Industrial data anomaly detection method, device, electronic device and storage medium

By using a neural network judgment model with an auto-encoding architecture in industrial data exception detection, combined with a classification model, the historical operation data of industrial equipment is trained, and the effective reconstruction and judgment of abnormal data is achieved, and the judgment in the existing technology is solved The problem of large errors improves robustness and production efficiency.

CN114580534BActive Publication Date: 2024-11-19WUHAN UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210209643.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-03
Publication Date
2024-11-19
Estimated Expiration
2042-03-03

AI Technical Summary

Technical Problem

The existing industrial data abnormality detection technology cannot effectively reconstruct abnormal data, resulting in large judgment errors and cannot meet the needs of industrial production.

Method used

By obtaining the historical operation data of industrial equipment, dividing the training data set, creating an initial classification model and neural network judgment model, training and combining the neural network judgment model with the auto-coding architecture, real-time data judgment and reconstruction are carried out.

Benefits of technology

It improves the robustness and accuracy of abnormal judgments of industrial data, reduces the false alarm rate, and enhances industrial production efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114580534B_ABST
    Figure CN114580534B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, electronic device and storage medium for detecting anomalies in industrial data, the method comprising: obtaining historical operation data of industrial equipment, obtaining a training data set according to the historical operation data, and dividing the training data set into a first training set and a second training set; creating an initial classification model and an initial neural network judgment model; inputting the first and second training sets into the initial classification model and the initial neural network judgment model for training, respectively, to obtain a fully trained classification model and a fully trained neural network judgment model; obtaining real-time industrial operation data, and inputting them into the fully trained classification model and the neural network judgment model, respectively, to determine whether the real-time industrial operation data is abnormal. The present invention improves the robustness and accuracy of judging abnormalities in industrial data, and while effectively judging abnormal data, also completes the reconstruction of data, has data cleaning capabilities, reduces the false alarm rate, and improves production efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of industrial data detection technology, and in particular to an industrial data anomaly detection method, device, electronic equipment and storage medium. Background Art

[0002] With the widespread popularization of Internet of Things (IOT) technology, a large amount of equipment operation data is generated in the industrial production process. Industrial data has the characteristics of large data capacity, strong correlation, and closed loop. Due to interference from the external environment, sensor errors and zero drift, temperature sources, signal sources, etc., some abnormal values ​​will be generated in the collected data. If the abnormal data is misjudged, the production equipment in good operating condition will be misjudged as faulty equipment, affecting the management work of the operation and maintenance personnel, and even causing large-scale downtime of the entire production line. Therefore, accurate judgment of abnormal data plays a very important role in the industrial production process.

[0003] At present, the anomaly detection method for industrial data is mainly to directly detect abnormal points by setting thresholds, or to use the 3sigma rule or box plot method to screen outliers. This type of statistical anomaly detection algorithm can effectively determine the anomaly of outliers, but it does not have the ability to distinguish complex situations such as time series anomalies and state anomalies. In addition, the determination of the threshold needs to rely on labeled data, and the more labeled data there is, the better the quality, and the more accurate the threshold. However, industrial data cannot provide a considerable amount of labeled data. In actual applications, the number of normal samples is much larger than the number of abnormal samples. In addition, abnormal data can often be reconstructed into normal data. Intelligent cleaning of abnormal data can reduce a large number of equipment failure alarm information and improve production efficiency.

[0004] Therefore, the existing industrial data anomaly detection technology can no longer meet the application needs of the existing industrial end. There are problems such as the inability to reconstruct abnormal data and large data judgment errors. It is necessary to design an industrial data anomaly detection method to solve the above problems. Summary of the invention

[0005] In view of this, it is necessary to provide an industrial data anomaly detection method, device, electronic device and computer-readable storage device to solve the problems in the prior art that abnormal data cannot be reconstructed and data judgment errors are large.

[0006] In order to solve the above problems, the present invention provides an industrial data anomaly detection method, comprising:

[0007] Acquire historical operation data of industrial equipment, obtain a training data set according to the historical operation data, and divide the training data set into a first training set and a second training set;

[0008] Create an initial classification model and an initial neural network judgment model;

[0009] Inputting the first training set into the initial classification model for training, and inputting the second training set into the initial neural network judgment model for training, to obtain a fully trained classification model and a fully trained neural network judgment model;

[0010] Real-time industrial operation data is acquired, and the real-time industrial operation data is respectively input into the well-trained classification model and the neural network judgment model to determine whether the real-time industrial operation data is abnormal.

[0011] Furthermore, the real-time industrial operation data is input into the well-trained classification model and the neural network judgment model respectively to judge whether the real-time industrial operation data is abnormal, including:

[0012] Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively to obtain a model classification result and a model judgment result;

[0013] Performing logical operations on the model classification results and the model judgment results to obtain a final judgment result;

[0014] According to the final judgment result, it is judged whether the real-time industrial operation data is abnormal.

[0015] Furthermore, the real-time industrial operation data is input into the neural network judgment model to obtain a model judgment result, including:

[0016] Inputting the real-time industrial operation data into the neural network judgment model to judge whether the real-time industrial operation data meets the abnormal data judgment condition;

[0017] When the real-time industrial operation data meets the abnormal data judgment condition, determining the real-time industrial operation data as data to be processed;

[0018] Reconstructing the data to be processed by using the neural network judgment model to obtain reconstructed operating data;

[0019] A model judgment result is obtained according to the real-time industrial operation data and the reconstructed operation data.

[0020] Furthermore, the initial neural network judgment model is established based on a fully connected network, including a diffusion neural network, a generative neural network, a conditional auxiliary neural network, a sparse dimensionality reduction neural network and a sparse generative neural network;

[0021] The diffusion neural network and the sparse dimension reduction neural network are arranged in parallel;

[0022] The diffusion neural network, the generative neural network, the sparse dimension reduction neural network and the sparse generative neural network are all logically connected to the conditional auxiliary network;

[0023] The generation network is logically connected with the diffusion network and the sparse generation network;

[0024] The sparse generation network and the sparse dimension reduction neural network are logically connected.

[0025] Furthermore, the second training set is input into the initial neural network judgment model for training to obtain a fully trained neural network judgment model, including:

[0026] Determine whether the trained neural network judgment model meets the preset performance standard; if the preset performance standard is met, output the neural network judgment model as a fully trained neural network judgment model, and obtain the abnormal data judgment condition.

[0027] Furthermore, the historical operation data is normalized using a first preset algorithm to obtain a training data set.

[0028] Furthermore, inputting the first training set into the initial classification model for training includes:

[0029] Constructing an optimization objective function of the initial classification model;

[0030] The optimization objective function is solved using a second preset algorithm and the training data set to obtain an optimal solution to the optimization objective function.

[0031] The present invention also provides an industrial data anomaly detection device, comprising:

[0032] A training set acquisition module, which acquires historical operation data of industrial equipment, obtains a training data set according to the historical operation data, and divides the training data set into a first training set and a second training set;

[0033] A model creation module is used to create an initial classification model and an initial neural network judgment model;

[0034] A model training module, used for inputting the first training set into the initial classification model for training, and inputting the second training set into the initial neural network judgment model for training, to obtain a fully trained classification model and a fully trained neural network judgment model;

[0035] The data judgment module is used to obtain real-time industrial operation data, input the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively, and judge whether the real-time industrial operation data is abnormal.

[0036] The present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, an industrial data anomaly detection method as described in any of the above technical solutions is implemented.

[0037] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, an industrial data anomaly detection method as described in any of the above technical solutions is implemented.

[0038] Compared with the prior art, the beneficial effects of the present invention include: the method of the present invention first obtains the first and second training sets based on the historical operation data of the industrial equipment, and uses the first and second training sets to respectively train the initial classification model and the initial neural network judgment model to obtain a fully trained classification model and a neural network judgment model. The present invention creates an initial classification model and an initial neural network judgment model based on the characteristics of the industrial equipment operation data, and uses the training set obtained from the historical industrial equipment operation data to train the two models to obtain a fully trained classification model and a neural network judgment model, and constructs a framework for anomaly detection of industrial big data, which greatly improves the robustness and accuracy of anomaly judgment of industrial data. The present invention designs a neural network judgment model with an auto-encoding architecture, which not only effectively judges abnormal data, but also completes the reconstruction of abnormal data, has data cleaning capabilities, reduces the false alarm rate, and improves industrial production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a flow chart of an embodiment of an industrial data anomaly detection method provided by the present invention;

[0040] Figure 2 A schematic diagram of the structure of a neural network judgment model provided by the present invention;

[0041] Figure 3 A flowchart of an embodiment of a specific algorithm for detecting anomalies in industrial data provided by the present invention;

[0042] Figure 4 An overall average loss function diagram of an embodiment of a neural network judgment model provided by the present invention;

[0043] Figure 5 A schematic diagram of the structure of an industrial data anomaly detection device provided by the present invention;

[0044] Figure 6 This is a structural block diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0045] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0046] The present invention provides an industrial data anomaly detection method, device, electronic device and computer-readable storage medium, which are described in detail below.

[0047] The embodiment of the present invention provides a method for detecting anomalies in industrial data, and the flowchart thereof is as follows: Figure 1 As shown, including:

[0048] Step S101, acquiring historical operation data of industrial equipment, obtaining a training data set according to the historical operation data, and dividing the training data set into a first training set and a second training set;

[0049] Step S102, creating an initial classification model and an initial neural network judgment model;

[0050] Step S103, inputting the first training set into the initial classification model for training, and inputting the second training set into the initial neural network judgment model for training, to obtain a fully trained classification model and a fully trained neural network judgment model;

[0051] Step S104: acquiring real-time industrial operation data, and inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively to determine whether the real-time industrial operation data is abnormal.

[0052] Compared with the prior art, the industrial data anomaly detection method provided in this embodiment creates an initial classification model and an initial neural network judgment model according to the characteristics of industrial equipment operation data, and trains the two models using a training set obtained from historical industrial equipment operation data to obtain a fully trained classification model and a neural network judgment model, thereby constructing a framework for anomaly detection of industrial big data, greatly improving the robustness and accuracy of industrial data anomaly judgment. The present invention, by designing a neural network judgment model with an auto-encoding architecture, not only effectively judges abnormal data, but also completes the reconstruction of abnormal data, has data cleaning capabilities, reduces the false alarm rate, and improves industrial production efficiency.

[0053] As a specific embodiment, in step S101, the historical operation data of industrial equipment is collected from a SCADA or DCS system; the historical operation data of industrial equipment includes a large amount of historical normal operation data of industrial equipment and a small amount of operation data with abnormal labels; the method of this embodiment makes full use of the real historical operation data of industrial equipment and does not need to rely on a large number of samples marked as abnormal data to improve the judgment accuracy of the model.

[0054] As a preferred embodiment, obtaining a training data set according to the historical operation data includes:

[0055] The historical operation data is normalized using a first preset algorithm to obtain a training data set.

[0056] As a specific embodiment, the training data set is divided into Data A and Data B, where Data A is a data set with a balanced number of normal data and abnormal data samples; Data B is a data set without abnormal data. The normal / abnormal label features of Data A are used as binary classification targets and divided into training set A-1 and test set A-2. The initial classification model is trained with the training set DataA, and the initial neural network judgment model is trained with Data B. Therefore, a large amount of labeled data is not required for the training of the neural network judgment model.

[0057] As a specific embodiment, the first preset algorithm is MinMax normalization processing. The data normalization method of MinMax adopts a custom MinMax normalization function in specific applications. First, based on the 3Ϭ principle, the 3Ϭ value of each dimensional feature is obtained, and the maximum and minimum 3Ϭ values ​​are used as the maximum and minimum values ​​to normalize the dimensional feature.

[0058] As a preferred embodiment, in step S102, the initial neural network judgment model is established based on a fully connected network, and the initial neural network judgment model includes a diffusion neural network, a generative neural network, a conditional auxiliary neural network, a sparse dimensionality reduction neural network and a sparse generative neural network;

[0059] The diffusion neural network and the sparse dimension reduction neural network are arranged in parallel;

[0060] The diffusion neural network, the generative neural network, the sparse dimension reduction neural network and the sparse generative neural network are all logically connected to the conditional auxiliary network;

[0061] The generation network is logically connected with the diffusion network and the sparse generation network;

[0062] The sparse generation network and the sparse dimension reduction neural network are logically connected.

[0063] The neural network judgment model of this embodiment adopts the AE self-encoding architecture, which can effectively judge abnormal samples and reconstruct abnormal samples, so that the neural network judgment model has data cleaning capabilities. The stability of the model is expanded by using a multi-layer neural network in parallel, and the label data is input into the neural network, which can better prevent the neural network from overfitting, which is conducive to activating more neurons, and can improve network performance without deepening the depth and width of the neural network.

[0064] As a preferred embodiment, in step S103, the second training set is input into the initial neural network judgment model for training to obtain a fully trained neural network judgment model, including:

[0065] Determine whether the trained neural network judgment model meets the preset performance standard; if the preset performance standard is met, output the neural network judgment model as a fully trained neural network judgment model, and obtain the abnormal data judgment condition.

[0066] As a specific embodiment, when the neural network judgment model training converges, the training is stopped to obtain a fully trained neural network judgment model. The last training loss after convergence is used as the judgment threshold for judging whether the data is abnormal. The judgment model can automatically define the error threshold for judging abnormalities based on the data generated during the neural network training process, and use the automatically adjusted threshold as the basis for judging whether the input data of the neural network judgment model is abnormal.

[0067] As a specific example, Figure 2 As shown, Figure 2 The schematic diagram of the structure of the neural network judgment model is a dual neural network parallel structure: a diffusion network and a sparse dimension reduction network. The training data are respectively input into the diffusion network and the sparse dimension reduction network to obtain the processing results R1 data and R2 data, and the R1 data and R2 data are input into the conditional auxiliary network.

[0068] The conditional auxiliary network performs matrix multiplication operation on R1 data and R2 data to obtain the operation result R3 data, obtains the classification label information (Label) of R3 data according to the abnormal data judgment threshold, and fuses the classification label information with R3 data to obtain R4 data.

[0069] The output R4 data of the conditional auxiliary network and the output R1 data of the diffusion network are fused and input into the generation network to obtain R5 data; the R4 data and the output R2 data of the sparse dimensionality reduction network are fused and input into the sparse generation network to obtain R6 data; the corresponding features of the R5 data and the R6 data are averaged to obtain the reconstructed data.

[0070] The root mean square error (RMSE) is combined with the classification error and cross entropy as the reconstruction error. The reconstructed data is used as the result of the forward propagation of the neural network, compared with the original data, and the Loss value is calculated and back-propagated. The RMSE is used as the optimization target to train the neural network judgment model. The last training Loss after the training converges is obtained as the threshold for judging whether the data is abnormal data.

[0071] As a preferred embodiment, in step S103, inputting the first training set into the initial classification model for training includes:

[0072] Constructing an optimization objective function of the initial classification model;

[0073] The optimization objective function is solved using a second preset algorithm and the training data set to obtain an optimal solution to the optimization objective function.

[0074] As a specific embodiment, the initial classification model is constructed based on XGBOOST; XGBOOST with higher computational efficiency is used as a supervised learning algorithm to reduce the time complexity of the model.

[0075] First, set the parameters of XGBOOST:

[0076] Set the general parameters of XGBOOST to:

[0077] 'booster':'gbtree','nthread':4,'num_feature':4, 'seed':1000;

[0078] Set the task parameters of XGBOOST to:

[0079] 'objective':'binary:logistic','gamma':0.1,'max_depth':20,'lambda':2,'subsample':0.8,'colsample_bytree':1,'min_child_weight':3,'eta':0.001,'nthread':2,'eval_metric':'error';

[0080] Secondly, the grid hyperparameter search (CV) method is used to search for key parameters and adjust the XGBOOST hyperparameters to obtain the best classification model;

[0081] Input Data A into the initial classification model and use K-fold cross-validation to train Data A;

[0082] As a specific embodiment, the specific solution process of the second preset algorithm is as follows:

[0083] The optimal model is established by minimizing the loss function of the training set, that is, formula (1):

[0084] (1)

[0085] in, represents the complexity of the model, is the complexity coefficient, M represents the number of all samples, and L is the name of the loss function. represents the target value, Represents the predicted value. The optimization objective function of the classification model in this embodiment consists of two parts: a loss function and a regular term representing the complexity of the model, and the classification model includes multiple CART trees, which are expressed as:

[0086] (2)

[0087] in, represents the optimization objective function value, L represents the loss function, It is obtained by adding the regularization terms of J trees. represents the target value, Representation and Asymptotically equal. From the forward step-by-step algorithm, we know that the structure of the first t-1 trees is constant , we can get:

[0088] (3)

[0089] After ignoring the constant term Constant and after a series of Taylor series expansions, combined with the complexity formula (4) of multiple CART trees, formula (5) can be derived:

[0090] (4)

[0091] (5)

[0092] In formula (4), T is the number of leaf nodes, is the modulus of the leaf node vector, is the difficulty of node segmentation, is the regularization coefficient. In formula (5), the function and Used to represent the loss function L and the predicted value The mathematical relationship of .

[0093] For the CART tree, we use Represents the leaf node number of the output, then Represents the output value, namely:

[0094] (6)

[0095] Further simplifying formula (5) yields:

[0096] (7)

[0097] It can be seen that the first part of formula (7) is the cumulative sum of all training set samples. At this time, all samples can be regarded as the mapping of leaf nodes.

[0098] If we start from the leaf node and accumulate, we can get:

[0099] (8)

[0100] make

[0101] (9)

[0102] , They represent the sum of the first-order derivative and the sum of the second-order derivative of all input samples of leaf node j, respectively. Substituting them into formula (8), we get

[0103] (10)

[0104] For the structure of the tth tree, the leaf nodes are independent of each other, that is and To determine the quantity, equation (10) can be regarded as a quadratic equation of w, and after differentiation, we get:

[0105] (11)

[0106] Here It can be regarded as the minimum value. Substituting it into formula (11) we get the final objective function:

[0107] (12)

[0108] Formula (12) is also called a scoring function, which can be used to measure the quality of the CART tree structure. The smaller the value, the better the tree structure.

[0109] As a preferred embodiment, in step S104, the real-time industrial operation data is respectively input into the well-trained classification model and the neural network judgment model to judge whether the real-time industrial operation data is abnormal, including:

[0110] Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively to obtain a model classification result and a model judgment result;

[0111] Performing logical operations on the model classification results and the model judgment results to obtain a final judgment result;

[0112] According to the final judgment result, it is judged whether the real-time industrial operation data is abnormal.

[0113] As a specific embodiment, the model classification result and the model judgment result are subjected to an AND or OR logic operation. For example, when both the model classification result and the model judgment result are data abnormalities, the real-time industrial operation data is determined to be abnormal data; when the model classification result is data abnormality and the model judgment result is normal, the real-time industrial operation data is determined to be normal data.

[0114] As a preferred embodiment, the real-time industrial operation data is input into the neural network judgment model to obtain a model judgment result, including:

[0115] Inputting the real-time industrial operation data into the neural network judgment model to judge whether the real-time industrial operation data meets the abnormal data judgment condition;

[0116] When the real-time industrial operation data meets the abnormal data judgment condition, determining the real-time industrial operation data as data to be processed;

[0117] The neural network judgment model is used to reconstruct the data to be processed to obtain reconstructed operating data.

[0118] As a specific embodiment, the judgment process of industrial data is as follows:

[0119] Input the collected raw data into the classification model and the neural network judgment model respectively;

[0120] The classification model and the neural network judgment model judge the original data;

[0121] The classification results of the neural network judgment model and the classification model are logically ANDed or operated, and the result after the operation is used as the judgment result of whether the original data is abnormal or not. When the model classification result and the model judgment result are both data abnormal, the real-time industrial operation data is determined to be abnormal data; when the model classification result is data abnormality and the model judgment result is normal, the real-time industrial operation data is determined to be normal data.

[0122] Wherein, the judgment method of the neural network judgment model is:

[0123] Determine whether the original data has anomalies; if the original data has no anomalies through the neural network judgment model, directly collect the original data, store the data in the historical database, and update the neural network judgment model in real time; if the original data is abnormal data, reconstruct the original data, and collect the reconstructed data as the processed data, and judge the reconstructed data through the neural network judgment model to obtain the model judgment result.

[0124] Combine the following Figure 3 This solution is described in detail with a specific application example:

[0125] This embodiment collects real industrial data as experimental data, builds a classification model and a neural network judgment model, further explains the method of the above technical solution in detail, and also verifies the effectiveness of the technical solution.

[0126] Step S201: obtaining original data, where an ultra-wideband (UWB) indoor positioning data set is used, and the anomaly is the data anomaly caused by environmental occlusion;

[0127] Step S202: Divide the data set into Data A and Data B, where Data A is a data set with a balanced number of normal data and abnormal data samples, and Data B is a data set without abnormal data. The division of the data set is shown in Table 1:

[0128] Table 1

[0129]

[0130] Step S203: construct an abnormal classification model based on XGBOOST, and use a grid hyperparameter search method to adjust the XGBOOST hyperparameters to obtain a model with the best classification effect;

[0131] Step S204: Set the parameters of XGBoost, including general parameters (General Parameters), model parameters (Booster Parameters) and learning task parameters (Learning Task Parameters). This embodiment performs grid search tuning on the two hyperparameters of shrinkage parameters eta and num_boost_round. eta refers to the learning rate, which is used to multiply the coefficient when updating the weight of the leaf node to avoid excessive step size. The larger the parameter value, the less likely it is to converge. The default value in the XGBoost method is 0.3. num_boost_round refers to the number of benchmark trees for the boosting iteration, which is used to control the number of iterations. The settings of other parameters are shown in Table 2;

[0132] Table 2

[0133]

[0134] Step S205: Use the normal / abnormal label features of Data A as the binary classification target, divide the training set and the test set, and perform MinMax normalization on the input data of the training set and the test set. Here, pandas is used to group each feature first, and then a custom 3Ϭ value is used for each group to obtain the maximum and minimum values, and each dimension is normalized with the maximum and minimum values. The normalized Data A data is used to train the XGBOOST-based abnormal classification model to obtain a fully trained classification model;

[0135] Step S206: Construct Figure 2 The conditional sparse diffusion autoencoder neural network (CSSAE) shown in the figure contains five neural networks: diffusion neural network, generation neural network, conditional auxiliary neural network, sparse dimension reduction neural network, and sparse generation neural network.

[0136] Step S207: MinMax normalize Data B (the same method as step S205); train the CSSAE network using the Data B data, such as Figure 4 As shown, Figure 4 is the overall average loss function of the neural network judgment model described in this embodiment;

[0137] Step S208: obtaining the last training Loss after training convergence (i.e., after training is completed) as a threshold for determining whether the data is abnormal data;

[0138] Step S209: input the real-time industrial data into CSSAE and XGBOOST respectively to obtain preliminary analysis results and model classification results respectively;

[0139] Step S210: If the error of the preliminary analysis result of the real-time industrial data after CSSAE is less than or equal to the threshold, the model judges that the data is normal, and proceeds to step S212; if the preliminary analysis result is greater than the threshold, proceeds to step S211;

[0140] Step S211: collect the reconstructed data after real-time industrial data reconstruction as processed data, and judge whether the reconstruction error of the data is greater than the threshold. If the reconstruction error is less than the threshold, the model judges that the data is normal, and stores the real-time industrial data in the historical database, and updates the CSSAE neural network in real time; if the reconstruction error is greater than the threshold, the model judges that the data is abnormal;

[0141] Step S212: Perform a logical AND or OR operation on the model judgment result of CSSAE and the classification result of XGBOOST, and use the result of the operation as the judgment result of whether the data is abnormal or not.

[0142] The embodiment of the present invention provides an industrial data anomaly detection device, the structural block diagram of which is as follows Figure 5 As shown, the industrial data anomaly detection device 500 includes:

[0143] A training set acquisition module 501 acquires historical operation data of industrial equipment, obtains a training data set according to the historical operation data, and divides the training data set into a first training set and a second training set;

[0144] A model creation module 502 is used to create an initial classification model and an initial neural network judgment model;

[0145] A model training module 503 is used to input the first training set into the initial classification model for training, and input the second training set into the initial neural network judgment model for training, so as to obtain a fully trained classification model and a fully trained neural network judgment model;

[0146] The data judgment module 504 is used to obtain real-time industrial operation data, input the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively, and judge whether the real-time industrial operation data is abnormal.

[0147] like Figure 6 As shown, the present invention also provides an electronic device 600, which can be a computing device such as a mobile terminal, a desktop computer, a notebook, a palm computer, and a server. The electronic device includes a processor 601, a memory 602, and a display 603.

[0148] In some embodiments, the memory 602 may be an internal storage unit of a computer device, such as a hard disk or memory of a computer device. In other embodiments, the memory 602 may also be an external storage device of a computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the memory 602 may also include both an internal storage unit of a computer device and an external storage device. The memory 602 is used to store application software and various types of data installed on the computer device, such as program codes for installing the computer device. The memory 602 may also be used to temporarily store data that has been output or is to be output. In one embodiment, an industrial data anomaly detection method program 604 is stored on the memory 602, and the industrial data anomaly detection method program 604 can be executed by the processor 601, thereby realizing an industrial data anomaly detection method of each embodiment of the present invention.

[0149] In some embodiments, the processor 601 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes stored in the memory 602 or process data, such as executing an industrial data anomaly detection program.

[0150] In some embodiments, the display 603 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 603 is used to display information on the computer device and to display a visual user interface. The components 601-603 of the computer device communicate with each other through a system bus.

[0151] This embodiment also provides a computer-readable storage medium, on which a program of an industrial data anomaly detection method is stored. When a processor executes the program, the industrial data anomaly detection method as described above is implemented.

[0152] The computer-readable storage medium and computing device provided according to the above embodiments of the present invention can be implemented with reference to the specific description of the industrial data anomaly detection method as described above according to the present invention, and have similar beneficial effects as the industrial data anomaly detection method as described above, which will not be repeated here.

[0153] The present invention discloses an industrial data anomaly detection method, device, electronic device and computer-readable storage medium. First, a first and a second training set are obtained according to the historical operation data of the industrial equipment. The first and the second training sets are used to train the initial classification model and the initial neural network judgment model respectively to obtain a fully trained classification model and a neural network judgment model.

[0154] According to the characteristics of industrial equipment operation data, the present invention creates an initial classification model and an initial neural network judgment model, and uses the training set obtained from historical industrial equipment operation data to train the two models, thereby obtaining a fully trained classification model and a neural network judgment model, and constructing a framework for abnormal detection of industrial big data, which greatly improves the robustness and accuracy of abnormal judgment of industrial data. By designing a neural network judgment model with an auto-encoding architecture, the present invention not only effectively judges abnormal data, but also completes the reconstruction of abnormal data, has data cleaning capabilities, reduces the false alarm rate, and improves industrial production efficiency.

[0155] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for detecting anomalies in industrial data, characterized in that: include: Acquire historical operation data of industrial equipment, obtain a training data set according to the historical operation data, and divide the training data set into a first training set and a second training set; Create an initial classification model and an initial neural network judgment model; Inputting the first training set into the initial classification model for training, and inputting the second training set into the initial neural network judgment model for training, to obtain a fully trained classification model and a fully trained neural network judgment model; Acquire real-time industrial operation data, and input the real-time industrial operation data into the well-trained classification model and the neural network judgment model, respectively, to determine whether the real-time industrial operation data is abnormal; Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively, and judging whether the real-time industrial operation data is abnormal, including: Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively to obtain a model classification result and a model judgment result; Performing logical operations on the model classification results and the model judgment results to obtain a final judgment result; According to the final judgment result, judging whether the real-time industrial operation data is abnormal; Inputting the real-time industrial operation data into the neural network judgment model to obtain a model judgment result includes: Inputting the real-time industrial operation data into the neural network judgment model to judge whether the real-time industrial operation data meets the abnormal data judgment condition; When the real-time industrial operation data meets the abnormal data judgment condition, determining the real-time industrial operation data as data to be processed; Reconstructing the data to be processed by using the neural network judgment model to obtain reconstructed operating data; Obtaining a model judgment result according to the real-time industrial operation data and the reconstructed operation data; The initial neural network judgment model is established based on a fully connected network, including a diffusion neural network, a generative neural network, a conditional auxiliary neural network, a sparse dimensionality reduction neural network and a sparse generative neural network; The diffusion neural network and the sparse dimension reduction neural network are arranged in parallel; The diffusion neural network, the generative neural network, the sparse dimension reduction neural network and the sparse generative neural network are all logically connected to the conditional auxiliary neural network; The generative neural network is logically connected with the diffusion neural network and the sparse generative neural network; The sparse generative neural network and the sparse dimensionality reduction neural network are logically connected.

2. The industrial data anomaly detection method according to claim 1, characterized in that: Inputting the second training set into the initial neural network judgment model for training to obtain a fully trained neural network judgment model, including: Determine whether the trained neural network judgment model meets the preset performance standard; if the preset performance standard is met, output the neural network judgment model as a fully trained neural network judgment model, and obtain the abnormal data judgment condition.

3. The industrial data anomaly detection method according to claim 1, characterized in that: A training data set is obtained according to the historical operation data, including: The historical operation data is normalized using a first preset algorithm to obtain a training data set.

4. The industrial data anomaly detection method according to claim 1, characterized in that: Inputting the first training set into the initial classification model for training includes: Constructing an optimization objective function of the initial classification model; The optimization objective function is solved using a second preset algorithm and the training data set to obtain an optimal solution to the optimization objective function.

5. An industrial data anomaly detection device, characterized in that: include: A training set acquisition module, which acquires historical operation data of industrial equipment, obtains a training data set according to the historical operation data, and divides the training data set into a first training set and a second training set; A model creation module is used to create an initial classification model and an initial neural network judgment model; A model training module, used for inputting the first training set into the initial classification model for training, and inputting the second training set into the initial neural network judgment model for training, to obtain a fully trained classification model and a fully trained neural network judgment model; A data judgment module, used for acquiring real-time industrial operation data, inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model, respectively, to judge whether the real-time industrial operation data is abnormal; Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively, and judging whether the real-time industrial operation data is abnormal, including: Inputting the real-time industrial operation data into the well-trained classification model and the neural network judgment model respectively to obtain a model classification result and a model judgment result; Performing logical operations on the model classification results and the model judgment results to obtain a final judgment result; According to the final judgment result, judging whether the real-time industrial operation data is abnormal; Inputting the real-time industrial operation data into the neural network judgment model to obtain a model judgment result includes: Inputting the real-time industrial operation data into the neural network judgment model to judge whether the real-time industrial operation data meets the abnormal data judgment condition; When the real-time industrial operation data meets the abnormal data judgment condition, determining the real-time industrial operation data as data to be processed; Reconstructing the data to be processed by using the neural network judgment model to obtain reconstructed operating data; Obtaining a model judgment result according to the real-time industrial operation data and the reconstructed operation data; The initial neural network judgment model is established based on a fully connected network, including a diffusion neural network, a generative neural network, a conditional auxiliary neural network, a sparse dimensionality reduction neural network and a sparse generative neural network; The diffusion neural network and the sparse dimension reduction neural network are arranged in parallel; The diffusion neural network, the generative neural network, the sparse dimension reduction neural network and the sparse generative neural network are all logically connected to the conditional auxiliary neural network; The generative neural network is logically connected with the diffusion neural network and the sparse generative neural network; The sparse generative neural network and the sparse dimensionality reduction neural network are logically connected.

6. An electronic device, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the industrial data anomaly detection method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the industrial data anomaly detection method as described in any one of claims 1-4 is implemented.

Citation Information

Patent Citations

  • Network information system operation and maintenance abnormity intelligent detection method based on reinforcement learning

    CN113780432A

  • Method and device for detecting data abnormality

    WO2021120775A1