaction between SO2 in the exhaust gas 114 and the limestone in the suspension 148 'will slow down and eventually cease to occur. When this happens, we commonly refer to it as limestone masking.

[0245] The amount of O2 that is dissolved in the recyclable suspension in the crystallizer 134 is not measurable. Accordingly, the dynamic 880 estimation model preferably includes the dissolved O2 model. Logical system for estimation, e.g. the 840 estimation logic performed by the 1500B MPCC 1500 estimator, in accordance with the dynamic 880 estimation model, calculates the estimated value of dissolved O2 in the recyclable mixture in the crystallizer 134. The calculated estimate is sent to the 1500A MPCC 1500 controller which uses the calculated estimate to update the dynamic control model, e.g. 870 dynamic control model. Next, the 1500A controller performs a prediction logic, e.g., the prediction logic 850, which compares the estimated value of dissolved O2 in suspension with the restriction of the value of dissolved O2 in suspension that has been introduced into MPCC 1500. The restriction of value of dissolved O2 in suspension is one of the restrictions 1555 shown in Fig . 15B and is shown in more detail in Fig. 19 as a requirement for O2 dissolved in suspension 1926.

[0246] Based on the result of the comparison, controller 1500A, while still performing prediction logic, determines whether correction is required for the operation of WFGD subsystem 130 'to ensure that the suspension 148' that is pumped into the absorption tower 132 does not have too little O2. It should be noted that ensuring that the slurry 148 'has sufficient dissolved O2 also helps ensure that SO2 emissions and the quality of the gypsum by-product continue to meet the required emission and quality limits.

[0247] As shown in Figs. 15A and 18, the blower 150 includes a load control mechanism 1501, which is sometimes called a blower speed control mechanism, which can match the flow of oxidizing air to the crystallizer 134. The load control mechanism 1501 can be used to regulate the blower load 150, and thus the amount of compressed oxidation air 154 'introduced into the crystallizer 134, and thus to facilitate any required adjustment of the operation of the WFGD 130' subsystem in the light of the results of the comparison. Preferably, the operation of the load control mechanism 1501 is controlled directly by the 1500A controller. However, if desired, the load control mechanism 1501 can be manually controlled by the subsystem operator based on the output from the controller 1500A, directing the operator to take appropriate manual control of the load control mechanism. In each case, depending on the result of the comparison, the 1500A controller performs 850 prediction logic, in accordance with the dynamic control model 870, to determine whether adjustment is needed to the amount of 154 'compressed oxidation air introduced into the crystallizer 134 to ensure that the slurry 148 'pumped into the absorption tower 132 was not low in O2 and, if so, the amount necessary for adjustment. The 1500A controller then performs a logic to generate control such as a logic to generate control 860, taking into account the blower load value 1502 received by MPCC 1500 from the load control mechanism 1501 to produce control signals to drive the load control mechanism

1501 to vary the load on the blower 150 to adjust the amount of compressed oxidation air 154 'introduced into the crystallizer 134 to the desired amount that will ensure that the slurry 148' pumped into the absorption tower has no O2 deficiency.

[0248] As mentioned earlier, O2 deficiency is a particular problem in the summer months when it heat reduces the amount of compressed oxidation air 154 'that may have been forced into crystallizer 134 by blower 150. Predictive logic 850 performed by controller 1500A may, for example , specify that the blower speed or load 150 that is entered into MPCC 1500 as the blower load value

1502, should be adjusted to increase the volume of compressed oxidation air 154 'fed into the crystallizer 134 by a certain amount. The logic to generate control performed by the 1500A controller then determines the blower load SP

1503, which leads to the desired increase in volume of the compressed oxidation air 154 '. Preferably, the SP load of the 1503 blower is sent from MPCC 1500 to the load control mechanism 1501, which directs the load increase corresponding to the SP load of the 1503 blower onto the 150 blower, thereby avoiding limestone masking

-60i ensuring that SO2 emissions and gypsum by-product quality are within applicable limits.

[0249] Increasing the speed or load of the blower 150 will of course also increase the blower power consumption and thus the operating costs of the WFGD subsystem 130 '. This increase in costs is also favorably monitored by MPCC 1500, while controlling the operation of the WFGD 130 'subsystem, and thus provides an economic incentive to control the blower 150 to direct only the necessary amount of compressed oxidation air 154' to the crystallizer 134.

[0250] As shown in Fig. 19, the current cost / unit of energy, shown as unit energy cost 1906, is preferably input into MPCC 1500 as one of the input signals not related to the 1550 process shown in Fig. 15B and included in the dynamic model control 870. Using this information, the 1500A MPCC 1500 controller can also calculate and display to the subsystem operator or other persons a change in operating cost, based on a correction of the oxidized compressed air flow 154 'to the crystallizer 134.

[0251] Therefore, provided that there is excess capacity of the blower 150, the controller 1500A will typically control the flow of compressed oxidation air 154 'to the crystallizer 134 to ensure that there is enough to prevent binding. However, if the blower 150 operates at full load and the amount of compressed oxidizing air 154 'flowing to the crystallizer 134 is still insufficient to prevent masking, i.e. there is a need to add air (oxygen) to oxidize all SO2 absorbed in the absorption tower 132, controller 1500A will have to implement an alternative control strategy. Therefore, when SO2 is absorbed by the suspension, it must be oxidized to gypsum, however, if there is no additional oxygen to oxidize a marginal amount of SO2, then the best solution is non-absorption of SO2, because if absorbed SO2 cannot be oxidized, it can finally limestone masking will occur.

[0252] In such circumstances, controller 1500A has another capability that it can use to control the operation of the WFGD 130 'subsystem to ensure that masking does not occur. More specifically, the 1500A controller, performing the predictive logic 850 in accordance with the dynamic control model 870 and the logic for generating control 860, can control the PID 180 to adjust the pH level of the suspension 141 'flowing to the crystallizer 134 and thus control the pH level of the suspension 148' , pumped to the absorption tower 132. By directing a decrease in the pH level of the suspension 148 'pumped into the absorption tower 132, the additional marginal SO2 absorption will be reduced and masking will be avoided.

[0253] Yet another alternative strategy that can be implemented by the 1500A controller is to operate beyond the limits 1555 shown in Fig. 15B. In particular, the 1500A controller may implement a control strategy that does not oxidize too much SO2 in the 148 'slurry in the crystallizer 134. Therefore, the amount of O2 required

In crystallizer 134 will decrease. However, this action will in turn impair the purity of the 160 'by-product gypsum produced by the WFGD 130' subsystem. By using this strategy, the 1500A controller overcomes one or more 1555 restrictions in controlling the operation of the WFGD 130 'subsystem. Preferably, the controller maintains stringent SO2 emission limits in clean exhaust 116 ', which is presented as a requirement for SO2 emission permits in Fig. 19 and exceeds and effectively reduces the specific purity of the 160 'by-product, as depicted as a requirement for the purity of plaster 1924 in Fig. 19.

[0254] Accordingly, once the maximum blower capacity limit is reached, the 1500A controller can control the operation of the WFGD 130 'subsystem to reduce the pH of the suspension 148' entering the absorption tower 132 and thus reduce SO2 absorption to the emission limit, i.e. permit requirement on SO2 emissions at the outlet of 1922. However, if further reduction of SO2 absorption would violate the SO2 emission permit requirement at outlet 1922 and the blower does not have sufficient capacity to provide the necessary amount of air (oxygen) to oxidize all absorbed SO2 to be removed, physical devices e.g. blower 150 and / or the crystallizer 134 is too small and neither SO2 removal requirements nor gypsum purity can be met. Because MPCC 1500 cannot "create" the additional oxygen required, an alternative strategy should be considered. As part of this alternative strategy, the 1500A controller controls the operation of the WFGD 130 'subsystem to maintain the current level of SO2 removal, i.e. meet the requirements of the SO2 1922 emission permit, and produce gypsum that meets the relaxed limitation of gypsum purity, i.e. meets the requirement for gypsum purity, which is smaller , than the requirement for plaster purity 1924. Preferably, the 1500A controller minimizes the deviation between the reduced gypsum purity requirement and the desired 1924 gypsum purity requirement. It should be understood that another alternative is for the 1500A controller to control the operation of the WFGD 130 'subsystem in accordance with a hybrid strategy that implements aspects of both of the above. These alternative control strategies can be performed by setting standard tuning parameters in MPCC 1500.

MPCC operations [0255] As described above, MPCC 1500 can control large WFGD subsystems for use within a distributed control system (DCS). The parameters that can be controlled with the MPCC 1500 are practically unlimited, but preferably include at least one or more of the following parameters: (1) pH of the suspension 148 'introduced into the absorption tower 132, (2) a series of suspension pumps that provides feeding the liquid suspension 148 'to various levels of the absorption tower 132 and (3) the amount of compressed oxidation air 154' introduced into the crystallizer 134. As you can see, this dynamic 870 control model contains basic process relationships that will be used by MPCC 1500 to control the WFGD process control. Accordingly, the relationships established in the dynamic 870 control model are of primary importance for the MPCC 1500. In this regard, the dynamic 870 control model applies to various parameters such as pH and air levels

Oxidizing, various restrictions such as levels of plaster purity and SO2 removal, and these relationships will allow dynamic and flexible control of the WFGD 130 'subsystem, as will be described in detail below.

[0256] Fig. 19 shows, in more detail, the preferred parameters and restrictions that are entered and used by the 1500A MPCC 1500 controller. As will be described below, controller 1500A performs prediction logic such as prediction logic 850 in accordance with dynamic control model 870 and based on input parameters and constraints to predict future WFGD process states and direct control of subsystem 130 ' so as to optimize the WFGD process. The 1500A then performs a logic to generate control, such as a logic to generate control 860, in accordance with the logic control directives for prediction to generate and send control signals to components specific to the control of the WFGD 130 'subsystem.

[0257] As previously described with reference to Fig. 15B, input parameters include measured process parameters 1525, parameters not related to process 1550, process limits WFGD 1555 and estimated parameters 1560 calculated using an MPCC 1500B estimator performing a logic system for estimating such as logic for estimating 840, according to the dynamic 880 estimation model.

[0258] In the preferred embodiment shown in Fig. 19, the measured process parameters 1525 include ambient conditions 1520, measured load of the power generating system (PGS) 1509, measured SO2 at inlet 189, blower load value 1502, measured pH of suspension 183, measured SO2 at outlet 1505, plaster purity measured in laboratory 1506, suspension pump condition 1512, and suspension pump load conditions 1515. The limitations of the WFGD 1555 process include the SO2 emission permit requirement at 1922, the plaster requirement 1924, the requirement for O2 dissolved in suspension 1926, and the requirement for pH in suspension 1928. Input signals not related to the 1550 process include tuning factors 1902, current credit price SO2 1904, current unit cost of energy 1906, current cost of organic acid 1908, current selling price of gypsum 1910 and future operational plans 1950. The estimated parameters 1560 calculated by the estimator 1500B include calculated gypsum purity 1932, calculated O2 dissolved in suspension 1934, and calculated suspension pH 1936.

Due to the inclusion of input parameters not related to the process, e.g. the current cost of the 1906 energy unit, MPCC 1500 can directly control the WFGD 130 'subsystem not only based on the current state of the process but also on the basis of matters outside the process.

Determining the Availability of SO2 Supplementary Absorption Capacity [0259] As discussed earlier with reference to Fig. 17, MPCC 1500 can control the state and load of 133A-133C pumps and thus control the flow of suspension 148 'to different levels of the absorption tower 132. MPCC 1500 can also calculate current energy consumption of the 133A-133C pumps based on the current pump ordering and current

-63 pump load values ​​1515A-1515C and, additionally, current operating costs of pumps based on calculated energy consumption and current unit cost of energy 1906.

[0260] MPCC 1500 is preferably configured to perform a predictive logic 850, in accordance with dynamic control model 870 and based on current pump condition values ​​1512A-1512C and current pump load values ​​1515A-1515C, to determine available additional pump capacities 133A -133C. The MPCC 1500 then determines, based on the determined amount of additional pump capacity available, the additional amount of SO2 that can be removed by regulating pump operation, e.g. switching on the pump for changing the order of pumps or increasing the pump power.

Determining the Extra SO2 Available To Remove [0261] As mentioned above, in addition to the measured SO2 content at inlet 189 detected by sensor 188, the load 1509 of the power generating system (PGS) 110 is preferably detected by load sensor 1508 and also entered as a measured parameter to MPCC 1500. The PGS 1509 load can, for example, reflect the BTU measure of coal consumed or the amount of energy produced by the 110 power generation system. However, PGS 1509 load may also reflect some other parameter of the power generating system 110 or associated power generation process, if only the measurement of this other parameter rationally corresponds to the inlet exhaust gas load, e.g. a parameter of the power generating system due to coal combustion or a process that rationally corresponds to the amount of exhaust at the entrance to the WFGD 130 'subsystem.

[0262] MPCC 1500 is preferably configured to perform prediction logic 850, in accordance with dynamic control model 870, to determine the exhaust gas load at the inlet, i.e. the volume or mass of the exhaust gas at the inlet 114, at the absorption tower 132 that corresponds to the PGS load 1509. The MPCC may, for example, calculate the exhaust gas inlet at the absorption tower 132 based on PGS 1509. Alternatively, the PGS 1509 load can itself be used as an inlet exhaust gas load, in which case no calculations are needed. In any case, MPCC 1500 will then determine the additional amount of SO2 that is available for removal from exhaust 114 based on the measured SO2 content at inlet 189, the exhaust gas load at the inlet, and the measured SO2 value at outlet 1505.

[0263] It will be appreciated that, if desired, the inlet exhaust gas load can be measured directly and entered into the MPCC 1500. This means that the actual measurement of the volume or mass of the exhaust gas at the inlet 114 directed to the absorption tower 132 can, optionally, be detected by a sensor (not shown) located in front of the absorption tower 132 and behind other APC 122 subsystems and be introduced into MPCC 1500. In this case, it may not be necessary for the MPCC 1500 to determine the exhaust gas inlet load that corresponds to PGS 1509.

Restrictive SO2 Removal Restrictions and Moving Average [0264] As described with reference to Fig. 12, the process history database 1210 includes the SO2 890 emission history database, for example described with reference to Fig. 8. Database

6464 process history data has interconnection with MPCC 1500. It should be understood that

MPCC 1500 may be of the type illustrated in, for example, Fig. 8 or it may be a multi-level controller such as the two-level controller shown in Fig.

10.

[0265] The SO2 emission history database 890 stores data reflecting SO2 emissions, not only in terms of SO2 content, but also in pounds of SO2 emitted, during the last moving average period. Thus, in addition to access to information reflecting current SO2 emissions by input of SO2 content measured at outlet 1505 from SO2 1504 analyzer, by interconnection with the 1210 process history database, MPCC 1500 also has access to historical information reflecting SO2 emissions, i.e. SO2 content measured at the outlet, during the last moving average of the time window by the SO2 890 emission history database. It should be understood that, although the current SO2 emissions correspond to a single value, the SO2 emissions over the last moving average time window correspond to the dynamic movement of SO2 emissions over the appropriate period.

Determining the Availability of Additional Oxidation Opportunity for SO2 [0266] As shown in Fig. 19 and discussed above, input signals to MPCC 1500 are measured values ​​(1) of SO2 content at outlet 1505, (2) of measured load of blower 1502 that corresponds to the amount of oxidizing air introduced into crystallizer 134, (3) the condition value of the slurry pump 1512, i.e. pump series and load values ​​of slurry pumps 1515 that correspond to the amount of limestone slurry flowing to the absorption tower 132, (4) measured pH 183 of the slurry flowing to the absorption tower 132. Additionally, MPCC 1500 introduces requirements for (1) purity 1924 by-product in the form of gypsum 160 ', (2) O2 1926 dissolved in a suspension in crystallizer 134, which corresponds to the amount of O2 dissolved in the suspension necessary to ensure sufficient oxidation and prevents masking of limestone and (3) SO2 at outlet 1922 in exhaust 116 'leaving the WFGD 130' subsystem. Currently, the SO2 1922 permit requirement typically includes restrictions on both instantaneous SO2 emissions and moving average SO2 emissions. MPCC 1500 has also introduced non-process input signals, including (1) 1906 unit cost of energy, e.g. electricity unit cost, and (2) current and / or predicted SO2 1904 loan price, which reflects the price for which such loans can be sold regulatory. In addition, MPCC 1500 calculates an estimate (1) of the current purity 1932 by-product in the form of 160 'gypsum, (2) O2 1934 dissolved in suspension in crystallizer 134 and (3) pH 1936 suspension flowing to the absorption tower 132.

[0267] MPCC 1500, performing a prediction logic according to the dynamic control program, processes these parameters to determine the amount of SO2 that has reacted with the suspension in the absorption tower 132. Based on this determination, MPCC 1500 can then determine the amount of undissolved O2 which remains available in suspension in the crystallizer 134 for oxidizing calcium sulfite to form calcium sulfate.

Determining Whether Additional Available Capacity Should Be Used [0268] If MPCC 1500 has determined that additional capacity is available to absorb and oxidize additional SO2, and there is additional SO2 available for removal, MPCC 1500 is also preferably configured to perform predictive logic 850, in accordance with the dynamic control model 870, to determine whether to control the WFGD 130 'subsystem, to regulate the operation of removing additional available SO2 from exhaust 114. To make this determination, MPCC 1500 can, for example, determine whether the production and sale of such SO2 credits will increase the cost-effectiveness of WFGD 130 'subsystem activities because it is more cost-effective to change removal activities additional SO2, besides required by working permits issued by relevant governmental regulatory units, i.e. otherwise required by the SO2 emission permit requirement at outlet 1922 and sell the regulatory credits received that will be obtained.

[0269] In particular, MPCC 1500, by performing prediction logic 850, in accordance with dynamic control model 870, will determine the necessary changes in the operation of the WFGD subsystem 130 'to increase SO2 removal. Based on this designation, MPCC 1500 will also calculate the number of additional regulatory loans that will be obtained. Based on designated operational changes and current or projected electricity cost, e.g. unit cost of energy, MPCC 1500 will designate the additional electricity costs obtained as required by the changes in WFGD 130 'subsystem activities that have been identified as necessary. Based on these last determinations and the current or predicted price of such loans, e.g. the SO2 1904 loan price, MPCC 1500 will then determine whether the cost of producing additional regulatory loans is greater than the price at which the loan can be sold.

[0270] If, for example, the loan price is low, the creation and sale of additional loans may not be beneficial. Removing SO2 at the smallest level necessary to meet the operating permits issued by relevant government regulatory bodies will rather minimize costs and thus maximize the cost-effectiveness of WFGD 130 'subsystem activities because it is more cost-effective to remove the amount of SO2 required to minimally meet SO2 emission permit requirements at outlet 1922 or working permits issued by relevant government regulatory bodies. If loans have already been generated under the current activities of the WFGD 130 'subsystem, MPCC 1500 may even direct changes in the activities of the WFGD 130' subsystem to reduce SO2 removal and thus stop further SO2 credit production and thereby reduce electricity costs and thus profitability actions.

Setting Work Priorities [0271] As also shown in Fig. 19, MPCC 1500 is also preferably configured to receive tuning factors 1902 as other input signals unrelated to the 1550 process. MPCC 1500, performing 850 prediction logic according to the dynamic model control 870 and tuning factors 1902, maybe

Set priorities for control variables using, for example, the appropriate meanings for each of the control variables.

[0272] Accordingly, preferably the 1555 constraints will establish, if desired, the required range for each limited parameter limit. Thus, for example, the SO2 emission permit requirement at 1922, gypsum purity requirement 1924, dissolved O2 content requirement 1926, and the pH requirement in suspension 1928 may have high and low limits, and MPCC 1500 will keep the WFGD 130 'subsystem within the range based on tuning factors 1902.

Assessment of Future WFGD Processes [0273] MPCC 1500, by performing prediction logic 850, in accordance with dynamic control model 870, preferably first evaluates the current state of process operation as discussed above. However, the assessment does not have to end here. MPCC 1500 is also preferably configured to perform logic prediction 850, in accordance with the dynamic process model 870, to assess whether process operations have shifted if no changes have been made to the activities of the WFGD 130 'subsystem.

[0274] More specifically, MPCC 1500 assesses the future state of the process based on relationships within dynamic control process 870 and process historical data stored in process history databases 1210. Process history data includes data in the SO2 history database as well as other data reflecting what has recently happened as part of the WFGD process over a certain pre-determined period. As part of this assessment, MPCC 1500 determines the current path on which the WFGD 130 'subsystem runs, and therefore the future values ​​of the various parameters associated with the WFGD process if no changes are made to the activities.

[0275] As will be understood by those skilled in the art, MPCC 1500 preferably determines, in a manner similar to that discussed above, the availability of additional SO2 absorption capacity, removable additional SO2, availability of additional SO2 oxidation capacity, and whether additional available available should be used ability based on determined future parameter values.

Implementation of the Working Strategy for the Activities of the WFGD Subsystem [0276] MPCC 1500 can also be used as a platform for implementing multiple working strategies without affecting the underlying process model and process control dependencies in the process model. MPCC 1500 uses goal functions to define the goals of the work. The goal function includes process information in terms of dependencies in process models, however, it also includes tuning or weight factors. Process dependencies reflected in the objective function by the process model are constant. Tuning factors can be adjusted before each controller execution. Subject to process limits and restrictions, the controller algorithm can maximize or minimize the value of the objective function to determine the optimal value of the objective function. Optimal operational goals for process values ​​are available to the controller from the optimal solution to the objective function. Adjusting the tuning factors or weights in the goal function changes the value of the goal function, and thus

-67 optimal solution. You can execute various working strategies using MPCC 1500 using the appropriate criteria or strategy to set the tuning constants of the goal function. Some more popular work strategies may include:

• Optimization of assets (maximizing profit / minimizing costs), • Maximizing pollution removal, • Minimizing the movements of variables manipulated in the problem of control.

Optimization of WFGD Subsystem Activities [0277] Based on the desired working criteria and the appropriately corrected objective function and tuning factors 1902, MPCC 1500 will perform a logic system for prediction 850, in accordance with the dynamic process model 870 and based on appropriate input or calculated parameters to first determine long-term work objectives for the WFGD 130 'subsystem. Thus, MPCC 1500 will map the appropriate course, such as optimal paths or paths, from the current state of process variables, for both manipulated and controlled variables, to the appropriate determination of long-term work goals for these process variables. MPCC 1500 then generates steering directives to modify the activities of the WFGD 130 'subsystem in accordance with its long-term work goals and optimal course mapping. Finally, the MPCC 1500, by executing the logic for generating the 860 control, generates and communicates the control signals to the WFGD 130 'subsystem, based on control directives.

[0278] Thus, MPCC 1500, in accordance with the dynamic control model 870 and the current measured and calculated parameter data, performs the first optimization of the WFGD 130 'subsystem based on the selected objective function, such as the one chosen based on current electricity costs or credit prices to determine the desired target stable state. MPCC 1500, in accordance with the dynamic control model 870 and process historical data, then performs a second optimization of the WFGD 130 'subsystem to determine the dynamic path along which it is necessary to move to process the variables from the current state to the desired stable target state. Preferably, the prediction logic implemented by MPCC 1500 determines a path that facilitates the control of WFGD 130 'subsystem operations by MPCC 1500 so as to shift process variables as quickly as practical to the desired target state of each process variable, while minimizing offset error between the desired target state of each process variable and the actual current state of each process variable at each point along the dynamic path.

[0279] Thus, MPCC 1500 solves the control problem not only for the current moment (T0), but at all other times during the period in which the process variables move from the current state in T0 to the target stable state in Tss. This enables the process variables to be moved for optimization by moving the entire path from the current state to the target stable state. This in turn provides additional stability

-68 compared to movements of process parameters using standard WFGD controllers such as PIDs described earlier in the background.

[0280] Optimized control of the WFGD subsystem is possible because the process relationships are included in the dynamic control model 870 and because changing the target function or non-process input signals such as economical input signals or tuning variables does not affect these relationships. Therefore, the way MPCC 1500 controls the WFGD 130 'subsystem and thus the WFGD process under various conditions, including different non-process conditions, can be manipulated or changed without further consideration of the process level after validation of the dynamic control model.

[0281] Referring again to Figs. 15A and 19, examples will be described of controlling the WFGD subsystem 130 'for the SO2 credit maximization function and for the objective function of maximizing cost-effectiveness or minimizing losses of the WFGD subsystem. It will be understood by specialists in this field that by creating tuning factors for other operating scenarios, it is possible to optimize, maximize or minimize other controlled parameters in the WFGD subsystem.

Maximizing SOg Loans [0282] To maximize SO2 loans, MPCC 1500 performs 850 prediction logic in accordance with the dynamic control model 870 having a target function with tuning constants configured to maximize SO2 loans. It will be understood that from the point of view of the WFGD process, maximizing SO2 loans requires that SO2 recovery is maximized.

[0283] Tuning constants that are introduced into the objective function will allow the objective function to offset changes in manipulated variables with respect to SO2 emissions relative to each other.

[0284] The net optimization results will include an increase by MPCC 1500:

• SO2 removal by increasing the pH of the suspension 186 'and • Amount of oxidizing air from the 154' blower to compensate for the additional amount of SO2 to be recovered.

• Subject to restrictions on:

• The lower limit of 1924 gypsum purity limit. It will be understood that this will usually be a value providing a small safety margin above the lowest allowable gypsum purity limit within the 1924 gypsum purity requirement.

• The lower limit of the required oxidizing air 154 'and • The maximum capacity of the oxidizing air blower 150.

[0285] In addition, if MPCC 1500 can adapt the alignment of pumps 133, MPCC 1500 will maximize suspension circulation and effective suspension height taking into account restrictions on the arrangement and loading of pumps 133.

[0286] As part of this working scenario, MPCC 1500 focuses entirely on increasing SO2 removal and SO2 credit production. MPCC 1500 will consider process limitations such as 1924 gypsum purity and oxidizing air requirements. However, this scenario does not provide for a balance between the price / value of electricity and the value of SO2 loans. This scenario would be appropriate if the value of SO2 loans would significantly exceed the cost / value of electricity.

Maximizing Cost-effectiveness or Minimizing Losses [0287] The objective function in MPCC 1500 can be configured to maximize profitability or minimize losses. This working scenario can be called the "asset optimization" scenario. This scenario also requires accurate and timely information on the cost / value of electricity, SO2 credits, limestone, gypsum and other additives such as organic acid.

[0288] The cost / value factors associated with each of the variables in the controller model are introduced into the objective function. Then the objective function in MPCC 1500 is directed to minimize cost / maximize profit. If profit is defined as a negative cost, then the cost / profit becomes a continuous function to be minimized by the objective function.

[0289] According to this scenario, the objective function will determine the minimum operating cost at the point where the marginal value of generating the additional SO2 loan is equal to the marginal cost of producing that loan. Note that the goal function is limited optimization, so the cost minimization solution will be limited in terms of:

• Minimal SO2 removal (for compliance with permits / emission targets), • Minimal gypsum purity, • Minimal requirement for oxidizing air, • Maximal blower load, • Pump arrangement and load limits, • Additive limits.

[0290] This operational scenario will be sensitive to changes in terms of both the value / cost of electricity and the value / cost of SO2 loans. For maximum profit, these cost factors should be updated in real time.

[0291] For example, assuming that the cost factors are updated before each 1500A controller is executed, because the demand for electricity increases during the day, the value with the instant supply of the electricity produced also increases. Assuming that public utilities can sell

-70 additional energy at this value with immediate delivery, and the value of SO2 loans is essentially constant at the moment, so if there is a way to transfer power from pumps 133 and blower 150 to the network, while still maintaining minimal SO2 removal, there is a significant economic incentive to introduce extra power to the network. The cost / value ratio of electricity as a function of the MPCC 1500 target will change as the value of electricity with immediate delivery changes and the objective function reaches a new solution that meets these operating constraints but uses less electricity.

[0292] Conversely, if the value of instant SO2 loan delivery increases, there is a market for additional SO2 loans and the cost / value of electricity is relatively constant, the objective function in MPCC 1500 will correspond to this change by increasing SO2 removal taking into account operational constraints.

[0293] In both example scenarios, MPCC 1500 will retain all operating constraints, and then the objective function in MPCC 1500 will look for the optimal operating point where the marginal value of the SO2 loan is equal to the marginal cost required to produce the loan.

Operation Not Feasible [0294] It is possible that sometimes WFGD 130 'will present a set of restrictions 1555 and operating conditions, measured 1525 and estimated 1560, for which there is no feasible solution; the area of ​​feasible solutions 525 as shown in Figs. 5A and 5B is the zero area. When this occurs, no solution will meet all of the 1555 system restrictions. This situation can be described as "impossible operation" because it is impossible to meet the system restrictions.

[0295] Impossible operation may be the result of operation beyond the capabilities of WFGD, the process is disrupted in WFGD or before WFGD. It may also be the result of overly restrictive, incorrect and / or incorrect 1555 restrictions in the WFGD and MPCC 1500 system.

[0296] During the period of impracticable operation, the objective function in MPCC 1500 focuses on minimizing weighted error. Each 1555 process limitation appears as a function of purpose. The weighing period is used for any error or violation of the limit limit by controlled / targeted process values. During commissioning of the 1500A controller, engineer (s) implementations choose appropriate values ​​for determining the weighting of errors so that during periods of impracticable the objective function will be "suspended" for the restrictions of the least importance to take into account the more important restrictions.

[0297] For example, in the WFGD 130 'subsystem, there are legal authorization limits associated with SO2 1505 outlet and sales specifications related to plaster purity 1506. Violations of SO2 emission permits have fines and other significant consequences. Violation of the sales specification for gypsum purity requires moving to a lower level or re-mixing the gypsum product. Moving the product to a lower level is not a desirable option, but it has less impact on the operating efficiency of the generation station than

-71 violation of an emission permit. Thus, the tuning factors will be set so that the limitation on the SO2 emission limit is more important, more weight than the limitation on gypsum purity. Thus, with these tuning factors, during periods of impracticable operation, the objective function in MPCC 1500 will advantageously keep SO2 emissions at or below the SO2 emission limit and violate the limitation of gypsum purity; MPCC 1500 will minimize violation of the plaster purity limitation, but will shift the impracticability to this variable to maintain a more significant emission limit.

Notification to Operators of Control Decisions [0298] MPCC 1500 is also preferably configured to provide notifications to operators about certain MPCC 1500 designations. Also here, logic for prediction 850, dynamic control model 870 or other programs can be used to configure MPCC 1500 to provide such notifications. For example, MPCC may act to direct the sound of alarms or to show text or other elements displayed on the monitors so that operators or other users are aware of certain MPCC 1500 markings, such as the indication that the quality of the remaining plaster has a low priority at a certain time because SO2 loans are so valuable.

WFGD Summary [0299] In summary, as described in detail above, optimization based control for the WFGD process is described. This control facilitates the manipulation of settings for the WFGD process in real time based on the optimization of a multi-input, multi-output model that is updated using process feedback. Optimization can take into account various goals and restrictions for the process. Without such control, the operator must specify the settings for WFGD. Due to the complexity of the process, the operator often chooses suboptimal settings to balance many restrictions and goals. Suboptimal settings / actions result in loss of removal efficiency, higher operating costs and potential violation of quality restrictions.

[0300] Virtual analysis for gypsum purity in the production line has also been described. The analysis allows you to calculate a direct estimate of the purity of the by-product in the form of gypsum produced in the WFGD process using measured process variables, laboratory analysis and a dynamic estimation model for gypsum purity. Since gypsum purity sensors in the production line are not standardly available, laboratory analysis outside the production line is commonly used to determine gypsum purity. However, since gypsum purity is only tested sporadically, and purity must be kept above the limit, usually set in the gypsum specification, process operators often use WFGD process settings, which means that gypsum purity is much higher than the required limit. This in turn makes concessions in terms of SO2 removal efficiency and / or unnecessary power consumption by the WFGD subsystem. Settings for the WFGD process can be controlled by estimating the gypsum purity in the production line, by

- 72 ensure that the gypsum purity is close to the purity reduction, and therefore, facilitate the increased SO2 removal efficiency.

[0301] As also described in detail above, virtual analysis of gypsum purity in the production line is carried out in a control loop, thus enabling the inclusion of estimates for closed-loop control, regardless of whether model predictive control (MPC) or PID control. By providing control loop feedback, SO2 removal efficiency can be increased by operating to produce gypsum with a purity closer to the corresponding purity reduction.

[0302] In addition, the virtual analysis in the operating costs production line has been described above. The analysis, as disclosed, uses WFGD process data as well as current market price data to remotely calculate the operating costs for the WFGD process. By default, operators do not include the current cost of the WFGD process. However, by calculating such costs remotely, operators can now track the effects of process changes, e.g. changes in operating cost settings.

[0303] In addition, the virtual remote operational cost analysis performance in the control loop is further described above, thus enabling the inclusion of closed system control estimates, regardless of whether MPC or PID is used. This closed system control can therefore be carried out to minimize operating costs.

[0304] The above also describes the technique of using MPC control to optimize the operation of the WFGD process to achieve maximum SO2 removal efficiency, minimum operating costs, and / or desired gypsum purity above the limit. Such control may use virtual analysis of gypsum purity and / or operating costs in the feedback loop as discussed above and can automatically optimize, for example SO2 removal performance and / or operating costs for the WFGD process.

[0305] Both necessary and optional parameters are described. Using the disclosed parameters, those skilled in the art can use well-known techniques in a routine manner to develop the appropriate model of the current WFGD process, which in turn can be used, for example, by MPCC 1550 to control the WFGD process to optimize the operation of the WFGD process. Models for gypsum purity, SO2 removal efficiency and / or operating costs as well as various other factors can be developed. Based on the WFGD process models developed in accordance with the principles, systems and processes described here, standard MPC or other logic circuits can be made to optimize the WFGD process. Thus, the limitations of traditional WFGD control processes have been overcome, for example using PIDs, which are limited to single-input / single-output designs and reliably rely on process feedback rather than process models. By incorporating models into a closed loop, WFGD process control can be even further increased to, for example,

73 keeping action closer to restrictions with less volatility than ever was possible.

[0306] The use of neural network based models in the WFGD process has also been described in the context of both process control and virtual remote analysis of the WFGD process. As described in detail above, the dependence of output and input signals in the WFGD process is non-linear, and therefore it is preferable to use a non-linear model, because such a model will best reflect the non-linearity of this process. In addition, the development of other models derived from data experienced from the WFGD process was also described.

[0307] The use of a combined model, which takes into account both basic principles data and the experimental process, for controlling and virtual analysis of the WFGD process is also described in detail above. Although some elements of the WFGD process are well understood and can be modeled using basic principles models, other elements are not so well understood and are therefore most conveniently modeled using experimental historical process data. By combining data from basic principles and the experimental process, you can quickly develop an accurate model without the need for a rolling test of all elements of the process.

[0308] The above also describes in detail the technique for validating the sensor measurements used in the WFGD process. As described, non-validated measurements can be replaced, thus avoiding improper control resulting from inaccurate sensor measurements in the WFGD process. By validating and replacing incorrect measurements, the WFGD process can now be carried out continuously, based on valid process data.

[0309] Rolling emissions have also been described in detail. Thus, in light of the disclosure, the WFGD process can be controlled so that one or more of the moving average of process emissions can be properly maintained. MPC can be performed using a single controller or multiple cascade controllers for process control. With the technique described, the WFGD process can be controlled, for example, in such a way that many moving averages are taken into account simultaneously and maintained, while operating costs are minimized at the same time.

SCR subsystem architecture:

[0310] The most important points of application of MPCC for SCR will be described to demonstrate the utility of the invention in other environments and embodiments. The main control objectives for SCR include:

• NOx removal for purposes of either regulatory compliance or asset optimization, • Ammonia residue control, and • Minimum operating cost - management of SCR catalyst and ammonia utilization.

[0311] Once again, a measurement and control methodology similar to that discussed for WFGD can be used:

Measurement: As mentioned, ammonia residues are an important control parameter that is often not measured. If no direct measurement of residual ammonia exists, it is possible to calculate residual ammonia based on NOx inlet and outlet measurements 2112 and 2111 and the flow of ammonia to SCR 2012. The accuracy of these measurements is uncertain because it requires accurate and reproducible measurements and involves assessing small differences between large quantities. Without direct measurement of ammonia residues, virtual remote analyzer techniques are used in addition to calculating ammonia residues directly to create an estimate of ammonia residues with greater fidelity.

[0312] The first step in VOA estimates the catalyst potential (response factor) and space velocity correlation variance (SVSV) in the SCR catalyst. They are calculated using the inlet gas flow, temperature, total catalyst run time, and NOx inlet and NOx outlet. Both the catalyst potential calculation and SVSV are time-averaged for many samples. The catalyst potential changes slowly, so many data points are used to calculate the potential, and SVSV changes more often, so relatively little data is used to calculate SVSV. Considering the catalyst potential (reaction coefficient), the volatility correlation variation (SVSV) and NOx at the inlet, the estimation of the ammonia residue can be calculated using the technique shown in Fig. 9.

[0313] If an ammonia residue sensor is available, the closed loop from such sensor to the process model will be used to automatically deviate the VOA. VOA would be used to significantly reduce the typically noisy output of a hardware sensor.

[0314] Finally, it should be noted that a virtual remote SCR operational cost analyzer can be used. As mentioned in the previous section, the operating cost model was developed based on basic principles. Operating costs can be calculated remotely, using the virtual remote analyzer again, and the technique shown in Fig. 9 is used for VOA.

[0315] Control: To achieve control objectives, MPCC was used for the SCR control problem. Fig. 22, similar to Fig. 8, shows the MPCC structure for MPCC 2500 in SCR. Because of the similarities to Fig. 8, a detailed discussion of Fig. 22 is not necessary because MPCC 2500 will be understood from the discussion of Fig. 8 above. Fig. 23A illustrates the use of MPCC 2500 in the SCR 2170 'subsystem. The biggest change in the SCR 2170 'subsystem control scheme is that the functionality of the NOx 2020 PID controller and the 2220 load feedback controller, each of which is shown in Fig. 20, has been replaced by MPCC 2500. MPCC

-752500 directly calculates the SP 2021A 'ammonia flow for use by the ammonia flow PID controller (s) (PID 2010).

[0316] MPCC 2500 can match one or more ammonia streams to control NOx removal efficiency and ammonia residues. Provided there are enough measurements from the 2003 and 2004 NOx inlet and outlet and residual ammonia 2611 measurements from the 2610 ammonia analyzer to determine NOx removal efficiency and ammonia profile information, MPCC 2500 will control the total and average NOx removal efficiency and ammonia residues as well as profile values. Coordinated control of many NOx removal efficiency values ​​and ammonia residue profile allows for significant reduction of variability of average process values. Lower variability translates into fewer "hot" breaks in the system. This profile control requires at least some form of profile measure and control of more than one NOx inlet, NOx outlet and measurement of residual ammonia, and more than one dynamically controlled flow of ammonia. It must be admitted that without the necessary input signals (measurements) and control execution (ammonia flows), MPCC 2500 will not be able to introduce profile control and capture the benefits that result from it.

[0317] From the MPCC 2500 perspective, additional parameters related to profile control increase the size of the controller, but the overall methodology, scheme and control objectives are unchanged. Therefore, future discussions will include controlling the SCR subsystem without profile control.

[0318] Fig. 23B shows an overview of MPCC 2500.

Optimization of the SCR Subsystem Activities [0319] Based on the desired operating criteria and the appropriately tuned objective function and tuning factors 2902, MPCC 2500 will perform the 2850 prediction logic in accordance with the dynamic 2870 control model and based on appropriate input or calculated parameters to first set long-term working goals for the SCR 2170 'subsystem. MPCC 2500 will then map the optimal course, such as appropriate tracks or paths, based on the current state of process variables, for both manipulated and controlled variables to properly set long-term working goals for these process variables. MPCC 2500 then generates control directives to modify the activities of the SCR 2170 'subsystem in accordance with established long-term work objectives and optimal waveform mapping. Finally, MPCC 2500, by executing the logic for generating the 2860 control, generates and communicates the control signals of the SCR 2170 'subsystem based on the control directives.

[0320] MPCC 2500, in accordance with a dynamic control model and measured and calculated parameter data, therefore carries out the first optimization of SCR 2170 'subsystem operations based on a selected objective function such as the one chosen on the basis of current electricity costs or the price of the regulatory loan for

Determining the desired target steady state. MPCC 2500, in accordance with the dynamic control model and process historical data, then performs a second optimization of SCR 2170 'subsystem operations to determine the dynamic path along which process variables should be moved from the current state to the desired target stable state. Preferably, the prediction logic implemented by MPCC 2500 defines a path that will facilitate the control of SCR 2170 'subsystem operations by MPCC 2500 so as to shift process variables as quickly as possible to the desired target state of each process variable, minimizing error or offset between the desired target state of each process variable and the actual current state of each process variable at each point along the dynamic path.

[0321] Thus, MPCC 2500 solves the control problem not only for the current moment of time (T0), but at all other times of the time when process variables pass from the current state at T0 to the stable target state at Tss. This allows the process variables to be shifted for optimization by traveling the entire path from the current state to the target stable state. This in turn provides additional stability compared to process parameter shifts using standard PID controllers such as the previously described PID.

[0322] Optimized control of the SCR subsystem is possible because the process dependencies were made in the dynamic control model 2870, and because changing the purpose function of non-process input signals such as economical input signals or tuning variables does not affect these relationships. Thus, it is possible to manipulate or change the way MPCC 2500 controls the SCR 2170 'subsystem, and thus the SCR process, under different conditions, including different non-process conditions, without further considering the process level after validating the dynamic control model.

[0323] Referring again to Figs. 23A and 23B, examples of SCR subsystem control 2170 'will be described for the objective of maximizing NOx credits and the objective function of maximizing cost-effectiveness or minimizing losses of operation of the SCR subsystem. It will be understood by those skilled in the art that by creating tuning factors through other operating scenarios, it is possible to optimize, maximize or minimize other controlled parameters in the SCR subsystem.

Maximization of NO loans_x [0324] To maximize NOx credits, MPCC 2500 performs 2850 prediction logic according to the 2870 dynamic control model having a target function with tuning constants configured to maximize NOx credits. It will be understood that from the SCR process point of view, maximizing NOx credits requires that NOx recovery is maximized.

[0325] Tuning constants that are introduced into the objective function will allow the objective function to offset the impact of changes in manipulated variables regarding NOx emissions. [0326] The net results of the optimization will be that MPCC 2500 will increase:

-77 • NOx removal by increasing the ammonia flow setting (s) subject to restrictions:

• Maximum residual ammonia.

[0327] In this work scenario, MPCC 2500 focuses entirely on increasing NOx removal for the production of NOx credits. The MPCC will take into account the process limitation in terms of ammonia residues. However, this scenario does not provide for a balance between the cost / value of ammonia or residual ammonia and the value of NOx credits. This scenario would be appropriate if the value of NOx credits significantly exceeds the cost / value of ammonia and residual ammonia.

Maximizing Profitability or Minimizing Losses [0328] The objective function in MPCC 2500 can be configured to maximize profit or minimize losses. This working scenario can be called the "asset optimization" scenario. This scenario also requires accurate and timely information on the cost / value of electricity, NOx credits, ammonia and the impact of ammonia residues on the equipment at the bottom.

[0329] Cost / value factors associated with each of the variables in the controller model are introduced into the objective function. Thus, the objective function in MPCC 2500 is focused on minimizing costs / maximizing profit. If profit is defined as a negative cost, the cost / profit ratio becomes a continuous function for the objective function to be minimized.

[0330] According to this scenario, the objective function will determine the operation with the minimum cost at the point where the marginal value of generating an additional NOx loan is equal to the marginal cost of producing that loan. It should be noted that the goal function is limited optimization, so minimizing costs will be subject to restrictions on:

• Minimal NOx removal (for compliance with permits / emission targets), • Minimal ammonia residues, • Minimizing ammonia consumption.

[0331] This work scenario will be sensitive to changes in both the value / cost ratio of electricity and the value / cost ratio of NOx credits. For maximum benefits, these cost factors should be updated in real time.

[0332] For example, assuming that cost factors are updated before each controller is made, as the demand for electricity increases during the day, the value with instant delivery of the electricity produced also increases. Assuming that the plant is able to sell additional energy at this value with immediate delivery and the value of NOx credits is basically constant at the moment, there is a significant incentive for minimizing ammonia residues, as this will allow greater purity of the air heater and more efficient energy production. There is a significant economic incentive to introduce an additional one

-78 energy to the network. Cost / value ratio related to electricity as a function of goal

The MPCC 2500 will change as the value of electricity changes with immediate delivery and the objective function reaches a new solution that meets the operating limitations but uses less electricity.

[0333] Conversely, if the value of NOx loan with immediate delivery increases, there is a market for additional NOx loans and the electricity cost / value ratio is relatively constant, the objective function in MPCC 2500 will respond to this change by increasing NOx removal subject to operational restrictions .

[0334] In both example scenarios, MPCC 2500 will meet all operating constraints, and therefore the objective function in MPCC 2500 will look for the optimal operating point where the marginal value of the NOx loan is equal to the marginal cost required to produce the loan.

Mirosława Ważyńska

Patent Attorney