An improved adversarial auto-encoding network-based batch process fault detection method

Abstract

The application discloses a kind of intermittent process fault detection methods based on improved adversarial auto-encoding network, belong to intermittent industrial process monitoring technical field, it includes: under the normal running state of intermittent process, collect multiple batches of process data, and the data collected are preprocessed, and the data after preprocessing are used as training sample;Increase prediction loss term in the loss function of adversarial auto-encoding network, and construct time window enhanced adversarial auto-encoding network;Wherein, prediction loss term is obtained based on pre-trained time series predictor;Training sample is used to train time window enhanced adversarial auto-encoding network;Intermittent process fault detection is realized using trained time window enhanced adversarial auto-encoding network.The application further considers the before and after dependence of time series on the basis of conventional adversarial auto-encoding network, when processing different batch data of intermittent process, it has better adaptability, improves the accuracy of fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of intermittent industrial process monitoring, and in particular to an intermittent process fault detection method based on an improved adversarial autoencoder network. Background Art

[0002] Batch processing is a production method widely used in modern chemical industry, biotechnology, food processing, and pharmaceutical industries. Unlike continuous processes, batch processes are cyclical operations, typically using "batches" as the basic production unit. Each batch operation includes the addition of raw materials, adjustment of reaction conditions, and product extraction, all of which are completed at specific time points. This production method can flexibly adapt to the production needs of a variety of products and is particularly suitable for production environments with small outputs, diverse product types, or special requirements. In the chemical industry, batch processes are commonly used in synthetic reactions, catalytic processes, and the production of fine chemicals. In these processes, the addition of reactants and the regulation of temperature and pressure are crucial to the quality of the final product. Batch processes are also widely used in biotechnology fields such as fermentation, enzyme catalysis, and cell culture, as these processes require precise control of microbial growth conditions to improve product yield and quality. In the food processing industry, particularly in winemaking and dairy production, batch processes are a key method for ensuring product flavor and quality consistency. However, a notable characteristic of batch processes is their discontinuous and cyclical nature. Within each batch, changes in raw materials, environmental fluctuations, and differences in operating conditions can lead to variability in production results. The unpredictability and complexity of these changes make monitoring and control a huge challenge. Especially in modern automated production, how to effectively detect and diagnose faults in intermittent processes in real time has become a key issue in improving production efficiency and ensuring product quality.

[0003] Fault detection in batch processes begins with collecting process data such as temperature, pressure, and flow through sensors. This data is then preprocessed, including denoising, imputing missing values, and normalizing. Next, feature extraction methods are used to extract key features from the data. These features provide input for model training, helping to distinguish between normal and abnormal operation. The trained fault detection model is then able to monitor new data in real time, promptly detect abnormal signals, and perform fault diagnosis. As data complexity and dynamism increase, traditional methods are struggling to meet these demands, especially when dealing with large-scale, complex data. To address these challenges, deep learning methods, particularly neural network models, are currently a key technology in batch process fault detection. They can automatically learn complex nonlinear features from large amounts of data and effectively capture temporal dependencies and hidden patterns in the data, significantly improving the accuracy and robustness of fault detection. Common deep learning models include deep belief networks (DBNs), autoencoders (AEs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). However, while these deep learning methods can effectively extract features from complex data, they still face challenges in fault detection in batch processes. First, batch processes have significant dynamic characteristics, and each batch of operations introduces certain variations, which places adaptability demands on traditional deep learning models. Traditional models often struggle to adapt to these inter-batch fluctuations, especially when faced with new, unseen batches, often failing to make effective predictions. Furthermore, as the process evolves, the data distribution can shift significantly, leaving traditional models lacking sufficient generalization capabilities. In this context, researchers have proposed several improved methods, such as the Adversarial Autoencoder (AAE), which aims to enhance the model's feature learning and generation capabilities through adversarial training, thereby improving its robustness. Nevertheless, existing unsupervised AAE methods still face difficulties in addressing the challenges of extracting features from dynamic data, particularly in the context of batch processes, where time dependencies are evident. These models often lack built-in mechanisms to handle temporal dependencies in time series, making them unable to effectively respond to changes in time series data. Therefore, further optimization and improvement are still needed for practical applications. Summary of the Invention

[0004] The present invention provides an intermittent process fault detection method based on an improved adversarial autoencoder network to solve the technical problem that the existing intermittent process fault detection method cannot efficiently respond to changes in time series data, resulting in unsatisfactory detection results.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In one aspect, the present invention provides a method for detecting intermittent process faults based on an improved adversarial autoencoder network, comprising:

[0007] Under normal operation of the intermittent process, multiple batches of process data are collected, and the collected process data are preprocessed, and the preprocessed process data are used as training samples;

[0008] Adding a prediction loss term to the loss function of the adversarial autoencoder network to construct a time window enhanced adversarial autoencoder network; wherein the prediction loss term is obtained based on a pre-trained time series predictor;

[0009] Using training samples to train the time window enhanced adversarial autoencoder network;

[0010] Intermittent process fault detection is achieved using trained time window enhanced adversarial autoencoder networks.

[0011] Furthermore, the process data of multiple batches are collected during the normal operation of the intermittent process, and the collected process data are preprocessed, and the preprocessed process data are used as training samples, including:

[0012] During the normal operation of the intermittent process, multiple batches of process data are collected and represented as a three-dimensional matrix of I×J×K; where I represents the data batch, J represents the number of observed variables, and K represents the number of sampling times;

[0013] The collected process data in the form of a three-dimensional matrix is ​​expanded in the batch direction to obtain a two-dimensional data matrix, and the matrix is ​​standardized in the batch direction to eliminate the influence of different dimensions.

[0014] The standardized two-dimensional data matrix is ​​rearranged to obtain two-dimensional time series matrix data expanded according to the observed variables; wherein the data in each column of the two-dimensional time series matrix data reflects the change of a certain variable over time; the two-dimensional time series matrix data is used as a training sample.

[0015] Furthermore, the training process of the time series predictor includes:

[0016] Use a sliding time window to divide the data in the training sample and convert the continuous time series data into multiple time slices to obtain windowed data; each time slice consists of observation data within the corresponding time period;

[0017] Construct a time series predictor, use the windowed data as the input of the time series predictor, calculate the mean square error between the output data of the time series predictor and the actual data, and perform the back propagation algorithm to determine the final network parameters of the time series predictor, and obtain a trained time series predictor.

[0018] Furthermore, the predicted loss item is obtained as follows:

[0019] The reconstructed output of the adversarial autoencoder network is input into the pre-trained time series predictor, the mean square error between the output of the time series predictor and the actual data is calculated, and the calculation result is used as the prediction loss term.

[0020] Furthermore, the loss function of the time window enhanced adversarial autoencoder network is calculated as follows:

[0021] Calculate the mean square error between the reconstructed output of the adversarial autoencoder network and the original input of the adversarial autoencoder network, and use the mean square error between the reconstructed output and the original input as the reconstruction loss term;

[0022] The sum of the prediction loss term and the reconstruction loss term is used as the loss function of the time window enhanced adversarial autoencoder network.

[0023] Furthermore, the method of using the trained time window enhanced adversarial autoencoder network to implement intermittent process fault detection includes:

[0024] Based on the trained time window enhanced adversarial autoencoder network, the normal working condition data in the feature space and reconstruction space are counted, the corresponding statistics are calculated and their upper limits are determined, and the control threshold is set;

[0025] Obtain the process data of the current intermittent process and preprocess it. Input the preprocessed current process data into the trained time window enhanced adversarial autoencoder network to obtain the statistics of the current process data.

[0026] Compare the statistics of current process data with the control threshold to complete fault detection.

[0027] Furthermore, the method of enhancing the adversarial autoencoder network based on the trained time window, counting the normal operating condition data in the feature space and the reconstruction space, calculating the corresponding statistics and determining the upper limit thereof, and setting the control threshold, includes:

[0028] Input the training sample into the time window enhanced adversarial autoencoder network and calculate the T corresponding to each normal sample 2 After calculating the statistics and SPE statistics, the fault detection control limit T is determined using the kernel density estimation method. α 2 and SPE α .

[0029] Furthermore, the comparison of the statistics of the current process data with the control threshold to complete the fault detection includes:

[0030] If any statistic of the current process data exceeds the corresponding fault detection control limit, it is determined that a fault has occurred.

[0031] In still another aspect, the present application also provides an electronic device, comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above method.

[0032] In still another aspect, the present application also provides a computer-readable storage medium, which stores at least one instruction, which is loaded and executed by a processor to implement the above method.

[0033] The technical scheme provided by the present application has at least the following beneficial effects:

[0034] The method fully considers the complexity and dynamic characteristics of the batch industrial process data, effectively extracts the feature distribution in the process data by enhancing the design of the adversarial autoencoder through the time window, and realizes the extraction of dynamic features in the fault detection process without increasing the model complexity. According to the extracted internal time-dependent relationship of the data, the method can cope with the changes of process data in different stages, thereby realizing high-precision detection of process faults. Compared with existing fault detection methods, the method not only improves the accuracy of fault detection, but also enhances the sensitivity to different fault modes, providing a more effective and feasible solution for batch fault detection in complex industrial processes. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0036] Figure 1 is the execution flow diagram of the batch process fault detection method based on the improved adversarial auto-encoding network provided by the embodiment of the present application;

[0037] Figure 2 is the T 2 and SPE fault detection diagram of the penicillin fermentation process fault data with batch number 4 to which the method of the present application is applied;

[0038] Figure 3 is the system block diagram of the electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0040] First, it should be noted that in the embodiments of the present application, the words "exemplary", "for example", and the like are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments or designs. Rather, use of the word exemplary is intended to present concepts in a concrete manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be either one of the two.

[0041] First embodiment

[0042] The present embodiment provides an intermittent process fault detection method based on improved adversarial auto-encoding network, aiming to overcome the shortcomings of existing adversarial auto-encoding network in intermittent process monitoring application, which proposes a time window enhanced adversarial auto-encoding network, aiming at the dynamic nature and rapid change between batches of intermittent process, extracts the predictable dynamic characteristics in the process behavior, so as to improve the accuracy of fault detection. The method can be realized by an electronic device, which can be a terminal or a server. The execution flow of the method is as shown in the figure, which introduces the related idea of predictability feature extraction method in multivariate statistics into the feature extraction method of adversarial auto-encoding network, and introduces the predictability loss into the training process of adversarial auto-encoding network. Figure 1 The present embodiment provides an intermittent process fault detection method based on improved adversarial auto-encoding network, aiming to overcome the shortcomings of existing adversarial auto-encoding network in intermittent process monitoring application, which proposes a time window enhanced adversarial auto-encoding network, aiming at the dynamic nature and rapid change between batches of intermittent process, extracts the predictable dynamic characteristics in the process behavior, so as to improve the accuracy of fault detection. The method can be realized by an electronic device, which can be a terminal or a server. The execution flow of the method is as shown in the figure, which introduces the related idea of predictability feature extraction method in multivariate statistics into the feature extraction method of adversarial auto-encoding network, and introduces the predictability loss into the training process of adversarial auto-encoding network.

[0043] Specifically, the intermittent process fault detection method of the present embodiment comprises the following steps:

[0044] S1, under the normal running state of the intermittent process, collecting process data of multiple batches, and pre-processing the collected process data, taking the pre-processed process data as training samples;

[0045] Specifically, the implementation process of S1 is as follows:

[0046] S11, data collection: under the normal running state of the intermittent process, collecting process data of multiple batches as training data, and expressing the collected data as a three-dimensional matrix of I×J×K, wherein I represents the batch number, J represents the number of observation variables, and K represents the sampling times.

[0047] S12, data preprocessing, specifically including:

[0048] S121, standardization processing: expanding the three-dimensional data samples collected from the intermittent process according to the batch direction to obtain a two-dimensional data matrix X b =[x1,x2,...,x i ] T ∈R I×KJ , wherein x i =[x i,1 ,xi,2 ,...,x i,KJ ] T , i=1,2,...,I ​​is a column vector. It is normalized in the batch direction as follows:

[0049]

[0050] The above standardization process can eliminate the influence of different dimensions, making the data more comparable. and σ i are batch data matrices X b The column mean and standard deviation of column i. After standardization, each column has a mean of 0 and a variance of 1.

[0051] S122, generate training data: rearrange the standardized two-dimensional matrix to obtain a two-dimensional time series matrix X expanded by the observed variables t =[x1,x2,...,x J ] T ∈R IK×J , where x j ,j=1,2,…,J is X t The column vector of , each column of data reflects the change of a certain variable over time. By arranging the time series, each column of data can reflect the change of the process, providing a basis for subsequent modeling.

[0052] S2, adding a prediction loss term to the loss function of the adversarial autoencoder network to construct a time window enhanced adversarial autoencoder network; wherein the prediction loss term is obtained based on a pre-trained time series predictor;

[0053] Specifically, the implementation process of the above S2 is as follows:

[0054] S21, establish a time series vector prediction network: Use a sliding time window to divide the data, use the divided three-dimensional time window data to train the time series predictor, extract its time dependency, and predict future process changes. Specifically, the process of establishing the time series vector prediction network is as follows:

[0055] S211, Windowed training data input: Divide a batch of continuous time series data into time periods (windows). Specifically, for each batch of input data, a fixed time window length s is selected, that is, each window contains data from s consecutive moments before a certain moment t. In this way, we can convert continuous time data into multiple time segments. For example, for batch data X t =[x1,x2,...,x J ] T ∈RIK×J , X i =[x i-s+1 ,...,x i-1 ,x i ] T That is, it represents the time window composed of the observation data of time t = i and the previous s consecutive moments. This batch of data contains a total of (IK-s) time windows, so the new data form X can be obtained. w =[X s ,X s+1 ,...,X IK-1 ] T ∈R (IK-s)×s×J , where X j ∈R s×J ,j=1,2,...,IK-s is the time window at different moments.

[0056] S212, training time series predictor: And initialize the weight parameter β i , bias r, the order is consistent with the time window length, where the column vector x k is the observation data collected at time t = k, x k-i with x k have the same dimensions, is the model prediction result at time t=k; so the windowed data X w As prediction input, the network will output the prediction result matrix The prediction model can be established based on regression model or deep neural network such as RNN, LSTM, etc. s =[x s+1 ,x s+2 ,...,x n ] T The mean square error is calculated and the back propagation algorithm is performed to determine the final network parameters and obtain the time series prediction model.

[0057] S22, build an adversarial autoencoder model, including encoder G e , decoder G d , discriminator G y Structure, and initialize the network parameters θ e ,θ d ,θ y , and the data prior distribution. The training samples are represented by row vector x r =[x1,x2,...,x J ], we can get the hidden layer z r =G e (x;θ e ), reconstruct output discrimination result y fake , wherein, derived from sampling of the prior distribution.

[0058] S3, training the time window enhanced adversarial auto-encoder network using the training samples;

[0059] Specifically, the above S3 is to train the adversarial auto-encoder model according to the time series predictor obtained by the previous pre-training, so as to extract the features of the time series. The composite loss function is used to ensure that the model can accurately capture the change pattern in the time series while extracting the key features, and the introduction of the adversarial mechanism enhances the robustness of the model, so that it can cope with complex process dynamics. The specific implementation process is as follows:

[0060] S31, input the training data into the batch adversarial auto-encoder network, for the input data X t ∈R IK×J , the hidden layer output Z t ∈R IK×h and the reconstruction output wherein h is the feature dimension of the hidden layer. Generally speaking, the mean square error between the original input X t and the reconstruction output is the reconstruction loss term of the adversarial auto-encoder network. In this method, the reconstruction output needs to be additionally input into the time series predictor obtained in the previous step, and the mean square error between the prediction output and the actual data x t+1 is calculated, and the calculation result is taken as the prediction loss term of the adversarial auto-encoder network. The reconstruction output is compared with the original input X t , and the mean square error is calculated to obtain the reconstruction loss term of the adversarial auto-encoder network. The composite loss function for training the adversarial auto-encoder network is established based on the prediction loss term and the reconstruction loss term, so as to complete the model parameter update of the encoder G e and the decoder G d in the adversarial auto-encoder reconstruction training stage, and save the model parameters after training. Specifically, the sum of the reconstruction loss term and the prediction loss term is taken as the total loss of the reconstruction training process in this embodiment.

[0061] S32, an adversarial training strategy is executed, in each round of training process, the network parameters of the encoder G e and the decoder G d are updated through the back propagation of the reconstruction loss and the prediction loss, and then the adversarial loss is calculated by the discriminator to obtain the difference between the feature layer and the prior distribution, and the encoder G e and the discriminator G yAfter that, the two stages are performed alternately, and the model parameters after the final training are saved.

[0062] S4, using the trained time window enhanced adversarial autoencoder network to realize intermittent process fault detection.

[0063] Specifically, S4 is based on the trained time-window enhanced adversarial autoencoder model. It collects statistics on normal operating data in the feature space and reconstruction space, calculates the corresponding statistics, determines their upper limit, and sets an appropriate control threshold. In this way, when new input data exceeds this threshold, it indicates a fault has occurred, and the model can issue a timely warning. Specifically, its implementation process is as follows:

[0064] S41, set fault detection control limits: establish Hotelling statistics T 2 And the mean square error statistic SPE monitors the feature space and reconstruction space data. For example, at a certain moment, the training sample data is x r , the feature data and reconstructed data obtained after inputting the model are z r , The statistical calculation method can be expressed as:

[0065] T 2 =z r z r T

[0066]

[0067] Input training data into the model to calculate the T corresponding to each normal sample 2 After obtaining the statistics and SPE statistics, the kernel density estimation method can be used to determine the control limit α and obtain the fault detection control limit T α 2 and SPE α .

[0068] S42, online monitoring: collect online data of the intermittent process and standardize it according to S1, then input it into the model to obtain the corresponding statistics and SPE test If SPE test >SPE α or This indicates a malfunction.

[0069] In summary, in order to solve the problem of fault detection in intermittent industrial processes, this embodiment provides an intermittent process fault detection method. First, this method collects intermittent process data under normal working conditions as training samples and performs batch normalization on the observed variables. Secondly, a sliding time window is used on the training samples to obtain three-dimensional time series data samples, train a time series predictor, extract time series dependencies, and achieve the effect of predicting the next moment through historical data. Then, a time window enhanced adversarial autoencoder network is trained. During the reconstruction training process, a prediction loss term based on the time series predictor is introduced, and it is used together with the reconstruction loss as the total loss of the training process. In the adversarial training process, the model takes into account the distribution characteristics and time dependencies of the data. Finally, the statistics corresponding to the normal working condition data are calculated, the fault detection control limits are determined by the kernel density estimation method, and the model is used for online monitoring. Based on this, the intermittent process fault detection method provided by this embodiment further considers the front-end dependency of the time series on the basis of the traditional adversarial autoencoder network, has better adaptability when processing different batches of data of the intermittent process, and improves the accuracy of fault detection.

[0070] Second embodiment

[0071] This embodiment uses the application of the method of the present invention to fault detection in a penicillin fermentation process as an example to illustrate the implementation process and implementation effect of the method of the present invention for fault detection in an intermittent process.

[0072] The industrial process of penicillin fermentation with batch feeding is a typical intermittent industrial production process and is of great significance in terms of fault detection. A monitoring and control group at the Illinois Institute of Technology in the United States developed a modular fermentation batch feeding simulator Pensim to simulate this process. During the process, penicillin colonies carry out metabolic activities to produce antibiotics under suitable fermentation conditions such as temperature, pH, and concentration. Because its model is complex and changeable, and usually has strong nonlinear characteristics, the difficulty of fault detection is further increased. The present invention uses the penicillin fermentation process as the simulation research object, and uses the Pensim2.0 standard simulation platform for penicillin fermentation process to generate intermittent process data. The method steps of the present invention will be further elaborated in conjunction with the specific process below:

[0073] 1) Data collection: The present invention sets the reaction time of each batch of the process to 400h, which includes a pre-culture phase of about 45h and a feeding phase of about 355h. The sampling interval is set to 0.5h, so each batch of data contains 800 sampling points, and each sampling point contains 16 monitored process variables, as shown in Table 1. Therefore, the present invention generates a total of 20 batches of normal operating data under different initial conditions and control parameters, which can constitute a three-dimensional data matrix X (20×800×16) as a training sample. In addition, in order to verify the effectiveness of the method proposed in the present invention, 6 batches of fault data were selected for fault detection effect verification. The specific fault setting method is shown in Table 2. The dimension of each batch of fault data is the same as that of the normal operating data.

[0074] Table 1 Monitoring process variable number and name

[0075]

[0076] Table 2 Fault data number and fault type description

[0077]

[0078]

[0079] 2) Data preprocessing;

[0080] 2-1) Expand the obtained three-dimensional data X(20×800×16) along the batch direction to obtain X b (20×12800), and the matrix is ​​standardized so that the column mean is 1 and the standard deviation is 0.

[0081] 2-2) For the standardized X b (20×12800) are rearranged to obtain the two-dimensional data matrix X expanded along the variable direction t (16000×16), as the final training data input of the model.

[0082] 3) Establish a time series prediction network;

[0083] 3-1) Windowed input data: Set the input data batch size to 80, then each batch of input data dimension is time series data of X (80×16), set the window size to 5, and 75 time windows of data X can be divided. i =[X5,X6,...,X 79 ] T , and get the next moment data X corresponding to each window t =[x6,x7,...,x 80 ] T .

[0084] 3-2) Train the time series predictor and transform X i As input to a time series forecaster to maximize the forecast output and the actual output X t The correlation between is used as the training target, and the network parameters are iteratively optimized through the back propagation of the loss function and the gradient descent method. When the number of iterations reaches the set upper limit or the loss function basically no longer decreases, the iteration process is stopped and the predictor model P is saved.

[0085] 4) Establishing a time window enhanced adversarial autoencoder network model

[0086] 4-1) Create encoder G e , decoder G d , Discriminator G y Network structure and initialize network parameters θ e ,θ d ,θ y . Batch input training data X t (80×16), and setting the hidden layer feature dimension to 10, we can get the hidden layer data Z with continuous distribution t (80×10) and reconstructed output For X t =[x1,x2,...,x 80 ], And calculate the reconstruction loss and prediction loss of the data respectively:

[0087]

[0088] Use the composite loss function L1=L rec +βL pre As the total loss of encoding and decoding process, for encoder G e and decoder G d Update of model parameters, where β is a hyperparameter, can be used as a regulating factor in the loss function to balance the contribution of different loss terms to the total loss. The value needs to be adjusted according to the actual task.

[0089] 4-2) Set the prior distribution of each observed variable to the standard normal distribution N(0,1), and in each round of encoder G e and decoder G d After the model parameters are updated, the hidden layer data is input into the discriminator, the difference between the current hidden layer data distribution and the prior distribution is compared, and the adversarial loss is calculated:

[0090]

[0091] Where, Derived from the sampling of the prior distribution. The loss L of the adversarial process is obtained adv After that, the encoder G is trained through the back propagation algorithm. e and the discriminator model G y The model parameters are updated.

[0092] 4-3) Encoder G e and decoder G d The self-encoder network and encoder G e and the discriminator G y The generative adversarial network constructed is trained alternately to achieve the final optimization goal, where λ and β are adjustable hyperparameters:

[0093]

[0094] 5) Load normal operating data and calculate T 2 and SPE statistics, setting the confidence level to 95%, determine the fault detection control limit T α 2 and SPE α Read the test data set, use the same preprocessing method as the training data set, input the test data into the trained model, and calculate the statistics corresponding to the test data and SPE test ,judge and SPE test Whether it exceeds the control limit T α 2 and SPE α If any one of the items exceeds the control limit, a fault is determined to have occurred. For the fault data of 6 batches, the fault detection results of the method proposed in this invention are shown in Table 3.

[0095] Table 3 Test data fault detection results

[0096]

[0097] As can be seen from the above, the method of the present invention fully considers the complexity and dynamic characteristics of intermittent industrial process data, has better adaptability when processing different batches of data of intermittent processes, and improves the accuracy of fault detection.

[0098] Third embodiment

[0099] This embodiment provides an electronic device, such as Figure 3As shown, the electronic device includes: a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. In addition, the electronic device may also include a transceiver; the processor and the transceiver can be connected via a communication bus, and the transceiver is used to communicate with other devices.

[0100] Next, combine Figure 3 A detailed introduction to the various components of the electronic device is given below:

[0101] Among them, the processor is the control center of the electronic device, and the electronic device may include multiple processors, each of which may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here can be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPUs), or other general-purpose processors, application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement an embodiment of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor can perform various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0102] In a specific implementation, as an embodiment, the processor may include one or more CPUs, such as Figure 3 The CPU0 and CPU1 shown in FIG are, of course, only exemplary.

[0103] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0104] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and accessed through the interface circuit ( Figure 3 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0105] The transceiver may include a receiver and a transmitter ( Figure 3 The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function. The transceiver can be integrated with the processor or exist independently and communicate with the electronic device through the interface circuit ( Figure 3 (not shown) is coupled to the processor, which is not specifically limited in this embodiment of the present invention.

[0106] In addition, it should be noted that Figure 3 The structure of the electronic device shown in the figure does not constitute a limitation on the device. The actual device may include more or fewer components than shown, or may combine certain components, or arrange the components differently. In addition, the technical effects achieved by the electronic device when executing the method of the first embodiment can refer to the technical effects described in the first embodiment above, and therefore will not be repeated here.

[0107] Fourth embodiment

[0108] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device. The instructions stored therein can be loaded by a processor in a terminal to execute the method described above.

[0109] Furthermore, it should be noted that the present invention may be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention may take the form of a fully or partially hardware embodiment, a fully or partially software embodiment, or an embodiment combining software and hardware aspects. Furthermore, when implemented using software, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired connection (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium. The semiconductor medium may be a solid state drive.

[0110] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s). Figure 1 the functions specified in the individual block or blocks. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable devices to generate a computer-implemented process, thus the instructions executed on the computer or other programmable devices provide a process for implementing the functions specified in the flowchart block(s).

[0112] It should also be noted that, in the present document, the terms such as first and second, etc. are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or terminal device. Without more limitations, the element defined by the statement "including a…", does not exclude the presence of other identical elements in the process, method, article or terminal device including the element. In addition, the term "and / or" is merely a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " in the present document generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood in the context before and after. "One or more" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0113] In addition, it can be understood that in various embodiments of the present application, the size of the sequence number of the above processes does not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0114] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0115] In the several embodiments provided herein, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of functional modules / units is merely a logical functional division. In actual implementation, other division methods may be used, such as multiple units or components being combined or integrated into another device, or some features being ignored or not implemented. Furthermore, the coupling or direct coupling or communication connection shown or discussed between each other may be through some interface, indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of the present embodiment according to actual needs. In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0116] If the method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0117] Finally, it should be noted that the above description is only the preferred embodiment of the application, it should be pointed out that although the preferred embodiment of the application has been described, for those skilled in the art, once the basic creative concept of the application is known, several improvements and refinements can be made without departing from the principles of the application, and these improvements and refinements should also be considered as the protection scope of the application. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the application.

Claims

1. A method for detecting intermittent process faults based on an improved adversarial autoencoder network, characterized in that: include: Under normal operation of the intermittent process, multiple batches of process data are collected, and the collected process data are preprocessed, and the preprocessed process data are used as training samples; Adding a prediction loss term to the loss function of the adversarial autoencoder network to construct a time window enhanced adversarial autoencoder network; wherein the prediction loss term is obtained based on a pre-trained time series predictor; Using training samples to train the time window enhanced adversarial autoencoder network; Using trained time window enhanced adversarial autoencoder network to realize intermittent process fault detection; A prediction loss term is added to the loss function of the adversarial autoencoder network to construct a time window enhanced adversarial autoencoder network. Specifically: Establish a time series predictor: Use a sliding time window to divide the data, use the divided three-dimensional time window data to train the time series predictor, extract the data time dependency, and predict future process changes. Specifically, the process of establishing a time series predictor is as follows: Windowed training data input: Divide a batch of continuous time series data into multiple time segments according to time periods to obtain windowed data; each time segment data consists of observation data within the corresponding time period; use the windowed data as the input of the time series predictor to train the time series predictor: And initialize the weight parameter β i and bias r, the order is consistent with the time window length s, the column vector x k is the observation data collected at time t = k, x k-i with x k have the same dimensions, The model prediction result at time t=k is obtained; by calculating the mean square error between the output data of the time series predictor and the actual data and performing the back propagation algorithm, the final network parameters of the time series predictor are determined to obtain the trained time series predictor; Build an adversarial autoencoder model, which includes the encoder G e , decoder G d and the discriminator G y , and initialize the network parameters θ e ,θ d ,θ y , and the data prior distribution; The predicted loss item is obtained as follows: Input the reconstructed output of the adversarial autoencoder network into the pre-trained time series predictor, calculate the mean square error between the output of the time series predictor and the actual data, and use the calculated result as the prediction loss term; The loss function of the time window enhanced adversarial autoencoder network is calculated as follows: Calculate the mean square error between the reconstructed output of the adversarial autoencoder network and the original input of the adversarial autoencoder network, and use the mean square error between the reconstructed output and the original input as the reconstruction loss term; The sum of the prediction loss term and the reconstruction loss term is used as the loss function of the time window enhanced adversarial autoencoder network.

2. The intermittent process fault detection method based on the improved adversarial autoencoder network according to claim 1, characterized in that: The process data of multiple batches are collected during normal operation of the intermittent process, and the collected process data are preprocessed, and the preprocessed process data are used as training samples, including: During the normal operation of the intermittent process, multiple batches of process data are collected and represented as a three-dimensional matrix of I×J×K; where I represents the data batch, J represents the number of observed variables, and K represents the number of sampling times; The collected process data in the form of a three-dimensional matrix is ​​expanded in the batch direction to obtain a two-dimensional data matrix, and the matrix is ​​standardized in the batch direction to eliminate the influence of different dimensions. The standardized two-dimensional data matrix is ​​rearranged to obtain two-dimensional time series matrix data expanded according to the observed variables; wherein the data in each column of the two-dimensional time series matrix data reflects the change of a certain variable over time; the two-dimensional time series matrix data is used as a training sample.

3. The intermittent process fault detection method based on the improved adversarial autoencoder network according to claim 1, characterized in that: The method of using a trained time window enhanced adversarial autoencoder network to detect intermittent process faults includes: Based on the trained time window enhanced adversarial autoencoder network, the normal working condition data in the feature space and reconstruction space are counted, the corresponding statistics are calculated and their upper limits are determined, and the control threshold is set; Obtain the process data of the current intermittent process and preprocess it. Input the preprocessed current process data into the trained time window enhanced adversarial autoencoder network to obtain the statistics of the current process data. Compare the statistics of current process data with the control threshold to complete fault detection.

4. The intermittent process fault detection method based on the improved adversarial autoencoder network according to claim 3, characterized in that: The method of enhancing the adversarial autoencoder network based on the trained time window, counting the normal working condition data in the feature space and the reconstruction space, calculating the corresponding statistics and determining the upper limit thereof, and setting the control threshold, includes: Input the training sample into the time window enhanced adversarial autoencoder network and calculate the T corresponding to each normal sample 2 After calculating the statistics and SPE statistics, the fault detection control limit T is determined using the kernel density estimation method. α 2 and SPE α .

5. The intermittent process fault detection method based on the improved adversarial autoencoder network according to claim 4, characterized in that: The process of comparing the statistics of the current process data with the control threshold to complete fault detection includes: If any statistic of the current process data exceeds the corresponding fault detection control limit, it is determined that a fault has occurred.

Images