Industrial fault classification method, system and device and readable storage medium

Abstract

The invention provides an industrial fault classification method, system and device and a readable storage medium, and the method comprises the steps: carrying out the serialization of fault data, carrying out the modeling of the data through a TimeGAN-based time sequence generation algorithm, generating the time sequence data of a fault, and achieving the expansion of the fault data. And finally, performing classification modeling on the expanded data by using a time sequence classification algorithm to realize fault classification. According to the method, the accuracy and robustness of fault classification can be improved, the problem of classification performance reduction caused by data scarcity is solved, and the adaptive capacity of the model in a complex and changeable industrial environment is enhanced. Besides, through a data generation mode, a training data set can be enriched, effective learning of a deep learning model is promoted, and then the intelligent level of industrial fault monitoring and management is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more particularly to an industrial fault classification method, system, device and readable storage medium. Background Art

[0002] In modern industry, the rapid development of information technology and big data technology has made the monitoring and management of industrial processes more and more intelligent and automated. However, the complexity and diversity of industrial processes will inevitably lead to failure problems, which is still one of the key factors affecting production efficiency and quality. With the popularization of sensor technology and the improvement of data collection capabilities, enterprises can obtain a large amount of operation data in real time. These data contain rich fault information, which makes it possible to detect and classify faults. However, traditional fault classification methods often rely on manual experience and are difficult to adapt to the increasingly complex industrial environment. In addition, many existing technologies are inefficient in processing high-dimensional data, resulting in long response times and unable to meet the needs of real-time monitoring.

[0003] Specifically, traditional fault classification methods mainly rely on rule-based methods, which are often limited by expert knowledge because the specification of rules requires professional background and is difficult to generate automatically. In addition, when faced with new fault modes or environmental changes, the rules may no longer be effective and need to be constantly updated, increasing maintenance costs. With the continuous accumulation of data in industrial processes and the continuous replacement of new technologies, methods based on big data analysis and artificial intelligence technology have become more mature. By extracting potential correlations and regularities from massive industrial data, a more reliable basis is provided for fault classification. However, in the problem of fault classification, data imbalance is very serious. The amount of fault data is much smaller than the amount of normal data, which makes the model training results often suboptimal. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide an industrial fault classification method, system, device and readable storage medium. By using a time series generative adversarial network, additional fault samples can be generated to alleviate the problems caused by insufficient samples, thereby improving the generalization ability and robustness of the fault model and improving the accuracy of fault classification.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions: In a first aspect, the present invention discloses an industrial fault classification method, comprising: Obtain fault data of power plant equipment from the time series database; Performing serialization processing on the fault data to obtain time series fault data of a fixed window size, and performing normalization processing on the time series data; Use the time series generation algorithm based on the time series generative adversarial network to model the normalized time series fault data and generate synthetic fault data; Performing a denormalization process on the synthesized fault data, and fusing the denormalized synthesized fault data with the original fault data to generate fused fault data; Perform time series data truncation or padding processing on the fused fault data to generate model training data; Build a fault classification model based on the Transformer network, and use the processed model training data to train the fault classification model; Use the trained fault classification model to identify faults in real-time data.

[0006] Further, the acquiring of fault data of power plant equipment from the time series database includes: Extract the historical one-year fault data of equipment under any unit of the power plant from the time series database PI system, and the data collection time interval is 1 minute; Divide the acquired time series data into training samples and test samples , the training samples Fault data in Extract it.

[0007] Further, the serialization processing of the fault data to obtain time series fault data of a fixed window size and the normalization processing of the time series fault data include: Specify a fixed window size , for fault data Traverse and obtain the time series fault data of fixed window size, recorded as ; Identify fault data by querying The maximum value of and minimum value ; For each set of timing fault data Perform maximum and minimum value normalization calculations to generate normalized time series fault data , the formula is as follows:

[0008] in, For timing fault data state variables.

[0009] Furthermore, the method of using a time series generation algorithm based on a time series generative adversarial network to model the normalized time series fault data to generate synthetic fault data includes: Based on the time series generative adversarial network, the embedding module is used to map the original feature space to a latent space containing static features and dynamic features. The mapping process is as follows:

[0010] in, and Represent static feature data and dynamic time series features respectively, and Respectively represent the mapping functions of static feature and dynamic feature data to latent space, and Corresponding to and Low-dimensional mapping in latent space; The feature dimension of the data is restored by training the inverse mapping. The inverse mapping process is as follows:

[0011] in: and Respectively represent the static features and dynamic features after inverse mapping; Use reconstruction loss to train the embedding network and reconstruction network in the autoencoder, reconstruction loss for:

[0012] in, is the length of the time series, is the distribution of data; The generation function based on the time series generative adversarial network is defined as:

[0013] in, and represents random static features and dynamic random features, and It represents the synthetic static features and synthetic dynamic features in the latent space; Introducing a supervised error loss function and defining an unsupervised adversarial loss function to improve the generator’s ability to learn the temporal correlation of real data; Supervised Error Loss Function for:

[0014] Unsupervised adversarial loss function for:

[0015] in, and Labels representing static and dynamic features of the real sequence; , represents the static feature label of the synthetic sequence, represents the RNN feature extraction function, represents a synthetic dynamic feature label; By using a time series generative adversarial network, the normalized time series fault data is used to learn the sequence correlation in the embedding space and generate synthetic fault data.

[0016] Further, the synthesized fault data is denormalized, and the denormalized synthesized fault data is fused with the original fault data to generate fused fault data, including: Based on failure data The maximum value of and minimum value , the synthetic fault data is denormalized by the denormalization formula to generate fused fault data ; The formula for denormalization is:

[0017] in, is the synthetic fault data.

[0018] Further, the fusion fault data is subjected to time series data truncation or filling processing to generate model training data, including: Based on the fused fault data, the longest time series data length is recorded as , and generates a line of length The mask vector of , whose vector values ​​are all set to 1; For other fusion fault data, we extend the data length by filling in arbitrary values.

[0019] Based on the extended fused fault data, a corresponding mask vector is generated; the vector value of the filling position is set to 0, and the vector value of the remaining positions is set to 1, which is used as the model training data.

[0020] Furthermore, the fault classification model is constructed based on the Transformer network, including: Set the output of the Transformer network's self-attention to:

[0021] in, is the attention mask, is the input feature dimension, is the query vector, is the key vector, is a value vector.

[0022] In a second aspect, the present invention further discloses an industrial fault classification system, comprising: A data acquisition module is used to obtain fault data of power plant equipment from a time series database; A data processing module, used for performing serialization processing on the fault data to obtain time series fault data of a fixed window size, and performing normalization processing on the time series data; A data synthesis module is used to model the normalized time series fault data using a time series generation algorithm based on a time series generative adversarial network to generate synthetic fault data; A data fusion module, used for performing a denormalization process on the synthetic fault data, and fusing the denormalized synthetic fault data with the original fault data to generate fused fault data; A training data generation module is used to perform time series data truncation or padding processing on the fused fault data to generate model training data; The model building and training module is used to build a fault classification model based on the Transformer network and train the fault classification model using the processed model training data; The identification module is used to identify faults in real-time data using the trained fault classification model.

[0023] In a third aspect, the present invention further discloses an industrial fault classification device, comprising: A memory for storing an industrial fault classification program; A processor is used to implement the steps of any of the industrial fault classification methods described above when executing the industrial fault classification program.

[0024] In a fourth aspect, the present invention further discloses a readable storage medium, on which an industrial fault classification program is stored. When the industrial fault classification program is executed by a processor, the steps of the industrial fault classification method as described in any one of the above items are implemented.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention discloses an industrial fault classification method, system, device and readable storage medium, firstly serializes the fault data, then uses the time series generation algorithm based on TimeGAN to model the data, generates the time series data of the fault, and realizes the expansion of the fault data. Finally, the expanded data is classified and modeled using the time series classification algorithm to realize the classification of the fault. The present invention can help improve the accuracy and robustness of fault classification, overcome the problem of reduced classification performance caused by data scarcity, and enhance the adaptability of the model in complex and changeable industrial environments. In addition, by means of data generation, the training data set can be enriched, and the effective learning of the deep learning model can be promoted, thereby improving the intelligent level of industrial fault monitoring and management.

[0026] By using TimeGAN, the present invention can generate additional fault samples to alleviate the problem caused by insufficient samples, thereby improving the generalization ability and robustness of the model.

[0027] The fault data synthesis method proposed in the present invention can be used for model training, verification and testing, helping researchers to better understand the behavior of the model, optimize model parameters, and evaluate the performance of the model under different fault conditions, thereby improving the overall fault detection and classification capabilities.

[0028] The present invention uses TimeGAN to simulate multiple failure modes and changing trends, so that the generated data includes not only simple failure instances, but also more complex failure situations. This diversity helps to train a more comprehensive model, so that it can still effectively identify failures that have not been seen in the real world.

[0029] It can be seen that compared with the prior art, the present invention has outstanding substantive features and significant progress, and the beneficial effects of its implementation are also obvious. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0031] Figure 1 It is a method flow chart of a specific implementation mode of the present invention.

[0032] Figure 2 It is a system structure diagram of a specific implementation mode of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] See also Figure 1 As shown, this embodiment provides an industrial fault classification method, comprising the following steps: S1: Obtain fault data of power plant equipment from the time series database.

[0035] In a specific implementation, the historical fault data of a certain unit of a certain power plant for one year is extracted from the time series database PI system, and the data collection time interval is 1 minute. The acquired time series data is divided into training samples and test samples , the fault data in the training set Extract it.

[0036] S2: Serialize the fault data to obtain time series fault data with a fixed window size, and normalize the time series data.

[0037] In a specific implementation, this step includes serialization and normalization of data.

[0038] 1. Data serialization. In the actual production process of power plants, even for the same fault type, the duration of equipment failure is inconsistent. However, the input of the model needs to ensure the same time series length. Therefore, we first specify a fixed window size. , for fault data Traverse and obtain the time series fault data of fixed window size, recorded as .

[0039] 2. Data normalization. Because there are dimensional differences between different features, and in order to make the performance of the model more stable and reliable, the original data needs to be normalized. Since the amount of fault data is small, in order to make the training more stable, this step uses maximum and minimum value normalization. First, find the fault data The maximum value of and minimum value , and then for each group of split The data is normalized to the maximum and minimum values, and the formula is as follows:

[0040] in, Fault data A state variable in .

[0041] S3: Use the time series generation algorithm based on the time series generative adversarial network to model the normalized time series fault data and generate synthetic fault data.

[0042] It should be noted that the traditional Generative Adversarial Network (GAN) often only includes a generator network and a discriminator network. This simple GAN model structure may not be able to capture the step-by-step dependencies of the time series well. Therefore, TimeGAN adds a self-encoding structure network based on the traditional GAN, including an embedder, a supervision network, and a reconstruction network.

[0043] In a specific implementation, the specific implementation process of this step includes: S301: Add the autoencoder structure network.

[0044] Since traditional GAN ​​models are difficult to capture the dynamic temporal features in time series data, this step introduces an autoencoder network structure based on TimeGAN to model static and dynamic features, thereby ensuring that the generated data and real data maintain temporal dependency in the latent space.

[0045] First, TimeGAN uses an embedding module to map the original feature space to a latent space containing static features and dynamic features. The mapping process is as follows:

[0046] in, and Represent static feature data and dynamic time series features respectively, and Respectively represent the mapping functions of static feature and dynamic feature data to the latent space, and Corresponding to and The low-dimensional mapping in the hidden space. The reconstruction process completes the feature dimension recovery of the data by training the inverse mapping. The inverse mapping process is as follows:

[0047] in, and Represent the static features and dynamic features after inverse mapping respectively. Finally, the reconstruction loss is used to train the embedding network and reconstruction network in the autoencoder. for:

[0048] in, is the length of the time series, The distribution of the data.

[0049] S302: Generate adversarial network. Generate adversarial network is to achieve a balance between generator and discriminator. The generator captures the potential distribution of time series by inputting random noise and generating new synthetic sequences. The discriminator provides correct classification between real sequences and synthetic sequences. The generating function is defined as:

[0050] in, and represents random static features and dynamic random features, and It represents the synthetic static features and synthetic dynamic features in the latent space. In order to improve the generator's ability to learn the temporal correlation of real data, a supervised error loss function is introduced:

[0051] Defining the unsupervised adversarial loss function for:

[0052] in, and Labels representing static and dynamic features of the true sequence. , represents the static feature label of the synthetic sequence, represents the RNN feature extraction function, It is obtained from the hidden state of RNN and represents the synthetic dynamic feature label.

[0053] S303: Fault data generation.

[0054] Based on the joint training of each network to achieve the minimum of three errors, the model completes the learning of sequence correlation in the embedding space, thereby generating time series data that conforms to the real fault as synthetic fault data.

[0055] S4: performing denormalization processing on the synthesized fault data, and fusing the denormalized synthesized fault data with the original fault data to generate fused fault data.

[0056] In a specific embodiment, in a specific embodiment, based on the fault data The maximum value of and minimum value , the synthetic fault data is denormalized through the denormalization formula and restored to the same range as the original fault data to ensure the consistency of the data at different stages, and then generate fused fault data ; The formula for denormalization is:

[0057] in, is the synthetic fault data.

[0058] S5: Perform time series data truncation or padding processing on the fused fault data to generate model training data.

[0059] In a specific implementation, the fault data serialization in step S2 converts non-equal length time series into equal length time series data in order to generate more fault samples, and the time series classification task also requires the input time series data to be equal length. Different from the sliding window in step S2 to obtain equal length time series data, time series classification requires the participation of the full amount of time series data.

[0060] Therefore, in this step, the length of the longest time series data is taken as the standard, and the attention mask method is used to process the time series data of unequal length. Specifically, based on the fused fault data, the length of the longest time series data is taken as , and generates a line of length The mask vector of , whose vector values ​​are all 1. For the rest of the fault timing data, any value can be filled until the length is , and generate the corresponding mask vector at the same time, setting the vector value at the filling position to 0 and the vector value at the remaining positions to 1.

[0061] After the above processing, the processed fused fault data is used as model training data.

[0062] S6: Build a fault classification model based on the Transformer network and use the processed model training data to train the fault classification model.

[0063] In a specific implementation, the present invention uses a standard Transformer network as a fault classification model, and adds an attention mask on its basis to eliminate the impact of the padding data in step S5. The main component of the Transformer network is the self-attention mechanism, which consists of a query vector , key vector Sum value vector The output of self-attention is:

[0064] in, is the attention mask consisting of 0 and 1, is the input feature dimension.

[0065] Subsequently, the model training data was divided into training set, validation set and test set according to the ratio of 7:2:1, and the mean absolute error was used as the evaluation index. The network was iteratively trained on the training set, and the model with the smallest mean absolute error on the validation set was used as the final fault classification model, and its accuracy was verified on the test set.

[0066] S7: Use the trained fault classification model to identify faults in real-time data.

[0067] In a specific implementation, when a fault warning occurs in a device, the fault classification model is started and a length of The real-time fault data is obtained by using a window of . When the length of the real-time fault data is less than the window length, the real-time fault data is padded and the corresponding mask vector is generated; when the length of the real-time fault data is greater than the window length, the fault data of the window length is intercepted. Finally, the fault data is input into the fault classification model to obtain the fault category.

[0068] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process in this embodiment, another industrial fault classification method is provided, which specifically includes the following steps: Step 1: Select a unit of a power plant from 0:00:00 on January 1, 2021 to 23:59:59 on December 31, 2021, including normal operating conditions and 6 fault conditions. Select 600 training samples under normal conditions, 500 training samples of fault 1, 400 training samples of fault 3, and 300 training samples of fault 6. The number of test samples for normal conditions, fault 1, fault 3, and fault 6 is 100 each.

[0069] Step 2: Fault data serialization and data normalization.

[0070] 2-1. Data serialization. The time series length is 10. After serialization, we can get 591 training samples of normal working conditions, 491 training samples of fault 1, 391 training samples of fault 3, 291 training samples of fault 6, 91 test samples of normal working conditions, 91 test samples of fault 1, 91 test samples of fault 3, and 91 test samples of fault 6.

[0071] 2-2. Data normalization. Take the maximum and minimum values ​​of the selected fault samples 1, 3, and 6, and calculate the normalized values ​​according to the maximum and minimum normalization formulas. The maximum and minimum normalization formulas are as follows:

[0072] in, Fault data A state variable in Is the fault data The minimum value, Is the fault data The maximum value.

[0073] Step 3: Model the fault data using TimeGAN.

[0074] 3-1. Using TimeGAN to analyze fault data Data enhancement is performed to expand the original fault data and alleviate the uneven distribution of data. TimeGAN adds a self-encoding network structure to fault data based on traditional GAN. Static feature modeling and dynamic feature modeling are performed to ensure the time dependency of generated data and real data in the latent space. The static feature is recorded as , the dynamic characteristics are The autoencoder network structure includes an embedding module and a reconstruction module. First, the embedding module is used to map the original feature space to the latent space. The embedding process is defined as:

[0075] in, and Respectively represent the mapping functions of static feature and dynamic feature data to the latent space, and Corresponding to and The low-dimensional mapping in the hidden space. The reconstruction module completes the feature dimension recovery of the data by training the inverse mapping. The inverse mapping process is as follows:

[0076] in, and Represent the static features and dynamic features after inverse mapping respectively. Finally, the reconstruction loss is used to train the embedding network and reconstruction network in the autoencoder. for:

[0077] in: is the length of the time series, The distribution of the data.

[0078] Secondly, in order to improve the generator's ability to learn the temporal correlation of real data, a supervised error loss function is introduced:

[0079] in is the dynamic feature vector in the latent space. Finally, the adversarial loss of unsupervised GAN is defined as:

[0080] in, and Labels representing static and dynamic features of the true sequence. , represents the static feature label of the synthetic sequence, represents the RNN feature extraction function, The hidden state of RNN is used to represent the synthetic dynamic feature label. Through joint training of three losses, the learning of the input sequence is completed on the basis of minimizing the loss, so as to generate the induced draft fan operation data that conforms to the real time series distribution to expand the original data set.

[0081] 3-2 Generate synthetic fault data. Perform step 3.1 on the selected fault data 1, fault data 3, and fault data 6 respectively to generate synthetic data. First, specify the number of data to be generated. For fault 1, 100 fault data need to be generated; for fault 3, 200 fault data need to be generated; for fault 6, 300 fault data need to be generated, so as to ensure that the total number of each type of fault data is consistent with the number of training data under normal working conditions. During the generation process, only the generator part needs to be retained. By specifying the number of random noises, the number of synthetic data generated can be controlled. In order to verify the reliability of the generated data, the T-SNE algorithm and PCA algorithm are used to reduce the data dimension, as shown in Figure 2 shown.

[0082] Step 4: Denormalize the synthetic fault data.

[0083] The maximum and minimum values ​​of each fault are calculated in step 2.2, and the data synthesized by the algorithm are denormalized using the following formula.

[0084]

[0085] in is the synthetic data generated by the algorithm. The original data is fused with the denormalized synthetic data to obtain the expanded data set, denoted as .

[0086] Step 5: Time series data filling process.

[0087] Find the longest time series data in the total training data (data under normal working conditions, data of fault 1, data of fault 3 and data of fault 6), and record its length as , and generate a mask vector . Fill the remaining data with 1 and add it to the original vector until the data length is , and generate the corresponding mask vector at the same time, setting the vector value at the filling position to 0 and the vector value at the remaining positions to 1.

[0088] Step 6: Train the Transformer temporal classification network.

[0089] 6-1 Data preprocessing. First, the expanded data set in step 4 Divide into training sets according to 7:2:1 , validation set and test set Then the maximum and minimum normalization length is The input training sequence data After normalization, we get . For the normalized input data Perform a linear transformation to obtain , , Three matrices, and then the final output value is obtained through self-attention calculation.

[0090]

[0091] in It will reduce the effect of padding data while ensuring the consistency of the input time series data length. yes or The feature dimension is used to prevent the gradient from disappearing.

[0092] Iteratively train the Transformer model and save the model weight parameters with the smallest loss in the test set.

[0093] 6-2 Test data verification. Use the 91 normal working condition test samples, 91 fault 1 test samples, 91 fault 3 test samples, and 91 fault 6 test samples in step 2 to perform fault classification verification. The verification results are shown in Table 1: Table 1: Fault classification effect evaluation table

[0094] Among them, the accuracy , recall rate , F1 score , They are:

[0095]

[0096]

[0097]

[0098] in, represents the number of true positive examples, is the number of true negative examples, is the number of false positives, is the number of false negatives.

[0099] Step 7: Use real-time data for fault identification.

[0100] When a fault warning occurs in the device, the fault identification algorithm is started and a length of The real-time fault data is obtained by using a window of . When the length of the real-time fault data is less than the window length, the real-time fault data is padded and the corresponding mask vector is generated. When the length of the real-time fault data is greater than the window length, the fault data of the window length is intercepted. Finally, the fault data is input into the Transformer network in step 6 to obtain the fault category.

[0101] The present invention provides an industrial fault classification method that can help improve the accuracy and robustness of fault classification, overcome the problem of reduced classification performance due to data scarcity, and enhance the adaptability of the model in complex and changeable industrial environments. In addition, through data generation, the training data set can be enriched, promoting the effective learning of deep learning models, thereby improving the intelligent level of industrial fault monitoring and management.

[0102] See also Figure 2 As shown, the present invention also discloses an industrial fault classification system, including: a data acquisition module 1, a data processing module 2, a data synthesis module 3, a data fusion module 4, a training data generation module 5, a model building training module 6 and an identification module 7. The data acquisition module 1 is used to acquire the fault data of the power plant equipment from the time series database.

[0103] The data processing module 2 is used to perform serialization processing on the fault data to obtain time series fault data with a fixed window size, and to perform normalization processing on the time series data.

[0104] The data synthesis module 3 is used to model the normalized time series fault data using a time series generation algorithm based on a time series generative adversarial network to generate synthetic fault data.

[0105] The data fusion module 4 is used to perform a denormalization process on the synthetic fault data, and fuse the denormalized synthetic fault data with the original fault data to generate fused fault data.

[0106] The training data generation module 5 is used to perform time series data truncation or padding processing on the fused fault data to generate model training data.

[0107] The model building training module 6 is used to build a fault classification model based on the Transformer network and train the fault classification model using the processed model training data.

[0108] The identification module 7 is used to perform fault identification on real-time data using the trained fault classification model.

[0109] The specific implementation of the industrial fault classification system of this embodiment is basically the same as the specific implementation of the above-mentioned industrial fault classification method, and will not be repeated here.

[0110] The present invention also discloses an industrial fault classification device, comprising a processor and a memory; wherein the processor implements the steps of the industrial fault classification method as described in any one of the above items when executing the industrial fault classification program stored in the memory.

[0111] Furthermore, the industrial fault classification device in this embodiment may also include: The input interface is used to obtain the industrial fault classification program imported from the outside and save the obtained industrial fault classification program to the memory. It can also be used to obtain various instructions and parameters transmitted by external terminal devices and transmit them to the processor so that the processor can use the above various instructions and parameters to carry out corresponding processing. In this embodiment, the input interface can specifically include but is not limited to a USB interface, a serial interface, a voice input interface, a fingerprint input interface, a hard disk reading interface, etc.

[0112] The output interface is used to output various data generated by the processor to the terminal device connected thereto, so that other terminal devices connected to the output interface can obtain various data generated by the processor. In this embodiment, the output interface may specifically include but is not limited to a USB interface, a serial interface, etc.

[0113] The communication unit is used to establish a remote communication connection between the industrial fault classification device and the external server so that the industrial fault classification device can mount the image file to the external server. In this embodiment, the communication unit may specifically include but is not limited to a remote communication unit based on wireless communication technology or wired communication technology.

[0114] The keyboard is used to obtain various parameter data or instructions input by the user by tapping the keycaps in real time.

[0115] A display for real-time display of relevant information for running an industrial fault classification process.

[0116] The mouse can be used to assist users in inputting data and simplify user operations.

[0117] The present invention also discloses a readable storage medium, wherein the readable storage medium includes a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable hard disk, a CD-ROM or any other form of storage medium known in the technical field. The readable storage medium stores an industrial fault classification program, and when the industrial fault classification program is executed by a processor, the steps of the industrial fault classification method described in any one of the above items are implemented.

[0118] In summary, the present invention can generate additional fault samples by using a time series generative adversarial network, alleviate the problem caused by insufficient samples, thereby improving the generalization ability and robustness of the fault model and improving the accuracy of fault classification.

[0119] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. As for the method disclosed in the embodiment, since it corresponds to the system disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0120] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0121] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0122] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit.

[0124] Similarly, each processing unit in each embodiment of the present invention may be integrated into one functional module, or each processing unit may exist physically, or two or more processing units may be integrated into one functional module.

[0125] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0126] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0127] The industrial fault classification method, system, device and readable storage medium provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. An industrial fault classification method, characterized in that: include: Obtain fault data of power plant equipment from the time series database; Performing serialization processing on the fault data to obtain time series fault data of a fixed window size, and performing normalization processing on the time series data; Use the time series generation algorithm based on the time series generative adversarial network to model the normalized time series fault data and generate synthetic fault data; Performing a denormalization process on the synthesized fault data, and fusing the denormalized synthesized fault data with the original fault data to generate fused fault data; Perform time series data truncation or padding processing on the fused fault data to generate model training data; Build a fault classification model based on the Transformer network, and use the processed model training data to train the fault classification model; Use the trained fault classification model to identify faults in real-time data.

2. The industrial fault classification method according to claim 1, characterized in that: The acquiring of fault data of power plant equipment from the time series database includes: Extract the historical one-year fault data of equipment under any unit of the power plant from the time series database PI system, and the data collection time interval is 1 minute; Divide the acquired time series data into training samples and test samples , the training samples Fault data in Extract it.

3. The industrial fault classification method according to claim 2, characterized in that: The serializing the fault data to obtain time series fault data of a fixed window size, and normalizing the time series fault data, includes: Specify a fixed window size , for fault data Traverse and obtain the time series fault data of fixed window size, recorded as ; Identify fault data by querying The maximum value of and minimum value ; For each set of timing fault data Perform maximum and minimum value normalization calculations to generate normalized time series fault data , the formula is as follows: in, For timing fault data state variables.

4. The industrial fault classification method according to claim 3, characterized in that: The method uses a time series generation algorithm based on a time series generative adversarial network to model the normalized time series fault data to generate synthetic fault data, including: Based on the time series generative adversarial network, the embedding module is used to map the original feature space to a latent space containing static features and dynamic features. The mapping process is as follows: in, and Represent static feature data and dynamic time series features respectively, and Respectively represent the mapping functions of static feature and dynamic feature data to the latent space, and Corresponding to and Low-dimensional mapping in latent space; The feature dimension of the data is restored by training the inverse mapping. The inverse mapping process is as follows: in: and Respectively represent the static features and dynamic features after inverse mapping; Use reconstruction loss to train the embedding network and reconstruction network in the autoencoder, reconstruction loss for: in, is the length of the time series, is the distribution of data; The generation function based on the time series generative adversarial network is defined as: in, and represents random static features and dynamic random features, and It represents the synthetic static features and synthetic dynamic features in the latent space; Introducing a supervised error loss function and defining an unsupervised adversarial loss function to improve the generator’s ability to learn the temporal correlation of real data; Supervised Error Loss Function for: Unsupervised adversarial loss function for: in, and Labels representing static and dynamic features of the real sequence; , represents the static feature label of the synthetic sequence, represents the RNN feature extraction function, represents a synthetic dynamic feature label; By using a time series generative adversarial network, the normalized time series fault data is used to learn the sequence correlation in the embedding space and generate synthetic fault data.

5. The industrial fault classification method according to claim 4, characterized in that: The step of performing denormalization processing on the synthesized fault data and fusing the denormalized synthesized fault data with the original fault data to generate fused fault data includes: Based on failure data The maximum value of and minimum value , the synthetic fault data is denormalized by the denormalization formula to generate fused fault data ; The formula for denormalization is: in, is the synthetic fault data.

6. The industrial fault classification method according to claim 5, characterized in that: The performing time series data truncation or padding processing on the fused fault data to generate model training data includes: Based on the fused fault data, the longest time series data length is recorded as , and generates a line of length The mask vector of , whose vector values ​​are all set to 1; For other fusion fault data, we extend the data length by filling in arbitrary values. Based on the extended fused fault data, a corresponding mask vector is generated; the vector value of the filling position is set to 0, and the vector value of the remaining positions is set to 1, which is used as the model training data.

7. The industrial fault classification method according to claim 1, characterized in that: The fault classification model is constructed based on the Transformer network, including: Set the output of the Transformer network's self-attention to: in, is the attention mask, is the input feature dimension, is the query vector, is the key vector, is a value vector.

8. An industrial fault classification system, characterized in that: The system adopts the industrial fault classification method as claimed in any one of claims 1 to 7; The system comprises: A data acquisition module is used to obtain fault data of power plant equipment from a time series database; A data processing module, used for performing serialization processing on the fault data to obtain time series fault data of a fixed window size, and performing normalization processing on the time series data; A data synthesis module is used to model the normalized time series fault data using a time series generation algorithm based on a time series generative adversarial network to generate synthetic fault data; A data fusion module, used for performing a denormalization process on the synthetic fault data, and fusing the denormalized synthetic fault data with the original fault data to generate fused fault data; A training data generation module is used to perform time series data truncation or padding processing on the fused fault data to generate model training data; The model building and training module is used to build a fault classification model based on the Transformer network and train the fault classification model using the processed model training data; The identification module is used to identify faults in real-time data using the trained fault classification model.

9. An industrial fault classification device, characterized in that: include: A memory for storing an industrial fault classification program; A processor is used to implement the steps of the industrial fault classification method according to any one of claims 1 to 7 when executing the industrial fault classification program.

10. A readable storage medium, characterized in that: The readable storage medium stores an industrial fault classification program, and when the industrial fault classification program is executed by a processor, the steps of the industrial fault classification method according to any one of claims 1 to 7 are implemented.

Images