Real-time equipment abnormality detection device and method based on time sequence prediction model

Real-time equipment abnormality detection device and method based on time series prediction modelTechnical field

[0001] The present invention relates to the technical field of equipment abnormality detection, and in particular to a real-time equipment abnormality detection device and method based on a time series prediction model.

Background technique

[0002] Judging the abnormal operation of equipment based on data has been fully applied in manufacturing enterprises. In summary, the technical solutions applied by enterprises can be divided into the following two types: judgments based on empirical thresholds and abnormal judgments based on machine learning. The two schemes are briefly introduced as follows: 1. Abnormal judgment based on empirical threshold, as shown in Figure 1. Figure 1 is the basic framework of equipment operation abnormal diagnosis based on empirical threshold. The framework includes 3 parts: 1) Threshold setting: Before the real-time system debugging and running, through the interactive module, the system operation engineer needs to input the operating experience threshold of the production system into the logic controller. 2) The implementation of anomaly detection system operation includes 4 parts: 1. Production system: The production system is the source of the entire system's operating data. The operating logic of equipment operating abnormality diagnosis based on empirical thresholds is to analyze the operating mode of the data to operate the system. State to judge. Therefore, it is necessary to transmit the equipment data generated by the production system to the data layer by adding a data acquisition system; 2. Data layer: The data layer includes two parts, real-time data and historical data. The real-time data layer needs to complete real-time data preprocessing and other operations. The main function of historical data is to display the running status of the system to the display module; 3. Logic control layer: The logic control layer is the core of the operation of the entire system. Its main function is to match real-time operating data with empirical threshold settings, and pass the output results to the result display module. 4. Display layer: The display layer includes the key parameter display of system operation and the judgment result of the abnormal state of the system operation state by the logic control layer. However, the abnormal judgment based on the experience threshold has the following shortcomings: 1) The operation is too dependent on the experience of the system operator. In the state where the threshold setting is deviated or unreasonable, a judgment will be deviated, which will affect the operation of the equipment by the system operator in severe cases; 2) The operating conditions of the system will change during the operation of the system, and an empirical threshold will be given The abnormal state judgment system does not have an effective mechanism to judge the normal operating condition changes and abnormal conditions of the equipment. 2. Abnormal judgment based on machine learning: As shown in Figure 2, Figure 2 is a machine learning-based abnormal diagnosis framework for equipment operation, including the following two processes: 1) Model training process: The basic idea of machine learning is from historical data Discover the basic patterns of the data in the machine learning tasks (classification or regression, etc.) of the new data according to the generated model; 2) Model prediction process: The process of model prediction is to apply the model generated during the training process to the production system The data generated by the device in the device, and the output results are passed to the display layer; 3), the result display process: the result display layer displays the results of the model prediction to the user. However, the abnormal diagnosis of equipment operation based on machine learning exists such as The following shortcomings: 1) The abnormal diagnosis model based on traditional machine learning algorithms can only run time slice data, that is, traditional machine learning algorithms cannot apply the data before and after time slice to analyze the current state of the running equipment; 2), and The framework of abnormality judgment based on empirical thresholds is similar. The operating conditions of the system will change during operation, while the abnormality judgment framework based on traditional machine learning algorithms cannot update the pattern of changes in operating conditions in time, so it will be based on the results of the abnormal state of the equipment. Deviations or errors occur.

Summary of the invention

[0003] The purpose of the present invention is to solve the problems mentioned in the background art section above through a real-time device anomaly detection device and method based on a time series prediction model.

[0004] To achieve this objective, the present invention adopts the following technical solutions:

[0005] A real-time device anomaly detection device based on a time series prediction model, the device includes: a data layer, a logic control layer, a model center, and a display layer; the data layer includes a real-time database, a data buffer, a historical database, and time series data Processing module; the logic control layer includes a trainer and a predictor; the model center includes a machine learning model; the display layer includes a result display module; wherein, the real-time database is used to store the operation of each device in the production system The generated data; the data buffer is used to buffer the data of the preset time window; the historical database is used to store the historical data of the operation of each device in the production system; the time series data processing module is used to: 1. For the production system The historical data of each device operation is processed, and then, through the trainer, the time series prediction model is trained according to the time series data processed by the time series data processing module, and the trained time series prediction model is saved to the model center; 2. Cache the slave data The data imported by the predictor is processed, and then the processed data and the time series prediction model of the model center are imported into the predictor at the same time, the result of the time series data is predicted, and the predictor transmits the result to the result display module for display.

[0006] In particular, the data layer also includes an incremental database; during the operation of the device, the data in the data buffer is periodically synchronized to the incremental database.

[0007] In particular, the trainer is also used to: in a fixed time interval, the time series prediction model is trained by incremental data in the incremental database, and the trained model is updated to the model center for application by the predictor .

[0008] The present invention also discloses a real-time equipment abnormality detection method based on a time-series prediction model. The method is implemented based on the above-mentioned real-time equipment abnormality detection device based on the time-series prediction model and includes the following steps:

[0009] S101. Initial model training: S1011. Import the historical data of the operation of each device in the production system into the time series data processing module for processing; S1012. The time series prediction model is trained according to the time series data processed by the time series data processing module, and the training is performed The subsequent time series prediction model is saved to the model center;

[0010] S102, equipment status diagnosis: S1021, update the data generated during the operation of each equipment in the production system to the real-time database through the data collection equipment; S1022, the data buffer caches the data of the preset time window; S1023, the slave data The data imported by the buffer is processed, and then the processed data and the time series prediction model of the model center are imported into the predictor at the same time, and the result of the time series data is predicted; S1024, the predictor transmits the result of the prediction in step S1023 to the result display module for execution Show.

[0011] In particular, the real-time device abnormality detection method based on the time series prediction model further includes:

[0012] S103, incremental training of the model: S1031, during the operation of the device, periodically synchronize the data in the data buffer to the incremental database; S1032, in a fixed time interval, the time series prediction model passes the incremental database Incremental data is trained, and the trained model is updated to the model center for predictor application.

[0013] Compared with the traditional equipment anomaly detection solution, the real-time equipment anomaly detection device and method based on the time series prediction model proposed by the present invention can use the information of a certain time window during the operation of the equipment to make the system The accuracy of the model is higher, and the result is more in line with reality; the present invention supports an incremental training mode, so that the model can capture the operating conditions of different devices in time, and then update the machine learning model of the model center in time.

Description of the drawings

[0014] FIG. 1 is a schematic diagram of the basic framework of equipment operation abnormal diagnosis based on empirical threshold;

[0015] FIG. 2 is a schematic diagram of a machine learning-based equipment operation abnormal diagnosis framework;

[0016] FIG. 3 is a schematic diagram of a real-time device anomaly detection framework based on a time series prediction model of the present invention.

Detailed ways

[0017] The present invention will be further described below with reference to the drawings and embodiments. It can be understood that the specific embodiments described here are only used to explain the present invention, but not to limit the present invention. In addition, it should be noted that, for ease of description, the drawings only show part but not all of the content related to the present invention. Unless otherwise defined, all technical and scientific terms used herein are related to the technical field of the present invention. The technical staff usually understand the same meaning. The terminology used herein is only to describe specific embodiments, and is not intended to limit the present invention.

[0018] Embodiment One

[0019] The real-time equipment anomaly detection device based on the time series prediction model in this embodiment specifically includes: a data layer, a logic control layer, a model center, and a display layer; the data layer includes a real-time database, a data buffer, a historical database, and time series data Processing module. The logic control layer includes a trainer and a predictor; the model center includes a machine learning model; and the display layer includes a result display module.

[0020] In this embodiment, the real-time database is used to store data generated during the operation of each device in the production system; wherein, the data generated during the operation of each device in the production system is collected by a data acquisition device. In this embodiment, the data buffer is used to buffer the data of a preset time window; the process of running the time series prediction model needs to pass the data of a certain time window to the predictor, so a data buffer needs to be set in the data layer to buffer Data for a certain time window. The historical database is used to store historical data of the operation of each device in the production system.

[0021] In this embodiment, the time-series data processing module is used to: 1. Process the historical data of the operation of each device in the production system, such as data cleaning and missing value filling, etc., and then, through the trainer, the time-series prediction model follows Train the time series data processed by the time series data processing module, and save the trained time series prediction model to the model center; 2. Process the data imported from the data buffer, such as data cleaning and missing value filling, etc., and then at the same time The processed data and the time series prediction model in the center of the model are imported into the predictor, the result of the time series data is predicted, and the predictor transmits the result to the result display module for display.

[0022] In this embodiment, in order to enable the model to capture the operating conditions of different equipment in time, and to update the machine learning model of the model center in time, the real-time equipment anomaly detection device based on the time series prediction model also sets an incremental database. During the running process, the data in the data buffer needs to be synchronized to the incremental database on a regular basis. In order to adapt to changes in equipment operating conditions, in a fixed time interval, the time series prediction model is trained through incremental data in the incremental database, and the trained model is updated to the model center for predictor application.

[0023] Embodiment Two

[0024] As shown in FIG. 3, FIG. 3 is a schematic diagram of a real-time device anomaly detection framework based on the time series prediction model of the present invention. The numbers 1- in the figure correspond to the following S1011, S1012, S1021, S1022, S1023, S1024, S1031, S1032, respectively. .

[0025] This embodiment provides a real-time device anomaly detection method based on a time-series prediction model. The method is implemented based on the above-mentioned real-time device anomaly detection device based on the time-series prediction model and includes the following steps:

[0026] S101. Initial model training: S1011. Import the historical data of the operation of each device in the production system into the time series data processing module for processing; S1012. The time series prediction model is trained according to the time series data processed by the time series data processing module, and the training is performed The subsequent time series prediction model is saved to the model center.

[0027] S102, equipment status diagnosis: S1021, update the data generated during the operation of each equipment in the production system to the real-time database through the data collection equipment; S1022, the data buffer caches the data of the preset time window; S1023, the slave data The data imported by the buffer is processed, and then the processed data and the time series prediction model of the model center are imported into the predictor at the same time, and the result of the time series data is predicted; S1024, the predictor transmits the result of the prediction in step S1023 to the result exhibition

Display module for display.

[0028] In this embodiment, in order to enable the model to capture the operating conditions of different devices in time, and to update the machine learning model of the model center in time, the real-time device abnormality detection method based on the time series prediction model further includes:

[0029] S103. Incremental training of the model: S1031. During the operation of the equipment, periodically synchronize the data in the data buffer to the incremental database; S1032, in order to adapt to changes in equipment operating conditions, within a fixed time interval , The time series prediction model is trained through incremental data in the incremental database, and the trained model is updated to the model center for predictor application.

[0030] The technical solution of the present invention uses the time sequence prediction model to use information of a certain time window during the operation of the device, so that the model accuracy of the system is higher, and the result is more in line with reality; the present invention supports an incremental training mode, so that the model can Timely capture the operating status of different equipment, and then update the machine learning model of the model center in time.

[0031] A person of ordinary skill in the art can understand that all or part of the above embodiments can be implemented by instructing relevant hardware through a computer program. The program can be stored in a computer readable storage medium. During execution, it may include the procedures of the above-mentioned method embodiments. Wherein, the storage medium can be a magnetic disk, an optical disc, a read-only storage memory or a random storage memory, etc.

[0032] Note that the above are only the preferred embodiments of the present invention and the applied technical principles. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments and substitutions can be made to those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and can also include more other equivalent embodiments without departing from the concept of the present invention. The scope of is determined by the scope of the appended claims.

CN 109934412 A Attached drawings

<img file="CN109934412A_D0001.tif" />

<img file="CN109934412A_D0002.tif" />