A method and device for predicting partial discharge patterns of a switchgear

Abstract

The application provides a kind of switch cabinet partial discharge mode prediction method and device.The prediction method includes: obtaining a plurality of partial discharge signals of switch cabinet;And based on partial discharge signal screening model, a plurality of partial discharge signals are screened, and obtainable predictable partial discharge signal and unpredictable partial discharge signal;Based on original partial discharge mode identification model, the predictable partial discharge signal is predicted, and the partial discharge mode of the predictable partial discharge signal is obtained, and the partial discharge mode corresponding to the unpredictable partial discharge signal is marked;Based on a plurality of partial discharge signals and the partial discharge mode corresponding to each partial discharge signal, generate sample expansion after identification training sample;Based on sample expansion after identification training sample, the original partial discharge mode identification model is retrained, and the retraining partial discharge mode identification model is obtained, and the real-time partial discharge signal of switch cabinet is predicted based on the retraining partial discharge mode identification model.The application can improve the identification accuracy of partial discharge mode identification model, and ensure the safe operation of switch cabinet.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transmission and distribution, and particularly relates to a switch cabinet partial discharge mode prediction method and device. BACKGROUND

[0002] The switch cabinet occupies a crucial position in the power system, and its good operating state is an important basis for ensuring the safe and reliable operation of the power system. With the rapid development of China's ultra-high voltage power grid, partial discharge has become a key factor causing the degradation of the insulation performance of power equipment. Different types of partial discharge can cause different degrees of harm to electrical equipment. Therefore, accurate identification of the partial discharge mode can provide reliable reference information for power equipment health status assessment and serve as an auxiliary means for partial discharge fault location, which helps to improve the accuracy of partial discharge fault location.

[0003] The artificial neural network realizes the recognition of the partial discharge mode of the partial discharge signal by mining, analyzing and learning the features in the partial discharge signal. However, the electrode shape and electric field environment of the partial discharge have a great influence on the waveform of the partial discharge signal. The production, transportation and on-site operation environment of the switch cabinet and other power equipment are complex, and the on-site equipment insulation defect conditions are diverse, resulting in various electrode shapes and complex electric field environments. The complex partial discharge environment causes the signal waveform to be diverse, and it is inevitable that the similarity of the partial discharge signal to the pulse waveform in the training sample library of the artificial neural network is low.

[0004] Since the features of this part of the partial discharge signal have not been mined and learned, and the partial discharge mode recognition model is difficult to learn the features of this part of the partial discharge signal autonomously, the partial discharge mode recognition model has difficulty in recognizing this type of partial discharge signal, which can easily lead to safety accidents of the switch cabinet and other power equipment. SUMMARY

[0005] The present application provides a switch cabinet partial discharge mode prediction method and device, which can improve the recognition accuracy of the partial discharge mode recognition model and ensure the safe operation of the switch cabinet and other power equipment.

[0006] In a first aspect, the present application provides a switch cabinet partial discharge mode prediction method, comprising: obtaining a plurality of partial discharge signals of a switch cabinet; and screening the plurality of partial discharge signals based on a partial discharge signal screening model to obtain predictable partial discharge signals and unpredictable partial discharge signals; predicting the predictable partial discharge signals based on an original partial discharge mode recognition model to obtain the partial discharge mode of the predictable partial discharge signals, and labeling the partial discharge mode corresponding to the unpredictable partial discharge signals; generating sample-expanded recognition training samples based on the plurality of partial discharge signals and the partial discharge modes corresponding to each partial discharge signal; retraining the original partial discharge mode recognition model based on the sample-expanded recognition training samples to obtain a retrained partial discharge mode recognition model, and predicting real-time partial discharge signals of the switch cabinet based on the retrained partial discharge mode recognition model.

[0007] The present application provides a switch cabinet partial discharge mode prediction method, which screens the partial discharge signals of the switch cabinet by setting a partial discharge signal screening model, obtains predictable partial discharge signals and unpredictable partial discharge signals, and labels the corresponding partial discharge mode of the unpredictable partial discharge signals. Then, sample-expanded recognition training samples are generated based on the plurality of partial discharge signals and the partial discharge modes corresponding to each partial discharge signal; and the original partial discharge mode recognition model is retrained with the expanded recognition training samples, so that the retrained partial discharge mode recognition model can recognize more partial discharge signals, improving the recognition accuracy of the partial discharge mode recognition model. In addition, the present application can actively mine and learn the unpredictable partial discharge signals through the partial discharge signal screening model, ensuring that the partial discharge mode recognition model can accurately recognize in real time, thereby ensuring the safe operation of the switch cabinet and other power equipment.

[0008] In a possible implementation, based on the partial discharge signal screening model, the plurality of partial discharge signals are screened to obtain predictable partial discharge signals and unpredictable partial discharge signals, and the method further comprises: obtaining historical partial discharge signals of the switch cabinet in a historical period, the historical partial discharge signals including first partial discharge signals and second partial discharge signals, the first partial discharge signals being partial discharge signals that can be recognized by the original partial discharge mode recognition model, and the second partial discharge signals being partial discharge signals that cannot be recognized by the original partial discharge mode recognition model; performing feature analysis on the historical partial discharge signals based on the original partial discharge mode recognition model to obtain feature maps of each historical partial discharge signal, wherein each feature map of each historical partial discharge signal includes feature maps output by each convolution layer of the original partial discharge mode recognition model and feature maps output by a neural network layer; generating screening training samples based on the feature maps of each historical partial discharge signal and whether the historical partial discharge signals are recognizable; and training a pre-set convolutional neural network model based on the screening training samples to obtain the partial discharge signal screening model.

[0009] In a possible implementation, the pre-set convolutional neural network model comprises an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer; the pre-set convolutional neural network model is trained based on the screening training samples to obtain a partial discharge signal screening model, including: step one, inputting the feature maps in each screening training sample into the input layer to obtain a fusion feature map; the fusion feature map is used to represent the features of a single feature map or the fusion features of multiple feature maps; step two, calculating the fusion feature map through the first convolutional neural network to output a feature matrix of the fusion feature map; step three, generating an attention matrix based on the attention module; the attention matrix is used to increase the weight of important features in the feature matrix of the fusion feature map; step four, converting the attention matrix and the feature matrix of the fusion feature map to obtain a feature mapping matrix based on the attention matrix; step five, inputting the feature mapping matrix into the second convolutional neural network and outputting a screening result through the output layer, wherein the screening result of the feature map includes that the partial discharge signal corresponding to the feature map is identifiable or the partial discharge signal corresponding to the feature map is not identifiable; step six, calculating a loss function of the current iteration training process based on the screening result and the screening training sample corresponding to the screening result; step seven, if the loss function meets a preset condition or the number of iterations reaches a maximum number of iterations, the training process is exited to obtain the partial discharge signal screening model; otherwise, the parameters in the attention module, the first convolutional neural network, and the second convolutional neural network are updated, and steps two to seven are repeatedly executed until the training process is exited.

[0010] In a possible implementation, inputting the feature maps in each screening training sample into the input layer to obtain a fusion feature map includes: for any feature map, if the feature map is a feature map output by a neural network layer, the feature map is directly determined as a fusion feature map; if the feature map is a feature map output by a last convolutional layer, the feature map is fused with a feature map output by the neural network layer to generate a fusion feature map; if the feature map is a feature map output by a convolutional layer other than the last convolutional layer, the feature map is fused with a feature map output by a next convolutional layer to generate a fusion feature map.

[0011] In a possible implementation, the partial discharge signal screening model is used to screen multiple partial discharge signals to obtain predictable partial discharge signals and unpredictable partial discharge signals, including: inputting the multiple partial discharge signals into the partial discharge signal screening model to obtain a prediction value corresponding to each partial discharge signal; determining a partial discharge signal with a prediction value greater than a screening threshold as a predictable partial discharge signal; and determining a partial discharge signal with a prediction value less than or equal to the screening threshold as an unpredictable partial discharge signal.

[0012] In a possible implementation, the step of labeling the partial discharge mode corresponding to the unpredictable partial discharge signal comprises: obtaining the partial discharge mode corresponding to each unpredictable partial discharge signal input by a user; and labeling the unpredictable partial discharge signal based on the partial discharge mode corresponding to each unpredictable partial discharge signal input by the user.

[0013] In a possible implementation, the step of generating the sample-expanded recognition training sample based on the plurality of partial discharge signals and the partial discharge mode corresponding to each partial discharge signal comprises: determining a first recognition training sample based on the predictable partial discharge signal and the partial discharge mode of the predictable partial discharge signal; determining a second recognition training sample based on the unpredictable partial discharge signal and the labeled partial discharge mode of the unpredictable partial discharge signal; and determining the sample-expanded recognition training sample based on the first recognition training sample and the second recognition training sample.

[0014] In a possible implementation, the communication module is further configured to obtain historical partial discharge signals of the switch cabinet in a historical period, the historical partial discharge signals comprising a first partial discharge signal and a second partial discharge signal, the first partial discharge signal being a partial discharge signal that can be recognized by the original partial discharge mode recognition model, and the second partial discharge signal being a partial discharge signal that cannot be recognized by the original partial discharge mode recognition model; and the processing module is further configured to perform feature analysis on the historical partial discharge signals based on the original partial discharge mode recognition model to obtain a feature map of each historical partial discharge signal, wherein the feature map of each historical partial discharge signal comprises a feature map output by each convolution layer of the original partial discharge mode recognition model and a feature map output by a neural network layer; generate a screening training sample based on the feature map of each historical partial discharge signal and whether the historical partial discharge signal is recognizable; and train a convolutional neural network model preset to obtain the partial discharge signal screening model.

[0015] In a possible implementation, the communication module is further configured to obtain historical partial discharge signals of the switch cabinet in a historical period, the historical partial discharge signals comprising a first partial discharge signal and a second partial discharge signal, the first partial discharge signal being a partial discharge signal that can be recognized by the original partial discharge mode recognition model, and the second partial discharge signal being a partial discharge signal that cannot be recognized by the original partial discharge mode recognition model; and the processing module is further configured to perform feature analysis on the historical partial discharge signals based on the original partial discharge mode recognition model to obtain a feature map of each historical partial discharge signal, wherein the feature map of each historical partial discharge signal comprises a feature map output by each convolution layer of the original partial discharge mode recognition model and a feature map output by a neural network layer; generate a screening training sample based on the feature map of each historical partial discharge signal and whether the historical partial discharge signal is recognizable; and train a convolutional neural network model preset to obtain the partial discharge signal screening model.

[0016] In a possible implementation, the pre-set convolutional neural network model comprises an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer; and the processing module is specifically configured to perform the following steps: step one, inputting the feature maps in each screening training sample into the input layer to obtain a fusion feature map; the fusion feature map is used to represent the features of a single feature map or the fusion features of multiple feature maps; step two, calculating a feature matrix of the fusion feature map through the first convolutional neural network; step three, generating an attention matrix based on the attention module; the attention matrix is used to increase the weight of important features in the feature matrix of the fusion feature map; step four, converting the attention matrix and the feature matrix of the fusion feature map to obtain a feature mapping matrix; step five, inputting the feature mapping matrix into the second convolutional neural network, and outputting a screening result through the output layer, wherein the screening result of the feature map includes that the partial discharge signal corresponding to the feature map is identifiable or that the partial discharge signal corresponding to the feature map is not identifiable; step six, calculating a loss function of the current iteration training process based on the screening result and the screening training sample corresponding to the screening result; step seven, if the loss function meets a preset condition or the number of iterations reaches a maximum number of iterations, the training process is exited to obtain a partial discharge signal screening model; otherwise, the parameters in the attention module, the first convolutional neural network, and the second convolutional neural network are updated, and steps two to seven are repeatedly performed until the training process is exited.

[0017] In a possible implementation, the processing module is specifically configured to, for any feature map, if the feature map is a feature map output by a neural network layer, directly determining the feature map as a fusion feature map; if the feature map is a feature map output by a last convolutional layer, fusing the feature map with a feature map output by the neural network layer to generate a fusion feature map; and if the feature map is a feature map output by a convolutional layer other than the last convolutional layer, fusing the feature map with a feature map output by a next convolutional layer to generate a fusion feature map.

[0018] In a possible implementation, the processing module is specifically configured to input multiple partial discharge signals into the partial discharge signal screening model to obtain predicted values corresponding to the partial discharge signals; determine a partial discharge signal with a predicted value greater than a screening threshold as a predictable partial discharge signal; and determine a partial discharge signal with a predicted value less than or equal to the screening threshold as an unpredictable partial discharge signal.

[0019] In a possible implementation, the processing module is specifically configured to obtain partial discharge modes corresponding to each unpredictable partial discharge signal input by a user; and label the unpredictable partial discharge signals based on the partial discharge modes corresponding to the unpredictable partial discharge signals input by the user.

[0020] In a possible implementation, the processing module is specifically configured to determine the first identification training sample based on the predictable partial discharge signal and the partial discharge mode of the predictable partial discharge signal; determine the second identification training sample based on the unpredictable partial discharge signal and the partial discharge mode labeled for the unpredictable partial discharge signal; and determine the sample-expanded identification training sample based on the first identification training sample and the second identification training sample.

[0021] In a third aspect, an electronic device is provided, and the electronic device includes a memory and a processor. The memory stores a computer program. The processor is configured to invoke and run the computer program stored in the memory to perform the steps of the method in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program. The computer program, when executed by a processor, implements the steps of the method in the first aspect and any possible implementation of the first aspect.

[0023] The technical effects brought by any possible implementation of the second aspect to the fourth aspect can refer to the technical effects brought by the corresponding implementation of the first aspect, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0025] Figure 1 is a flowchart of a partial discharge mode prediction method of a switch cabinet provided by an embodiment of the present application;

[0026] Figure 2 is a flowchart of another partial discharge mode prediction method of a switch cabinet provided by an embodiment of the present application;

[0027] Figure 3 is a structural diagram of a partial discharge mode prediction device provided by an embodiment of the present application;

[0028] Figure 4 is a structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0029] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0030] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this document is only a description of the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can mean that A exists alone, A and B exist together, and B exists alone. In addition, "at least one" and "multiple" mean two or more. "First", "second", and the like do not limit the quantity and execution order, and "first", "second", and the like do not necessarily mean different.

[0031] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Rather, the use of the words "exemplary" or "for example" is intended to present concepts in a concrete manner to facilitate understanding.

[0032] In addition, the terms "include" and "have" and any variations thereof in the description of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the listed steps or modules, but can optionally include other steps or modules not listed or other steps or modules inherent to the process, method, product, or device.

[0033] In order to make the purpose, technical scheme and advantages of the present application clearer, specific embodiments will be described below with reference to the accompanying drawings of the present application.

[0034] With the rapid development of advanced sensing technology and computer technology, artificial intelligence technology has been widely applied in the field of fault diagnosis of power transformation equipment. Machine learning algorithm is fully combined with partial discharge fault diagnosis. The fault diagnosis system based on artificial neural network (ANN), support vector machine (SVM) and random forest (RF) has achieved great results in transformer operation and maintenance management. Traditional machine learning algorithm is essentially a shallow learning, which is difficult to mine high-dimensional features of partial discharge signals, resulting in low accuracy of pattern recognition of partial discharge signals. Deep learning has gradually become the fastest developing and most promising method in the field of partial discharge fault diagnosis due to its superior data feature mining ability.

[0035] Insulation defects are the main cause of partial discharge. The electrode shape and electric field environment of partial discharge have a great influence on the waveform of partial discharge pulse. The production, transportation and on-site operation environment of power equipment are complex, and the insulation defect conditions of on-site equipment are diverse, resulting in various electrode shapes and complex electric field environment of partial discharge. The complex partial discharge environment causes various pulse waveforms on site. It is inevitable that there will be partial discharge pulses with low similarity to the pulse waveforms in the training sample library. Due to the lack of feature knowledge, traditional partial discharge pattern recognition methods relying solely on data-driven are difficult to learn the features of such pulses, which further causes the pattern recognition model to have low generalization for such pulses. It is difficult to identify such pulses in on-site production, and the fault cause is often determined after the accident is inspected.

[0036] To solve the above technical problems, the present application provides a switch cabinet partial discharge pattern prediction method and device. The method builds a multilevel convolutional neural network integrating attention mechanism (AM-MLCNN), uses the network to filter the unknown feature partial discharge signals collected on site, and after labeling the partial discharge pattern of such partial discharge signals, expands the original training set of the partial discharge pattern recognition model, so that the partial discharge pattern recognition model retrained based on the expanded training set can recognize more partial discharge signals, and improves the recognition accuracy of the partial discharge pattern recognition model.

[0037] As shown in Figure 1 The present application provides a switch cabinet partial discharge pattern prediction method, and the execution subject is a switch cabinet partial discharge pattern prediction device. The prediction method includes steps S101-S105.

[0038] S101, acquire a plurality of partial discharge signals of a switch cabinet.

[0039] In some embodiments, the plurality of partial discharge signals of the switch cabinet can include predictable partial discharge signals and unpredictable partial discharge signals.

[0040] As a possible implementation manner, the prediction device can acquire the time-frequency spectrogram of the plurality of partial discharge signals to acquire the plurality of partial discharge signals.

[0041] In some embodiments, the prediction device can perform adaptive weighted frame-by-frame sparse method on the partial discharge signals to remove noise in the partial discharge signals.

[0042] S102, screening the plurality of partial discharge signals based on the partial discharge signal screening model to obtain the predictable partial discharge signals and the unpredictable partial discharge signals.

[0043] As a possible implementation manner, the prediction device can determine the predictable partial discharge signals and the unpredictable partial discharge signals based on steps S1021-S1023.

[0044] S1021, inputting the plurality of partial discharge signals into the partial discharge signal screening model to obtain the prediction value corresponding to each partial discharge signal.

[0045] S1022, determining the partial discharge signal with the prediction value greater than the screening threshold as the predictable partial discharge signal.

[0046] S1023, determining the partial discharge signal with the prediction value less than or equal to the screening threshold as the unpredictable partial discharge signal.

[0047] S103, predicting the predictable partial discharge signal based on the original partial discharge pattern recognition model to obtain the partial discharge pattern of the predictable partial discharge signal.

[0048] In some embodiments, the partial discharge pattern can include tip discharge, floating discharge, bubble discharge, and surface discharge.

[0049] As a possible implementation manner, the prediction device can input the predictable partial discharge signal into the original partial discharge pattern recognition model to obtain the partial discharge pattern of the predictable partial discharge signal.

[0050] S104, labeling the partial discharge pattern corresponding to the unpredictable partial discharge signal.

[0051] As a possible implementation manner, the prediction device can acquire the partial discharge pattern corresponding to each unpredictable partial discharge signal input by the user; and label the unpredictable partial discharge signal based on the partial discharge pattern corresponding to each unpredictable partial discharge signal input by the user.

[0052] S105, generating the sample expansion recognition training sample based on the plurality of partial discharge signals and the partial discharge pattern corresponding to each partial discharge signal.

[0053] As a possible implementation manner, the prediction device can generate the sample-expanded identification training sample based on steps S1051-S1053.

[0054] S1051, determine the first identification training sample based on the predictable partial discharge signal and the partial discharge mode of the predictable partial discharge signal.

[0055] The first identification training sample takes the predictable signal as input and takes the partial discharge mode of the predictable partial discharge signal as output.

[0056] S1052, determine the second identification training sample based on the unpredictable partial discharge signal and the partial discharge mode labeled for the unpredictable partial discharge signal.

[0057] The second identification training sample takes the unpredictable signal as input and takes the partial discharge mode labeled for the unpredictable partial discharge signal as output.

[0058] S1053, determine the sample-expanded identification training sample based on the first identification training sample and the second identification training sample.

[0059] S106, retrain the original partial discharge mode identification model based on the sample-expanded identification training sample to obtain a retrained partial discharge mode identification model.

[0060] As a possible implementation manner, the prediction device can update the classifier parameters of the original partial discharge mode identification model during the retraining process, so that the classification of the classifier is more accurate.

[0061] S107, predict the real-time partial discharge signal of the switch cabinet based on the retrained partial discharge mode identification model.

[0062] The present application provides a kind of prediction method of switch cabinet partial discharge mode, by setting partial discharge signal screening model, the partial discharge signal of switch is screened, and the predictable partial discharge signal and unpredictable partial discharge signal are obtained, and the corresponding partial discharge mode is labeled for unpredictable partial discharge signal.Afterwards, based on multiple partial discharge signals and the partial discharge mode corresponding to each partial discharge signal, generate sample-expanded identification training sample;And with the expanded identification training sample, the original partial discharge mode identification model is retrained, so that the retrained partial discharge mode identification model can identify more partial discharge signals, improve the identification accuracy of partial discharge mode identification model.In addition, the present application can realize active mining and learning for unpredictable partial discharge signal by partial discharge signal screening model, guarantee that partial discharge mode identification model can be identified in real time accurately, so as to guarantee the safe operation of switch cabinet and other power equipment.

[0063] Optionally, as Figure 2As shown, the switch cabinet partial discharge mode prediction method provided by the embodiment of the application further comprises steps S201-S204 before step S102.

[0064] S201, acquire historical partial discharge signals of the switch cabinet in a historical period.

[0065] In some embodiments, the historical partial discharge signals include first partial discharge signals and second partial discharge signals.

[0066] For example, the first partial discharge signals are partial discharge signals that can be recognized by the original partial discharge mode recognition model.

[0067] For example, the second partial discharge signals are partial discharge signals that cannot be recognized by the original partial discharge mode recognition model.

[0068] S202, perform feature analysis on the historical partial discharge signals based on the original partial discharge mode recognition model to obtain feature maps of the historical partial discharge signals.

[0069] Each feature map of the historical partial discharge signals includes feature maps output by each convolution layer of the original partial discharge mode recognition model and feature maps output by a neural network layer.

[0070] S203, generate screening training samples based on the feature maps of the historical partial discharge signals and whether the historical partial discharge signals can be recognized.

[0071] Each sample in the screening training samples takes a feature map as input and takes whether the historical partial discharge signal corresponding to the feature map can be recognized as output.

[0072] In some embodiments, the screening training samples include first screening training samples and second screening training samples.

[0073] For example, the first screening training samples take feature maps of the first partial discharge signals as input and take that the first partial discharge signals can be recognized as output.

[0074] For example, the second screening training samples take feature maps of the second partial discharge signals as input and take that the second partial discharge signals cannot be recognized as output.

[0075] S204, train a pre-set convolutional neural network model to obtain a partial discharge signal screening model.

[0076] In some embodiments, the pre-set convolutional neural network model includes an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer.

[0077] As a possible implementation, the prediction device can train the partial discharge signal screening model based on steps one to seven.

[0078] Step one, inputting the feature maps in each screening training sample into an input layer to obtain a fused feature map.

[0079] In some embodiments, the fused feature map is used to represent the features of a single feature map, or the fused features of multiple feature maps.

[0080] As a possible implementation, for any feature map, the prediction device can determine the fused feature map based on the convolutional pooling layer and the neural network layer corresponding to the feature map.

[0081] For example, if the feature map is the feature map output by the neural network layer, the feature map is directly determined as the fused feature map.

[0082] For another example, if the feature map is the feature map output by the last convolutional layer, the feature map is fused with the feature map output by the neural network layer to generate the fused feature map.

[0083] For another example, if the feature map is the feature map output by a convolutional layer other than the last convolutional layer, the feature map is fused with the feature map output by the next convolutional layer to generate the fused feature map.

[0084] Step two, inputting the fused feature map into a first convolutional neural network to calculate a feature matrix of the fused feature map.

[0085] Step three, generating an attention matrix based on an attention module; the attention matrix is used to increase the weight of important features in the feature matrix of the fused feature map.

[0086] Step four, converting the attention matrix and the feature matrix of the fused feature map to obtain a feature mapping matrix.

[0087] As a possible implementation, the prediction device can determine the feature mapping matrix based on the following formula.

[0088] C i2 = C i1 · H i ;

[0089] Wherein, Ci1 is the feature matrix of the fused feature map corresponding to the i-th layer in the original partial discharge pattern recognition model, Ci2 is the feature mapping matrix corresponding to the i-th layer in the original partial discharge pattern recognition model, and Hi is the attention matrix corresponding to the i-th layer.

[0090] Step five, inputting the feature mapping matrix into a second convolutional neural network, and outputting a screening result through an output layer.

[0091] Wherein, the screening result of the feature map includes that the partial discharge signal corresponding to the feature map is identifiable or that the partial discharge signal corresponding to the feature map is not identifiable.

[0092] Step six, based on the screening result and the screening training sample corresponding to the screening result, the loss function of the current iteration training process is calculated.

[0093] In some embodiments, the loss function can be expressed as the following formula.

[0094] L AM-MLCNN =-E[A S | case1 ]+E[A S | case2 ];

[0095] Wherein, L AM-MLCNN is the loss function, E[A S | case1 ] is the accuracy of the screening result of the first screening sample, E[A S | case2 ] is the accuracy of the screening result of the second screening sample.

[0096] Step seven, if the loss function reaches the preset condition, or the iteration number reaches the maximum iteration number, the training process is exited, and the partial discharge signal screening model is obtained; Otherwise, update the parameters in the attention module, the first convolutional neural network and the second convolutional neural network, and repeat steps two to seven until the training process is exited.

[0097] In this way, the partial discharge signal screening model can be obtained by training, the partial discharge signal can be screened, and the recognition training sample can be expanded, so as to improve the recognition accuracy of the partial discharge mode recognition model.

[0098] It should be noted that the embodiment of the present application introduces the attention mechanism to realize the automatic recognition of the unrecognized features in the partial discharge signal, so that more partial discharge signals that may have partial discharge problems can be introduced, the real-time expansion of the recognition training sample is realized, the real-time and accurate recognition of the partial discharge mode recognition model is ensured, and the reliable operation of the switch cabinet is ensured.

[0099] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0100] The following is a device embodiment of the present application, and the details not described in detail can be referred to the corresponding method embodiments described above.

[0101] Figure 3 A structure schematic diagram of a switch cabinet partial discharge mode prediction device provided by an embodiment of the present application is shown. The prediction device 300 includes a communication module 301 and a processing module 302.

[0102] The communication module 301 is configured to acquire a plurality of partial discharge signals of the switch cabinet. The processing module 302 is configured to: perform screening on the plurality of partial discharge signals based on a partial discharge signal screening model to obtain predictable partial discharge signals and unpredictable partial discharge signals; perform prediction on the predictable partial discharge signals based on an original partial discharge pattern recognition model to obtain a partial discharge pattern of the predictable partial discharge signals, and label a partial discharge pattern corresponding to the unpredictable partial discharge signals; generate a sample-expanded recognition training sample based on the plurality of partial discharge signals and the partial discharge pattern corresponding to each partial discharge signal; and retrain the original partial discharge pattern recognition model based on the sample-expanded recognition training sample to obtain a retrained partial discharge pattern recognition model, and perform prediction on real-time partial discharge signals of the switch cabinet based on the retrained partial discharge pattern recognition model.

[0103] In a possible implementation, the communication module 301 is further configured to acquire historical partial discharge signals of the switch cabinet in a historical period, the historical partial discharge signals including first partial discharge signals and second partial discharge signals, the first partial discharge signals being partial discharge signals that can be recognized by the original partial discharge pattern recognition model, and the second partial discharge signals being partial discharge signals that cannot be recognized by the original partial discharge pattern recognition model; and the processing module 302 is further configured to: perform feature analysis on the historical partial discharge signals based on the original partial discharge pattern recognition model to obtain a feature map of each historical partial discharge signal, wherein the feature map of each historical partial discharge signal includes feature maps output by each convolution layer of the original partial discharge pattern recognition model and feature maps output by a neural network layer; generate screening training samples based on the feature map of each historical partial discharge signal and whether the historical partial discharge signal is recognizable; and train a convolutional neural network model that is pre-set to obtain the partial discharge signal screening model based on the screening training samples.

[0104] In a possible implementation, the pre-set convolutional neural network model comprises an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer; the processing module 302 is specifically configured to perform the following steps: step one, inputting the feature maps in each screening training sample into the input layer to obtain a fusion feature map; the fusion feature map is used to represent the features of a single feature map or the fusion features of multiple feature maps; step two, calculating the feature matrix of the fusion feature map through the first convolutional neural network; step three, generating an attention matrix based on the attention module; the attention matrix is used to increase the weight of important features in the feature matrix of the fusion feature map; step four, converting the attention matrix and the feature matrix of the fusion feature map to obtain a feature mapping matrix; step five, inputting the feature mapping matrix into the second convolutional neural network, and outputting a screening result through the output layer, wherein the screening result of the feature map includes that the partial discharge signal corresponding to the feature map is identifiable or the partial discharge signal corresponding to the feature map is not identifiable; step six, calculating the loss function of the current iteration training process based on the screening result and the screening training sample corresponding to the screening result; step seven, if the loss function meets a preset condition or the number of iterations reaches a maximum number of iterations, the training process is exited to obtain a partial discharge signal screening model; otherwise, the parameters in the attention module, the first convolutional neural network, and the second convolutional neural network are updated, and steps two to seven are repeatedly performed until the training process is exited.

[0105] In a possible implementation, the processing module 302 is specifically configured to, for any feature map, if the feature map is a feature map output by a neural network layer, directly determining the feature map as a fusion feature map; if the feature map is a feature map output by a last convolutional layer, fusing the feature map with a feature map output by the neural network layer to generate a fusion feature map; if the feature map is a feature map output by a convolutional layer other than the last convolutional layer, fusing the feature map with a feature map output by a next convolutional layer to generate a fusion feature map.

[0106] In a possible implementation, the processing module 302 is specifically configured to input multiple partial discharge signals into the partial discharge signal screening model to obtain predicted values corresponding to the partial discharge signals; determine a partial discharge signal with a predicted value greater than a screening threshold as a predictable partial discharge signal; and determine a partial discharge signal with a predicted value less than or equal to the screening threshold as an unpredictable partial discharge signal.

[0107] In a possible implementation, the processing module 302 is specifically configured to obtain partial discharge modes corresponding to each unpredictable partial discharge signal input by a user; and label the unpredictable partial discharge signals based on the partial discharge modes corresponding to each unpredictable partial discharge signal input by the user.

[0108] In a possible implementation, the processing module 302 is specifically configured to determine the first identification training sample based on the predictable partial discharge signal and the partial discharge mode of the predictable partial discharge signal, determine the second identification training sample based on the unpredictable partial discharge signal and the partial discharge mode labeled for the unpredictable partial discharge signal, and determine the sample-expanded identification training sample based on the first identification training sample and the second identification training sample.

[0109] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. As shown in the figure, the electronic device 400 of this embodiment includes a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. The processor 401 implements the steps in each of the method embodiments described above when executing the computer program 403, for example, steps 101 to 107 shown in the figure. Alternatively, the processor 401 implements the functions of each module / unit in each of the device embodiments described above when executing the computer program 403, for example, the functions of the communication module 301 and the processing module 302 shown in the figure. Figure 4 Figure 1 Figure 3

[0110] Illustratively, the computer program 403 can be divided into one or more modules / units, which are stored in the memory 402 and executed by the processor 401 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 403 in the electronic device 400. For example, the computer program 403 can be divided into the communication module 301 and the processing module 302 shown in the figure. Figure 3

[0111] The processor 401 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0112] ​​​​The memory 402 can be an internal storage unit of the electronic device 400, for example, a hard disk or a memory of the electronic device 400. The memory 402 can also be an external storage device of the electronic device 400, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 400. Further, the memory 402 can include both the internal storage unit and the external storage device of the electronic device 400. The memory 402 is used to store the computer program and other programs and data required by the terminal. The memory 402 can also be used to temporarily store data that has been output or will be output.

[0113] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or software. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0114] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0115] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art 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 application.

[0116] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0117] The integrated module / unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0118] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for predicting partial discharge modes in switchgear, characterized in that, include: Acquire multiple partial discharge signals from the switchgear; Based on the partial discharge signal screening model, the multiple partial discharge signals are screened to obtain predictable partial discharge signals and unpredictable partial discharge signals. Based on the original partial discharge pattern recognition model, predictable partial discharge signals are predicted to obtain the partial discharge patterns of the predictable partial discharge signals, and the partial discharge patterns corresponding to the unpredictable partial discharge signals are labeled. Based on the multiple partial discharge signals and the corresponding partial discharge modes of each partial discharge signal, a recognition training sample with sample augmentation is generated. Based on the expanded recognition training samples, the original partial discharge pattern recognition model is retrained to obtain a retrained partial discharge pattern recognition model. Based on the retrained partial discharge pattern recognition model, the real-time partial discharge signal of the switchgear is predicted. The partial discharge (PD) signal filtering model filters the multiple PD signals to obtain predictable and unpredictable PD signals. Prior to this, the method further includes: acquiring historical PD signals of the switchgear within a historical period, where the historical PD signals include a first PD signal and a second PD signal. The first PD signal is a PD signal that can be recognized by the original PD pattern recognition model, and the second PD signal is a PD signal that cannot be recognized by the original PD pattern recognition model. Based on the original PD pattern recognition model, feature analysis is performed on the historical PD signals to obtain feature maps for each historical PD signal. The feature map of the partial discharge signal includes the feature maps output by each convolutional layer and the feature map output by the neural network layer of the original partial discharge pattern recognition model; based on the feature maps of each historical partial discharge signal and whether the historical partial discharge signal is identifiable, screening training samples are generated; based on the screening training samples, a pre-set convolutional neural network model is trained to obtain the partial discharge signal screening model, including: Step 1, inputting the feature maps from each screening training sample into the input layer of the convolutional neural network model to obtain a fused feature map; the fused feature map is used to characterize the features of a single feature map, or the fused features of multiple feature maps; The pre-configured convolutional neural network model includes an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer. Step two: The fused feature map is processed by the first convolutional neural network to calculate and output a feature matrix. Step three: An attention matrix is ​​generated based on the attention module; the attention matrix is ​​used to increase the weights of important features in the feature matrix of the fused feature map. Step four: Based on the attention matrix and the feature matrix of the fused feature map, a feature mapping matrix is ​​obtained. Step five: The feature mapping matrix is ​​input into the second convolutional neural network, and the output layer outputs the filtering results, where the filtering results of the feature map include whether the partial discharge signal corresponding to the feature map is identifiable or not. Step six: Based on the filtering results and the corresponding filtered training samples, the loss function for this iteration of training is calculated. Step seven: If the loss function reaches a preset condition, or the number of iterations reaches the maximum number of iterations, the training process is exited, and a partial discharge signal filtering model is obtained; otherwise, the parameters in the attention module, the first convolutional neural network, and the second convolutional neural network are updated, and steps two through seven are repeated until the training process is exited.

2. The method for predicting partial discharge modes in switchgear according to claim 1, characterized in that, The step of inputting the feature maps from each of the selected training samples into the input layer to obtain the fused feature map includes: For any feature map, if the feature map is the output of a neural network layer, then the feature map is directly determined as the fused feature map; If the feature map is the output of the last convolutional layer, then the feature map is fused with the feature map output by the neural network layer to generate a fused feature map; If the feature map is the output of a convolutional layer other than the last convolutional layer, then the feature map is fused with the feature map output of the next convolutional layer to generate a fused feature map.

3. The method for predicting partial discharge modes in switchgear according to claim 1, characterized in that, The partial discharge signal filtering model filters the multiple partial discharge signals to obtain predictable and unpredictable partial discharge signals, including: The multiple partial discharge signals are input into the partial discharge signal screening model to obtain the predicted value corresponding to each partial discharge signal; Partial discharge signals with predicted values ​​greater than the screening threshold are identified as predictable partial discharge signals. Partial discharge signals whose predicted values ​​are less than or equal to the filtering threshold are identified as unpredictable partial discharge signals.

4. The method for predicting partial discharge modes in switchgear according to claim 1, characterized in that, The annotation of the partial discharge mode corresponding to the unpredictable partial discharge signal includes: Obtain the partial discharge mode corresponding to each unpredictable partial discharge signal input by the user; Based on the partial discharge modes corresponding to each unpredictable partial discharge signal input by the user, the unpredictable partial discharge signals are labeled.

5. The method for predicting partial discharge modes in switchgear according to claim 1, characterized in that, The step of generating expanded recognition training samples based on the multiple partial discharge signals and the corresponding partial discharge modes of each partial discharge signal includes: Based on the predictable partial discharge signal and the partial discharge mode of the predictable partial discharge signal, a first identification training sample is determined. Based on the unpredictable partial discharge signal and the partial discharge pattern labeled on the unpredictable partial discharge signal, a second identification training sample is determined. Based on the first recognition training sample and the second recognition training sample, the expanded recognition training sample is determined.

6. A device for predicting partial discharge modes in switchgear, characterized in that, include: The communication module is used to acquire multiple partial discharge signals from the switchgear. The processing module is used to filter the multiple partial discharge signals based on the partial discharge signal filtering model to obtain predictable partial discharge signals and unpredictable partial discharge signals; predict the predictable partial discharge signals based on the original partial discharge pattern recognition model to obtain the partial discharge patterns of the predictable partial discharge signals, and label the partial discharge patterns corresponding to the unpredictable partial discharge signals; and generate recognition training samples after sample expansion based on the multiple partial discharge signals and the partial discharge patterns corresponding to each partial discharge signal. Based on the expanded recognition training samples, the original partial discharge pattern recognition model is retrained to obtain a retrained partial discharge pattern recognition model. Based on the retrained partial discharge pattern recognition model, the real-time partial discharge signal of the switchgear is predicted. The communication module is also used to acquire historical partial discharge signals of the switchgear during a historical period. The historical partial discharge signals include a first partial discharge signal and a second partial discharge signal. The first partial discharge signal is a partial discharge signal that can be recognized by the original partial discharge pattern recognition model, and the second partial discharge signal is a partial discharge signal that cannot be recognized by the original partial discharge pattern recognition model. The processing module is further configured to perform feature analysis on the historical partial discharge signals based on the original partial discharge pattern recognition model to obtain feature maps of each historical partial discharge signal, wherein the feature map of each historical partial discharge signal includes feature maps output by each convolutional layer and feature maps output by each neural network layer of the original partial discharge pattern recognition model; generate screening training samples based on the feature maps of each historical partial discharge signal and whether the historical partial discharge signal is identifiable; and train the pre-set convolutional neural network model based on the screening training samples to obtain the partial discharge signal screening model, including: Step 1, inputting the feature maps from each screening training sample into the input layer of the convolutional neural network model to obtain a fused feature map; the fused feature map is used to characterize the features of a single feature map, or the fused features of multiple feature maps; the pre-set convolutional neural network model includes an input layer, an attention module, a first convolutional neural network, a second convolutional neural network, and an output layer; Step 2, calculating the fused feature map through the first convolutional neural network. Step 3: Output the feature matrix of the fused feature map; Step 4: Generate an attention matrix based on the attention module; the attention matrix is ​​used to increase the weight of important features in the feature matrix of the fused feature map; Step 5: Convert the attention matrix and the feature matrix of the fused feature map to obtain a feature mapping matrix; Step 6: Input the feature mapping matrix into the second convolutional neural network and output the filtering result through the output layer, wherein the filtering result of the feature map includes whether the partial discharge signal corresponding to the feature map is identifiable or the partial discharge signal corresponding to the feature map is not identifiable; Step 7: Calculate the loss function of this iteration training process based on the filtering result and the filtered training samples corresponding to the filtering result; Step 8: If the loss function reaches the preset condition, or the number of iterations reaches the maximum number of iterations, exit the training process and obtain the partial discharge signal filtering model; otherwise, update the parameters in the attention module, the first convolutional neural network and the second convolutional neural network, and repeat steps 2 to 7 until exit the training process.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor being configured to invoke and run the computer program stored in the memory to perform the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5 above.

Images