GIS Partial Discharge Pattern Recognition Model and Its Training Method

Through the training method of phase matching neural network, the phase characteristics of GIS local discharge mode are extracted and the similarity is calculated, which solves the problem of degradation of the recognition accuracy of the model under different conditions, and achieves higher recognition accuracy and robustness.

CN118608876BActive Publication Date: 2024-11-08ELECTRIC POWER RES INST OF STATE GRID ZHEJIANG ELECTRIC POWER COMAPNY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411070944.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-08
Estimated Expiration
2044-08-06

AI Technical Summary

Technical Problem

The existing GIS local discharge pattern recognition model has decreased recognition accuracy under different data sources and experimental conditions, lacks robustness and versatility, and is difficult to adapt to changing scenarios.

Method used

A phase matching neural network is used to optimize the loss function to improve the model recognition accuracy and robustness by obtaining the training set, extracting phase features, building a grayscale map, calculating similarity and correcting the probability.

Benefits of technology

It significantly improves the accuracy of discharge pattern recognition and the adaptability of the system, and enhances the sensitivity to complex discharge phenomena and the reliability of identification results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118608876B_ABST
    Figure CN118608876B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of partial discharge classification and recognition, and discloses a GIS partial discharge pattern recognition model and a training method thereof. The method of the present invention comprises obtaining a training atlas; extracting phase features to generate corresponding grayscale images and grayscale values, constructing a phase matching neural network and inputting the training atlas into the phase matching neural network to output the probability of each discharge pattern of each grayscale image; classifying the grayscale images, normalizing the grayscale values ​​of the same grayscale images and the probability of each discharge pattern and then accumulating them to obtain statistical characteristic values ​​characterizing various types of discharge patterns; based on the statistical characteristic values ​​of various types of discharge patterns and the grayscale values ​​of the grayscale images, calculating the similarity between each grayscale image and each discharge pattern, presetting a correction coefficient and correcting the similarity based on the coefficient; setting a model training loss function to optimize the parameters of the phase matching neural network. The present invention improves the reliability of discharge pattern recognition and enhances the adaptability and robustness of the system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of partial discharge classification and recognition, and in particular relates to a GIS partial discharge pattern recognition model and a training method thereof. Background Art

[0002] Gas insulated switchgear (GIS) is an electrical device that uses SF6 as an insulating medium and is known for its excellent insulation performance and arc extinguishing efficiency. It has the advantages of small size and high reliability and is widely used in high-voltage power transmission and distribution systems.

[0003] With the increase in the number of operations, in the complex operating environment of the power system, the equipment will inevitably face the challenges of insulation material aging and the emergence of potential defects. These problems may induce partial discharge, thus posing a serious threat to the safe and stable operation of the power system. The main forms of partial discharge include floating electrode discharge, free particle discharge, air gap discharge, needle tip discharge and insulator surface discharge. The degree of damage to equipment caused by different types of partial discharge varies significantly. Therefore, it is necessary to identify the discharge mode based on the partial discharge data, so as to warn and evaluate possible faults, guide the power system to take corresponding measures and reduce losses. In recent years, machine learning technology has been introduced into the classification of partial discharge spectrum signals, which has significantly improved the classification efficiency. By capturing the frequency phase and discharge amount of the discharge pulse, the partial discharge PRPD spectrum is constructed, and the graphical presentation of the discharge information is realized. Each spectrum contains rich discharge point information, which effectively reduces the data size of the PD record. At the same time, relying on the continuous monitoring of phase information, long-term tracking of the discharge phenomenon is achieved. However, the current PRPD spectrum research is highly dependent on data sets under specific equipment and unified experimental parameters, which to a certain extent limits the generalization ability of the model. When experimental conditions or data sources change, such as equipment replacement or on-site GIS data collection, model performance is often affected and recognition accuracy decreases. To solve this problem, it is necessary to explore strategies to improve the adaptability of the model so that it can bridge the gap between different data sources and experimental conditions and maintain robust performance in different scenarios, thereby promoting the further development of gas-insulated switchgear fault diagnosis technology. Summary of the invention

[0004] Based on the above-mentioned shortcomings and deficiencies in the prior art, one of the objects of the present invention is to at least solve one or more of the above-mentioned problems in the prior art. In other words, one of the objects of the present invention is to provide a GIS local discharge pattern recognition model and its training method that meet one or more of the above-mentioned requirements, so as to improve the recognition accuracy of the model while ensuring the robustness and versatility under different conditions.

[0005] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a GIS local discharge pattern recognition model training method, comprising the following steps: S1, obtaining a training atlas, wherein the training atlas comprises a plurality of original images and their corresponding discharge pattern labels; S2, extracting phase features from each original image in the training atlas to generate a corresponding grayscale image and grayscale value, constructing a phase matching neural network, and inputting each grayscale image and its corresponding discharge pattern label into the phase matching neural network to output the probability of each discharge pattern of each grayscale image; S3, classifying the grayscale images based on the discharge pattern labels, normalizing the grayscale values ​​of the same grayscale images and the probability of each discharge pattern, and then accumulating them to obtain statistical characteristic values ​​characterizing each type of discharge pattern; S4, calculating the similarity between each grayscale image and each discharge pattern based on the statistical characteristic values ​​of each type of discharge pattern and the grayscale value of the grayscale image, presetting a correction coefficient and correcting the probability of each discharge pattern of each grayscale image based on the correction; S5, setting a phase matching neural network training loss function, optimizing the phase matching neural network parameters based on the training loss function and the probability of each discharge pattern of each grayscale image obtained after correction in step S4 to complete the training and obtain the recognition model.

[0007] As a preferred solution, step S2 extracts phase features from each original image in the training atlas to generate a corresponding grayscale image and grayscale value, including the steps of: preprocessing each original image in the training atlas to hide information unrelated to the phase features, wherein the information unrelated to the phase features includes coordinate axes and text; binarizing each preprocessed original image to obtain a grayscale image; assigning a black discharge point in the grayscale image to 1 and a white background point to 0, and using bilinear interpolation to adjust each grayscale image to the same resolution N to obtain the grayscale value of each grayscale image, and the grayscale value of the i-th grayscale image is recorded as A i , ,and A i The value range is [0 , 1].

[0008] As a preferred solution, step S2 of constructing a phase matching neural network includes the steps of: setting multiple layers of neurons to form an input layer, an output layer and at least one intermediate layer, thereby constructing the phase matching neural network; the output layer is provided with an activation function f(x).

[0009] As a preferred solution, each grayscale image input from the input layer and its corresponding discharge mode label are recorded as X , and the probability of each discharge mode of each grayscale image output by the output layer is recorded as Y , then the expression of the probability of a discharge mode of a grayscale image is

[0010] In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

[0011] As a preferred solution, the statistical characteristic value expression of a discharge mode of a pixel point in a grayscale image is: In the formula, , i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode.

[0012] As a preferred solution, before the preset correction coefficient is used in step S4 and the probability of each discharge mode of each grayscale image is corrected based on the preset correction coefficient, the following steps are further included: mapping the similarity to a preset mapping interval and then using an activation function to calculate to obtain the similarity after mapping; the preset mapping interval is [0,1]; the similarity between each grayscale image before mapping and each discharge mode is recorded as S ic , and the similarity between each grayscale image after mapping and each discharge mode is recorded as C ic .

[0013] As a preferred solution, the similarity S ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, A i Indicates i The grayscale value of the grayscale image, W c Represents the statistical characteristic value of a discharge mode of a pixel in a grayscale image; C ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, S ic Indicates i Grayscale image andc The similarity of the discharge patterns.

[0014] As a preferred solution, the expression of the probability of a discharge mode of a certain grayscale image after correction is: In the formula, λ represents the correction factor, i Indicates i A grayscale image, c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

[0015] As a preferred solution, the loss function is: In the formula, i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode, d ic is a symbolic function, if the sample i The true discharge type is equal to c If yes, it takes 1, otherwise it takes 0.

[0016] In a second aspect, the present invention provides a GIS local discharge pattern recognition model, based on the training method described in the first aspect, including a phase matching neural network, wherein the phase matching neural network is provided with multiple layers of neurons; the multiple layers of neurons form an input layer, an output layer and at least one intermediate layer; the output layer is provided with an activation function f(x).

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. Combining neural networks, black box structures, and activation functions to build a phase matching neural network, and improving the existing loss function to improve the analysis accuracy of the phase matching neural network, so that this network structure can more effectively capture the phase information in the image, not only improving the speed of data processing, but also significantly enhancing the sensitivity to weak phase differences in complex discharge phenomena. Compared with traditional methods, this network can more accurately analyze the phase features in the image, providing a solid foundation for subsequent discharge pattern recognition.

[0019] 2. The original image is preprocessed, binarized, and processed at the same resolution. These processing measures effectively remove noise and redundant information in the image while retaining key phase features. This series of optimization measures work together to significantly improve the accuracy of discharge pattern recognition.

[0020] 3. Set the correction coefficient to correct the probability of each discharge mode in each grayscale image, effectively compensating for the recognition deviation caused by image quality, environmental factors or inherent defects of the algorithm. By continuously learning and optimizing the correction coefficient, the present invention can gradually approach the real discharge mode distribution, making the recognition result closer to the actual situation. This innovation not only improves the reliability of discharge mode recognition, but also enhances the adaptability and robustness of the system.

[0021] Further or more detailed beneficial effects will be described in detail in conjunction with specific examples in the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 It is a flowchart of the model training method described in Example 1 of the present invention.

[0024] Figure 2 It is a schematic diagram of the structure of the bit matching neural network described in Example 1 of the present invention.

[0025] Figure 3 Schematic diagram of the original image according to the first embodiment of the present invention.

[0026] Figure 4 It is a schematic diagram of the grayscale image described in the first embodiment of the present invention.

[0027] Figure 5 It is a schematic diagram of the accuracy of the phase matching neural network training in Example 3 of the present invention.

[0028] Figure 6 It is a schematic diagram of the number of loss iterations during phase matching neural network training in Example 3 of the present invention.

[0029] Figure 7 It is a schematic diagram of the predicted value-true value matrix before the phase matching neural network is corrected in Example 3 of the present invention.

[0030] Figure 8 It is a schematic diagram of the predicted value-true value matrix after correction of the phase matching neural network in Example 3 of the present invention. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] In the following description, multiple embodiments of the present invention are provided, and different embodiments may be replaced or combined, so the present invention may also be considered to include all possible combinations of the same and / or different embodiments described. Therefore, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more of A, B, C, and D, all other possible combinations, even though the embodiment may not be clearly described in the following text.

[0033] The following description provides examples and does not limit the scope, applicability or examples set forth in the claims. Changes may be made to the functions and arrangements of the elements described without departing from the scope of the present invention. Various processes or components may be appropriately omitted, substituted or added to each example. For example, the described method may be performed in an order different from the described order, and various steps may be added, omitted or combined. In addition, the features described in some examples may be combined in other examples.

[0034] In order to facilitate a better understanding of the embodiments of the present invention, before explaining the specific implementation modes of the present invention in detail, its application scenarios are first described.

[0035] The discharge pattern recognition model and its training method described in the embodiments of this specification are applied to power equipment status monitoring, industrial equipment fault diagnosis, and battery health management in the new energy field. In these scenarios, the application of the discharge pattern recognition model and its training method can timely warn of potential faults or abnormalities, avoiding damage or shutdown of equipment due to long-term poor condition, thereby ensuring the reliability and continuity of the production system to provide high-quality user experience and reliable performance.

[0036] Embodiment 1:

[0037] like Figure 1As shown, the present embodiment provides a GIS partial discharge pattern recognition model training method, comprising the following steps: S1, obtaining a training atlas, wherein the training atlas comprises a plurality of original images and their corresponding discharge pattern labels; S2, extracting phase features from each original image in the training atlas to generate a corresponding grayscale image and grayscale value, constructing a phase matching neural network, and inputting each grayscale image and its corresponding discharge pattern label into the phase matching neural network to output the probability of each discharge pattern of each grayscale image; S3, classifying the grayscale images based on the discharge pattern labels, normalizing the grayscale values ​​of the same grayscale images and the probability of each discharge pattern, and then accumulating them to obtain statistical characteristic values ​​characterizing various types of discharge patterns; S4, calculating the similarity between each grayscale image and each discharge pattern based on the statistical characteristic values ​​of various types of discharge patterns and the grayscale values ​​of the grayscale images, presetting a correction coefficient, and correcting the probability of each discharge pattern of each grayscale image based on the correction; S5, setting a phase matching neural network training loss function, optimizing the phase matching neural network parameters based on the training loss function and the probability of each discharge pattern of each grayscale image obtained after the correction in step S4 to complete the training and obtain the recognition model.

[0038] Understandably, most neural networks are like this Figure 2 The black box structure shown may contain convolutional layers, pooling layers, fully connected layers, activation functions, etc. of various structures. The specific parameters corresponding to each layer do not need to be calculated manually. It is only necessary to determine the dimensions of the input layer and the output layer.

[0039] Specifically, this embodiment provides a preferred implementation of step S2, wherein the phase feature extraction is performed on each original image in the training atlas to generate a corresponding grayscale image and grayscale value, including the steps of: preprocessing each original image to hide information irrelevant to the phase feature, wherein the information irrelevant to the phase feature includes coordinate axes and text; binarizing each preprocessed original image to obtain a grayscale image; assigning a black discharge point in the grayscale image to 1 and a white background point to 0, and adjusting each grayscale image to the same resolution N using bilinear interpolation to obtain the grayscale value of each grayscale image, and the grayscale value of the i-th grayscale image is recorded as A i , ,and A i The value range is [0 , 1].

[0040] It is understandable that before phase feature extraction, the image needs to be preprocessed to remove irrelevant information such as coordinate axes and text. Figure 3 and Figure 4As shown in Figure 2, since each discharge mode has different distribution characteristics in phase, in order to unify the phase and discharge intensity information of different images and obtain the statistical data of discharge characteristics, it is necessary to use bilinear interpolation to adjust the images to the same resolution N.

[0041] Specifically, this embodiment provides a preferred implementation method, in which step S2 constructs a phase matching neural network, including the steps of: setting multiple layers of neurons to form an input layer, an output layer and at least one intermediate layer, thereby constructing the phase matching neural network; the output layer is provided with an activation function f(x).

[0042] Specifically, this embodiment provides a preferred implementation method, where each grayscale image input from the input layer and its corresponding discharge mode label are recorded as X , and the probability of each discharge mode of each grayscale image output by the output layer is recorded as Y , then the expression of the probability of a discharge mode of a grayscale image is

[0043] In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

[0044] Specifically, this embodiment provides a preferred implementation mode, in which the statistical characteristic value expression of a discharge mode of a pixel point in a grayscale image is: In the formula, , i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode.

[0045] It is understandable that resolution refers to the number of pixels contained in a unit length. After all the images in the training atlas are accumulated, the feature-extracted images will show the statistical law of the local discharge phase. This statistical law can be used to match the images in the data atlas. In essence, the statistically obtained discharge information matrix is ​​matched with the actual discharge matrix, so the method of calculating the similarity of the matrix can be used.

[0046] Specifically, this embodiment provides a preferred implementation method, before the preset correction coefficient in step S4 is used to correct the probability of each discharge mode of each grayscale image based on the preset correction coefficient, the steps further include: mapping the similarity to a preset mapping interval and then using an activation function to calculate to obtain the similarity after mapping; the preset mapping interval is [0,1]; the similarity between each grayscale image before mapping and each discharge mode is recorded as S ic , and the similarity between each grayscale image after mapping and each discharge mode is recorded as C ic .

[0047] Specifically, this embodiment provides a preferred implementation method, the similarity S ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, A i Indicates i The grayscale value of the grayscale image, W c Represents the statistical characteristic value of a discharge mode of a pixel in a grayscale image; C ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, S ic Indicates i The grayscale image and c The similarity of the discharge patterns.

[0048] It is understandable that since the discharge density of images with different sampling rates is quite different, The similarities obtained may also be very different, even with differences in magnitude. To avoid interference to the network when similarity information with differences in magnitude is introduced into the neural network, this embodiment firstly converts the similarity S ic Map it to the range of 0 to 1, and then calculate it using the softmax function.

[0049] Specifically, this embodiment provides a preferred implementation mode, and the expression of the probability of a certain discharge mode of a certain grayscale image after correction is: In the formula, λ represents the correction factor, i Indicates i A grayscale image,c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

[0050] It can be understood that the similarity between each grayscale image after mapping and each discharge mode is C ic (denoted as difference one) directly reflects the preliminary performance of the model in feature extraction and pattern recognition. The higher the similarity, the more accurately the model can capture the features related to the specific discharge pattern from the grayscale image. However, due to the design limitations of the model itself or the complexity of the data itself, this preliminary judgment may not be completely accurate. On the other hand, there is a difference between the probability of each discharge mode of each grayscale image output by the output layer of the phase matching neural network and each grayscale image and its corresponding discharge mode label input by the input layer (denoted as difference two), which further reveals the deviation of the model in the prediction or classification process. This difference two not only contains the accuracy information of the model in identifying the discharge pattern, but also may reflect the generalization ability and robustness of the model in the classification of complex discharge patterns. This embodiment introduces the correction coefficient λ , cleverly solves the problem of weight distribution of the above two differences (difference 1 and difference 2). λ , the system can adaptively adjust the contribution of these two differences to the final optimization goal. Specifically, when the model performs well in the initial similarity judgment but the classification probability is biased, λ The weight of the difference value 2 may be increased, prompting the model to pay more attention to improving the classification accuracy during the training process; on the contrary, if the model is relatively accurate in classification probability, but the similarity judgment needs to be improved, λ This may reduce the weight of difference 2 and instead strengthen the optimization of feature extraction and pattern recognition capabilities. This adaptive weight allocation mechanism not only improves the flexibility and efficiency of model training, but also enables the model to dynamically adjust the optimization direction according to the current performance during the training process at different stages, ultimately achieving more accurate and robust discharge pattern recognition capabilities. Through the self-learning process of the network, λ The specific value of will gradually converge to an optimal solution, making the model reach the best state in terms of overall performance.

[0051] Specifically, this embodiment provides a preferred implementation, and the loss function is: In the formula, i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode,d ic is a symbolic function, if the sample i The true discharge type is equal to c If yes, it takes 1, otherwise it takes 0.

[0052] It is understandable that for multi-classification tasks, the commonly used loss function is the cross entropy loss function

[0053] In the above formula, N is the number of samples, M is the number of categories, d ic is a sign function, which takes 1 if the true label of sample i is equal to c, otherwise it takes 0. y ic is the probability that sample i belongs to category c after being calculated by the neural network. Although this classic loss function can train the model well, it can only reflect the probability value corresponding to the correct discharge mode, and does not consider the corresponding probability of the wrong discharge mode of other items. For the judgment results of the two models a = {0.6, 0.4, 0, 0, 0}, b = {0.6, 0.1, 0.1, 0.1, 0.1}, although both of them identify the correct discharge mode as 0.6, the result a has a greater misleading effect on the judgment of the discharge mode. If the loss value of a is expected to be slightly greater than b, the above formula cannot meet this condition. Therefore, this embodiment improves the loss function. The phase matching neural network is trained by the improved loss function of this embodiment. Compared with the completely random initial parameters, the phase correction parameters can provide the model with a better initial setting, so that the network has better performance and improves the local discharge mode recognition and early warning capabilities of gas insulated switchgear.

[0054] Embodiment 2:

[0055] This embodiment provides a GIS local discharge pattern recognition model based on the training method described in Example 1, including a phase matching neural network, wherein the phase matching neural network is provided with multiple layers of neurons; the multiple layers of neurons form an input layer, an output layer and at least one intermediate layer; the output layer is provided with an activation function f(x).

[0056] Embodiment three:

[0057] In order to verify the effectiveness of a GIS partial discharge pattern recognition model based on a neural network and its training method described in this specification, this embodiment is based on a multi-layer perceptron, and introduces additional information of the phase into the network by modifying the bias. In this embodiment, a multi-layer perceptron MLP control group is set up during the experiment. The control group has the same network structure as the phase-corrected one, that is, the number of neurons, the learning rate and other hyperparameters, but no feature extraction correction is performed, that is, λSet to 0. At the same time, this implementation selected convolutional neural network (CNN), lightweight neural network (MobileNet), and deep residual network (Resnet-50) for comparison. During the training process, the ratio of the training set to the test set is 7:3.

[0058] Table 1 shows the accuracy of different networks after training. The accuracy of MLP is generally lower than that of other networks. This is because MLP has relatively few network parameters and is slightly worse than other neural networks in extracting image feature information. Figure 5 After the MLP was modified by introducing the phase matching proposed in this paper, the MLP-PM model had an accuracy rate of 82.63% in identifying GIS partial discharge, which was 4.8% higher than the MLP network, and 3.59%, 8.38% and 0.6% higher than Resnet-50, MobileNet and CNN respectively. This is of great significance for the recognition of partial discharge patterns of gas-insulated switchgear on site. The relationship between the accuracy and loss value of the neural network model training in Table 1 and the iteration is shown in the figure. Figure 5 and Figure 6 As shown in the figure, CNN, MobileNet, and Resnet-50 can converge quickly in fewer iterations. After adding the phase matching module PM, the convergence speed of MLP-PM is significantly faster than that of MLP.

[0059] Figure 7 and Figure 8 The predicted value-true value matrix before and after phase matching correction is shown. Labels 0 to 4 represent five types of discharges, namely floating electrode discharge, free particle discharge, air gap discharge, needle tip discharge, and insulator surface discharge. It can be seen that the recognition rate of surface discharge is low. This is because the mechanism of surface defects is more complicated and has no obvious phase characteristics compared to other discharge types. After the phase correction network is introduced, the recognition accuracy of other discharge types is improved except for floating electrode discharge.

[0060] Table 1 The accuracy of different network training

[0061]

[0062] Based on the above, this embodiment verifies the effectiveness of the phase matching neural network of this specification.

[0063] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0064] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0065] The above is only an exemplary embodiment of the present disclosure, and the scope of the present disclosure cannot be limited thereto. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the specification and practicing the disclosure here, those skilled in the art will easily think of the implementation scheme of the present disclosure. The present invention is intended to cover any variation, use or adaptive change of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the technical field not recorded in the present disclosure. The description and examples are regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. GIS partial discharge pattern recognition model training method, characterized in that: Includes steps: S1. Obtain a training atlas, wherein the training atlas includes a plurality of original images and their corresponding discharge mode labels; S2, extracting phase features from each original image in the training atlas to generate corresponding grayscale images and grayscale values, constructing a phase matching neural network, and inputting each grayscale image and its corresponding discharge mode label into the phase matching neural network to output the probability of each discharge mode of each grayscale image; S3, classifying the grayscale images based on the discharge mode labels, normalizing the grayscale values ​​of the same grayscale images and the probability of each discharge mode, and then accumulating them to obtain statistical characteristic values ​​representing various discharge modes; S4, based on the statistical characteristic values ​​of various discharge modes and the grayscale values ​​of the grayscale images, calculating the similarity between each grayscale image and each discharge mode, presetting a correction coefficient and correcting the probability of each discharge mode of each grayscale image based on the correction coefficient; S5. Setting a phase matching neural network training loss function, optimizing the phase matching neural network parameters based on the training loss function and the probability of each discharge mode of each grayscale image corrected in step S4 to complete the training and obtain a recognition model.

2. The GIS partial discharge pattern recognition model training method according to claim 1 is characterized in that: Step S2 extracts phase features from each original image in the training atlas to generate a corresponding grayscale image and grayscale value, including the following steps: Preprocessing each original image in the training atlas to hide information irrelevant to the phase feature, wherein the information irrelevant to the phase feature includes coordinate axes and text; Binarize each preprocessed original image to obtain a grayscale image; The black discharge points in the grayscale image are assigned a value of 1 and the white background points are assigned a value of 0. Each grayscale image is adjusted to the same resolution N using bilinear interpolation to obtain the grayscale value of each grayscale image. The grayscale value of the i-th grayscale image is recorded as A i , ,and A i The value range is [0 , 1].

3. The GIS partial discharge pattern recognition model training method according to claim 2 is characterized in that: Step S2 constructs a phase matching neural network, including the following steps: Arranging multiple layers of neurons to form an input layer, an output layer and at least one intermediate layer, thereby constructing the phase matching neural network; The output layer is provided with an activation function f(x).

4. The GIS partial discharge pattern recognition model training method according to claim 3 is characterized in that: Each grayscale image input from the input layer and its corresponding discharge mode label are recorded as X , and the probability of each discharge mode of each grayscale image output by the output layer is recorded as Y , then the expression of the probability of a discharge mode of a grayscale image is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

5. The GIS partial discharge pattern recognition model training method according to claim 4 is characterized in that: The statistical characteristic value expression of a discharge mode of a certain pixel point in a grayscale image is: In the formula, , i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode.

6. The GIS partial discharge pattern recognition model training method according to claim 5 is characterized in that: Before the preset correction coefficient in step S4 and the correction of each discharge mode probability of each grayscale image based on the preset correction coefficient, the following steps are also included: After mapping the similarity to the preset mapping interval, the activation function is used for calculation to obtain the mapped similarity; The preset mapping interval is [0,1]; The similarity between each grayscale image before mapping and each discharge mode is recorded as S ic , and the similarity between each grayscale image after mapping and each discharge mode is recorded as C ic .

7. The GIS partial discharge pattern recognition model training method according to claim 6 is characterized in that: The similarity S ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, A i Indicates i The grayscale value of the grayscale image, W c Represents the statistical characteristic value of a discharge mode of a certain pixel in a certain grayscale image; The similarity C ic The expression is In the formula, i Indicates i A grayscale image, c Indicates c Discharge mode, S ic Indicates i The grayscale image and c The similarity of the discharge patterns.

8. The GIS partial discharge pattern recognition model training method according to claim 7 is characterized in that: The expression of the probability of a discharge mode of a grayscale image after correction is: In the formula, λ represents the correction factor, i Indicates i A grayscale image, c Indicates c Discharge mode, n Indicates the layer number of neurons that output the probability results of the discharge pattern in the phase matching neural network.

9. The GIS partial discharge pattern recognition model training method according to claim 8, characterized in that: The loss function is: In the formula, i Indicates i A grayscale image, j Represents a pixel in the grayscale image. N Represents the total number of pixels in the grayscale image. c Indicates c Discharge mode, d ic is a symbolic function, if the sample i The true discharge type is equal to c If yes, it takes 1, otherwise it takes 0.

10. A GIS partial discharge pattern recognition model, based on the training method according to any one of claims 1 to 9, characterized in that: It includes a phase matching neural network, wherein the phase matching neural network is provided with multiple layers of neurons; The multiple layers of neurons form an input layer, an output layer and at least one intermediate layer; The output layer is provided with an activation function f(x).