Method, device and equipment for selection of partial discharge classification model interpretability method
By comparing the interpretability methods of partial discharge classification models under different neural network architectures, and selecting the method that is similar to or has little difference from the preset results, the problem of choosing a suitable interpretability method is solved, and the accuracy and efficiency of power equipment insulation status assessment and fault diagnosis are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
- Filing Date
- 2023-08-14
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies make it difficult to select suitable interpretability methods for partial discharge classification models, affecting the accuracy and efficiency of insulation status assessment and fault diagnosis of power equipment.
By acquiring partial discharge classification models under different neural network architectures, we use various interpretability methods to interpret them, obtain the interpretation information of each method, and compare it with the preset interpretability results. We then select the interpretability method with high similarity or low difference as the target method.
A fast, robust, and stable method for selecting interpretable partial discharge classification models has been achieved, improving the accuracy and efficiency of insulation condition assessment and fault diagnosis for power equipment.
Smart Images
Figure CN117034083B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of partial discharge technology, and in particular to a method, apparatus and equipment for selecting a partial discharge classification model interpretability method. Background Technology
[0002] With the rapid development of the power industry and the high-speed expansion of the power grid, more and more new technologies are needed to support and maintain the safe operation of the power grid. The safe and stable operation of the power grid is the foundation of reliable power supply. Once the power grid fails, it will not only damage power equipment and stop power supply, affecting people's normal production and life, but also seriously endanger public safety, causing significant economic losses and adverse social impacts.
[0003] Partial discharge typically refers to the discharge process caused by the concentration of local electric fields inside or on the surface of electrical equipment insulation. During the discharge process, phenomena such as sound, light, heat, and chemical reactions occur. Under the influence of partial discharge, the aging of electrical equipment insulation materials is accelerated, eventually leading to insulation breakdown and failure. Therefore, detecting partial discharge signals is crucial for assessing the insulation condition and locating faults in electrical equipment. It directly relates to the formulation of maintenance and repair plans for electrical equipment, and thus affects the stable operation of the power system. Classifying partial discharge signals allows for the diagnosis of the insulation condition of on-site electrical equipment based on the collected partial discharge signals.
[0004] To improve the efficiency and accuracy of partial discharge fault diagnosis, machine learning and artificial intelligence technologies are currently widely used for online detection and evaluation of power equipment. With the rapid development and application of artificial intelligence, explaining the algorithm's output to users has become crucial, a concept known as ex-post interpretability. Researching ex-post interpretability methods allows us to understand the decisions made by the model, identify the causes of biases, and thus improve model performance. It also helps users understand the decisions made by artificial intelligence, increasing the credibility of the results and enabling users to use the model more effectively. However, there are numerous interpretability methods available, covering a wide range of scenarios, making it difficult to select a suitable interpretability method for partial discharge classification models. Summary of the Invention
[0005] Therefore, it is necessary to provide a method, apparatus, and equipment for selecting the interpretability method of a partial discharge classification model that can quickly determine the interpretability of the partial discharge classification model, in order to address the above-mentioned technical problems.
[0006] Firstly, this application provides a method for selecting an interpretability method for a partial discharge classification model. This method includes:
[0007] We obtain partial discharge classification models under different neural network architectures and use various interpretability methods to interpret these models. We obtain interpretation information for each interpretation method, which is used to characterize the contribution of each attribute feature in the partial discharge classification model to the classification results.
[0008] Obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results;
[0009] Based on the comparison results, the target interpretability method was selected from among various interpretability methods.
[0010] In one embodiment, obtaining a preset interpretability result corresponding to each partial discharge classification model includes:
[0011] Feature extraction was performed on the ultra-high frequency partial discharge data under different fault types to obtain initial feature data with multiple attribute features;
[0012] Based on the initial feature data under different fault types, the correlation between each attribute feature and the fault type is obtained, and based on the correlation, at least one target attribute feature is obtained among the attribute features.
[0013] For each target attribute feature, the feature data corresponding to the target attribute feature in the initial feature data is perturbed to obtain the perturbed feature data corresponding to the target attribute feature. Combined with the initial feature data and the perturbed feature data, the preset interpretability result corresponding to each partial discharge classification model is determined.
[0014] In one embodiment, by combining initial feature data and perturbation feature data, a preset interpretability result corresponding to each partial discharge classification model is determined, including:
[0015] By combining initial feature data, disturbance feature data, and partial discharge classification model, the degree of contribution of target attribute features to the target of the partial discharge classification model is determined.
[0016] For each partial discharge classification model, the contribution degree of each target attribute feature to the target of the partial discharge classification model is used to rank the target contribution degree of each target, and the target ranking result of each target attribute feature in the partial discharge classification model is obtained. The target ranking result is used to characterize the preset interpretability result corresponding to the partial discharge classification model.
[0017] In one embodiment, the interpretation information corresponding to the interpretability method is compared with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results, including:
[0018] For each partial discharge classification model, the explanatory sub-information corresponding to the partial discharge classification model in the explanatory information is compared with the preset interpretability result to obtain the comparison sub-result. The explanatory sub-information is used to characterize the degree of contribution of each attribute feature in a partial discharge classification model to the partial discharge classification result.
[0019] The comparison results of the interpretable methods are obtained by averaging the comparison results of various partial discharge classification models in the interpretation information of the interpretability method.
[0020] In one embodiment, the interpretable sub-information corresponding to the partial discharge classification model in the interpretable information is compared with a preset interpretability result to obtain a comparison sub-result, including:
[0021] Obtain the similarity between the explanatory sub-information and the preset interpretability results;
[0022] Based on the degree of similarity, the comparison sub-results are obtained.
[0023] In one embodiment, obtaining the similarity between the interpretability sub-information and the preset interpretability result includes:
[0024] Based on the degree of contribution of each attribute feature in the interpretation sub-information to the partial discharge classification result, determine the ranking result of each target attribute feature in the interpretation sub-information;
[0025] The similarity is obtained based on the ranking results and the target ranking results in the preset interpretability results.
[0026] Secondly, this application also provides a selection device for the interpretability method of a partial discharge classification model. The device includes:
[0027] The acquisition module is used to acquire partial discharge classification models under different neural network architectures, and to interpret various partial discharge classification models using multiple interpretability methods. The interpretation information corresponding to each interpretability method is obtained, and the interpretation information is used to characterize the contribution of each attribute feature in various partial discharge classification models to the partial discharge classification results.
[0028] The comparison module is used to obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results.
[0029] The selection module is used to select the target interpretability method from multiple interpretability methods based on the comparison results.
[0030] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:
[0031] We obtain partial discharge classification models under different neural network architectures and use various interpretability methods to interpret these models. We obtain interpretation information for each interpretation method, which is used to characterize the contribution of each attribute feature in the partial discharge classification model to the classification results.
[0032] Obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results;
[0033] Based on the comparison results, the target interpretability method was selected from among various interpretability methods.
[0034] Fourthly, this application also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:
[0035] We obtain partial discharge classification models under different neural network architectures and use various interpretability methods to interpret these models. We obtain interpretation information for each interpretation method, which is used to characterize the contribution of each attribute feature in the partial discharge classification model to the classification results.
[0036] Obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results;
[0037] Based on the comparison results, the target interpretability method was selected from among various interpretability methods.
[0038] Fifthly, this application also provides a computer program product. This computer program product includes a computer program that, when executed by a processor, performs the following steps:
[0039] We obtain partial discharge classification models under different neural network architectures and use various interpretability methods to interpret these models. We obtain interpretation information for each interpretation method, which is used to characterize the contribution of each attribute feature in the partial discharge classification model to the classification results.
[0040] Obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results;
[0041] Based on the comparison results, the target interpretability method was selected from among various interpretability methods.
[0042] The above-mentioned method, apparatus, and equipment for selecting interpretability methods for partial discharge classification models involve acquiring partial discharge classification models under different neural network architectures, interpreting various partial discharge classification models using multiple interpretability methods, obtaining interpretation information corresponding to each interpretability method, obtaining preset interpretability results corresponding to each partial discharge classification model, comparing the interpretation information corresponding to each interpretability method with the preset interpretability results corresponding to various partial discharge classification models for each interpretability method, obtaining comparison results, and selecting a target interpretability method from multiple interpretability methods based on the comparison results. This application selects partial discharge classification models with different architectures and applies multiple interpretability methods to these models. It obtains the interpretation of each interpretability method for each partial discharge classification model, thus acquiring interpretability information for each method. By comparing this interpretability information with a preset interpretability result, the similarity or difference between the interpretability information of each method and the preset interpretability result is obtained. This allows for the selection of interpretability methods that are more similar to the preset interpretability result from multiple partial discharge classification models with different architectures, resulting in greater robustness and stability. It also enables the rapid selection of a suitable target interpretability method for a partial discharge classification model from multiple interpretability methods, and allows for the rapid interpretation and evaluation of other partial discharge classification models using the target interpretability method. Attached Figure Description
[0043] Figure 1 This is a diagram illustrating the application environment of the method for selecting the interpretability of the partial discharge classification model in one embodiment.
[0044] Figure 2 This is a flowchart illustrating the selection method for the interpretability of a partial discharge classification model in one embodiment.
[0045] Figure 3 This is a flowchart illustrating the process of determining a preset interpretability result in one embodiment;
[0046] Figure 4 This is a flowchart illustrating the process of obtaining comparison results in one embodiment;
[0047] Figure 5A structural block diagram of a device for selecting an interpretability method for a partial discharge classification model in one embodiment;
[0048] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0050] The method for selecting the interpretability of the partial discharge classification model provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.
[0051] In one embodiment, such as Figure 2 As shown, a method for selecting the interpretability of a partial discharge classification model is provided, and this method is applied to... Figure 1 Taking the server in the example, the following steps are included:
[0052] Step 202: Obtain partial discharge classification models under different neural network architectures, and use various interpretability methods to interpret the various partial discharge classification models to obtain the interpretation information corresponding to each interpretability method.
[0053] Among them, the explanatory information is used to characterize the degree of contribution of each attribute feature in various partial discharge classification models to the partial discharge classification results.
[0054] The preferred embodiment of this application is to obtain partial discharge classification models under different neural network architectures, such as CNN partial discharge classification models based on convolutional neural networks (CNN), Bi-LSTM partial discharge classification models based on bidirectional long short-term memory networks (Bi-LSTM), and Transformer partial discharge classification models based on Transformer neural networks.
[0055] For example, UHF signal datasets labeled with different partial discharge fault types can be obtained first, such as UHF signal datasets for different fault types like floating potential defects, tip defects, air gap discharge defects, surface discharge defects, and particle discharge defects. A basic model architecture of convolutional neural networks, bidirectional long short-term memory networks, and Transformer neural networks is then constructed. This basic model architecture is trained using UHF signal datasets for different fault types to obtain partial discharge classification models under different neural network architectures.
[0056] This application utilizes multiple interpretability methods to interpret various partial discharge classification models, obtaining interpretation information corresponding to each interpretability method. This interpretation information includes the explanation of each interpretability method for each partial discharge classification model. Interpretable methods can identify attribute features that influence the model's prediction results, such as analyzing the weights of model parameters, the contribution of features, or other relevant indicators to quantify the importance of features. The interpretation information in this application is used to characterize the degree of contribution of each attribute feature in various partial discharge classification models to the partial discharge classification results. Each attribute feature is a feature extracted from a UHF signal dataset.
[0057] For example, six interpretability methods—DeepLift, GradShap, Integrated Gradients, KernelShap, DeepLiftShap, and Shapley—are selected and applied to the three neural network architectures mentioned above. Taking the DeepLift interpretability method as an example, the explanation information corresponding to the DeepLift interpretability method includes its explanation of the CNN partial discharge classification model, the Bi-LSTM partial discharge classification model, and the Transformer partial discharge classification model. That is, the contribution of each attribute feature in the CNN partial discharge classification model, the Bi-LSTM partial discharge classification model, and the Transformer partial discharge classification model to the partial discharge classification result as explained by the DeepLift interpretability method.
[0058] Step 204: Obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to each partial discharge classification model to obtain the comparison results.
[0059] This application's embodiments obtain a preset interpretability result corresponding to each partial discharge classification model, and compare the explanation of each partial discharge classification model in each interpretability method with the corresponding preset interpretability result to obtain a comparison result for each interpretability method. This comparison result can be the degree of similarity or difference between the explanation information corresponding to the interpretability method and the corresponding preset interpretability result.
[0060] Step 206: Based on the comparison results, select the target interpretability method from among the various interpretability methods.
[0061] Based on the comparison results, the embodiments of this application select the interpretability method that is highly similar to or has a small difference from the preset interpretability result as the target interpretability method.
[0062] In the above method for selecting interpretability methods for partial discharge classification models, partial discharge classification models under different neural network architectures are obtained, and various interpretability methods are used to interpret the various partial discharge classification models to obtain the interpretation information corresponding to each interpretability method. The preset interpretability results corresponding to each partial discharge classification model are obtained, and for each interpretability method, the interpretation information corresponding to the interpretability method is compared with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results. Based on the comparison results, the target interpretability method among the various interpretability methods is selected. This application embodiment selects partial discharge classification models with different architectures and applies multiple interpretability methods to these models. It obtains the interpretation of each interpretability method for each partial discharge classification model, thus acquiring the interpretation information for each method. By comparing this interpretation information with a preset interpretability result, the similarity or difference between the interpretation information of each method and the preset interpretability result is obtained. This allows for the selection of interpretability methods that are more similar to the preset interpretability result from multiple partial discharge classification models with different architectures, resulting in greater robustness and stability. It also enables the rapid selection of a suitable target interpretability method for a partial discharge classification model from multiple interpretability methods, and allows for the rapid interpretation and evaluation of other partial discharge classification models using the target interpretability method.
[0063] In one embodiment, such as Figure 3 As shown, the preset interpretability results for each partial discharge classification model include:
[0064] Step 302: Extract features from the UHF partial discharge data under different fault types to obtain initial feature data with multiple attribute features.
[0065] This application embodiment can first acquire raw partial discharge data carrying different fault type tags, such as ultra-high frequency (UHF) partial discharge data, which is a type of time-series signal data. After feature extraction from the UHF partial discharge data, initial feature data with various attribute features is obtained.
[0066] Step 304: Based on the initial feature data under different fault types, obtain the correlation degree between each attribute feature and the fault type, and based on the correlation degree, obtain at least one target attribute feature among the attribute features.
[0067] After obtaining feature data of multiple attribute features under different fault types, this application embodiment can obtain the correlation degree between each attribute feature and the fault type based on the feature data of each attribute feature and the corresponding fault type label. The correlation degree can reflect the correlation coefficient between each attribute feature and the fault type label in the feature data.
[0068] In this embodiment, after determining the correlation degree, at least one target attribute feature is obtained from each attribute feature. For example, the correlation degree is compared with a preset threshold, and the attribute feature with a correlation degree greater than the preset threshold is selected as the target attribute feature. Another example is to sort the correlation degrees between each attribute feature and the fault type, and select a preset number of attribute features with high correlation degrees as the target attribute features.
[0069] Step 306: For each target attribute feature, perturb the feature data corresponding to the target attribute feature in the initial feature data to obtain the perturbed feature data corresponding to the target attribute feature, and combine the initial feature data and the perturbed feature data to determine the preset interpretability result corresponding to each partial discharge classification model.
[0070] In this embodiment, for each target attribute feature, the feature data corresponding to the target attribute feature in the initial feature data is perturbed to obtain the perturbed feature data corresponding to the target attribute feature.
[0071] For example, noise processing or transformation processing can be performed on the feature data corresponding to target attribute feature 1 in the initial feature data to obtain the perturbation feature data corresponding to target attribute feature 1.
[0072] For example, noise processing or transformation processing can be performed on the feature data corresponding to target attribute feature 2 in the initial feature data to obtain the perturbed feature data corresponding to target attribute feature 2.
[0073] The perturbation feature data in this embodiment is based on the initial feature data and only targets a specific target attribute feature. Therefore, by combining the initial feature data, the perturbation feature data, and the partial discharge classification model, the target contribution of the target attribute feature to the classification result of the partial discharge classification model can be determined.
[0074] Specifically, by combining the initial feature data, perturbation feature data, and various partial discharge classification models, the degree of contribution of each target attribute feature to the target of each partial discharge classification model is determined. For each partial discharge classification model, the degree of contribution of each target attribute feature to the target of the partial discharge classification model is ranked to obtain the target ranking result of each target attribute feature in the partial discharge classification model. The target ranking result is used to characterize the preset interpretability result corresponding to the partial discharge classification model.
[0075] In one embodiment, such as Figure 4 As shown, the interpretable information corresponding to the interpretability method is compared with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results, including:
[0076] Step 402: For each partial discharge classification model, compare the interpretation sub-information corresponding to the partial discharge classification model in the interpretation information with the preset interpretability result to obtain the comparison sub-result.
[0077] Among them, the explanatory sub-information is used to characterize the degree of contribution of each attribute feature in a partial discharge classification model to the partial discharge classification result.
[0078] The interpretation information corresponding to each interpretability method in this application includes interpretation sub-information for each partial discharge classification model. The interpretation sub-information is compared with the corresponding preset interpretability result to obtain a comparison sub-result.
[0079] For example, the similarity between the explanatory sub-information and the preset interpretability results can be obtained, and a comparison sub-result can be obtained through the similarity between the explanatory sub-information and the preset interpretability results. For instance, the ranking result of each target attribute feature in the explanatory sub-information can be determined based on the contribution of each attribute feature in the explanatory sub-information to the partial discharge classification result; and the similarity between the ranking result and the target ranking result in the preset interpretability results can be obtained.
[0080] For example, we can obtain the consistent and inconsistent pairs between the ranking results of each target attribute feature in the explanatory sub-information and the target ranking results, and then calculate the correlation coefficient to determine the degree of similarity, such as Kendall's rank correlation coefficient and Spearman's rank correlation coefficient.
[0081] Step 404: Take the average of the comparison results corresponding to various partial discharge classification models in the interpretation information of the interpretability method to obtain the comparison results of the interpretability method.
[0082] In this embodiment, after obtaining the results of each comparison sub-result in the interpretation information of each interpretability method, the average value of each comparison sub-result is taken, thereby representing the interpretation performance of the interpretability method for partial discharge classification models under different neural network architectures.
[0083] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0084] Based on the same inventive concept, this application also provides a device for selecting the interpretability method of a partial discharge classification model to implement the selection method of the partial discharge classification model interpretability method described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for selecting the interpretability method of a partial discharge classification model provided below can be found in the limitations of the selection method of the partial discharge classification model interpretability method described above, and will not be repeated here.
[0085] In one embodiment, such as Figure 5 As shown, a selection device for the interpretability method of a partial discharge classification model is provided, comprising: an acquisition module 502, a comparison module 504, and a selection module 506, wherein:
[0086] The acquisition module 502 is used to acquire partial discharge classification models under different neural network architectures, and to interpret various partial discharge classification models using multiple interpretability methods to obtain interpretation information corresponding to each interpretation method. The interpretation information is used to characterize the contribution of each attribute feature in various partial discharge classification models to the partial discharge classification results.
[0087] The comparison module 504 is used to obtain the preset interpretability results corresponding to each partial discharge classification model, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain the comparison results.
[0088] Selection module 506 is used to select the target interpretability method from a variety of interpretability methods based on the comparison results.
[0089] In one embodiment, when the comparison module 504 performs the task of obtaining the preset interpretability result corresponding to each partial discharge classification model, it is configured to: extract features from the ultra-high frequency partial discharge data under different fault types to obtain initial feature data of multiple attribute features; based on the initial feature data under different fault types, obtain the correlation degree between each attribute feature and the fault type, and based on the correlation degree, obtain at least one target attribute feature among the attribute features; for each target attribute feature, perturb the feature data corresponding to the target attribute feature in the initial feature data to obtain the perturbed feature data corresponding to the target attribute feature, and combine the initial feature data and the perturbed feature data to determine the preset interpretability result corresponding to each partial discharge classification model.
[0090] In one embodiment, when the comparison module 504 performs the task of combining initial feature data and perturbation feature data to determine the preset interpretability result corresponding to each partial discharge classification model, it is used to: combine the initial feature data, perturbation feature data and partial discharge classification model to determine the target contribution degree of the target attribute feature to the partial discharge classification model; for each partial discharge classification model, sort the target contribution degree of each target attribute feature to the target contribution degree of the partial discharge classification model to obtain the target ranking result of each target attribute feature in the partial discharge classification model, and the target ranking result is used to characterize the preset interpretability result corresponding to the partial discharge classification model.
[0091] In one embodiment, the comparison module 504 compares the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models. When obtaining the comparison results, it is used to: for each partial discharge classification model, compare the interpretation sub-information corresponding to the partial discharge classification model in the interpretation information with the preset interpretability results to obtain comparison sub-results. The interpretation sub-information is used to characterize the contribution of each attribute feature in a partial discharge classification model to the partial discharge classification results; and take the average of the comparison sub-results corresponding to various partial discharge classification models in the interpretation information of the interpretability method to obtain the comparison results of the interpretability method.
[0092] In one embodiment, when the comparison module 504 compares the explanatory sub-information corresponding to the partial discharge classification model in the explanatory information with the preset interpretability result to obtain the comparison sub-result, it is used to: obtain the similarity between the explanatory sub-information and the preset interpretability result; and obtain the comparison sub-result based on the similarity.
[0093] In one embodiment, when the comparison module 504 performs the similarity assessment between the obtained interpretability sub-information and the preset interpretability result, it is used to: determine the ranking result of each target attribute feature in the interpretability sub-information based on the contribution of each attribute feature in the interpretability sub-information to the partial discharge classification result; and obtain the similarity based on the ranking result and the target ranking result in the preset interpretability result.
[0094] Each module in the aforementioned partial discharge classification model interpretability method selection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0095] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores partial discharge data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a selection method for the interpretability of a partial discharge classification model.
[0096] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0097] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0098] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.
[0099] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0100] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0101] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0102] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0103] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for selecting an interpretability method for a partial discharge classification model, characterized in that, The method includes: Partial discharge classification models under different neural network architectures are obtained, and various interpretability methods are used to interpret the various partial discharge classification models to obtain the interpretation information corresponding to each interpretation method. The interpretation information is used to characterize the contribution of each attribute feature in the various partial discharge classification models to the partial discharge classification results. Obtain the preset interpretability result corresponding to each of the partial discharge classification models, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability result corresponding to each of the partial discharge classification models to obtain a comparison result; the comparison result is used to characterize the degree of similarity or difference between the interpretation information and the preset interpretability result. Based on the comparison results, the target interpretability method is selected from the multiple interpretability methods; The step of obtaining the preset interpretability result corresponding to each partial discharge classification model includes: Feature extraction was performed on the ultra-high frequency partial discharge data under different fault types to obtain initial feature data with multiple attribute features; Based on the initial feature data under the different fault types, the correlation between each attribute feature and the fault type is obtained, and the attribute feature with a correlation greater than a preset threshold is taken as the target attribute feature, thereby obtaining at least one target attribute feature among the attribute features. For each target attribute feature, the feature data corresponding to the target attribute feature in the initial feature data is perturbed to obtain the perturbed feature data corresponding to the target attribute feature. The preset interpretability result corresponding to each partial discharge classification model is determined by combining the initial feature data and the perturbed feature data.
2. The method according to claim 1, characterized in that, The step of combining the initial feature data and the perturbation feature data to determine the preset interpretability result corresponding to each partial discharge classification model includes: By combining the initial feature data, the disturbance feature data, and the partial discharge classification model, the degree of contribution of the target attribute feature to the target of the partial discharge classification model is determined. For each partial discharge classification model, the contribution degree of each target attribute feature to the target of the partial discharge classification model is ranked to obtain the target ranking result of each target attribute feature in the partial discharge classification model. The target ranking result is used to characterize the preset interpretability result corresponding to the partial discharge classification model.
3. The method according to claim 2, characterized in that, The step of comparing the interpretation information corresponding to the interpretability method with the preset interpretability results corresponding to various partial discharge classification models to obtain comparison results includes: For each partial discharge classification model, the explanatory sub-information corresponding to the partial discharge classification model in the explanatory information is compared with the preset interpretability result to obtain a comparison sub-result. The explanatory sub-information is used to characterize the degree of contribution of each attribute feature in a partial discharge classification model to the partial discharge classification result. The comparison results of the interpretability method are obtained by averaging the comparison results corresponding to various partial discharge classification models in the interpretation information of the interpretability method.
4. The method according to claim 3, characterized in that, The step of comparing the explanatory sub-information corresponding to the partial discharge classification model in the explanatory information with the preset interpretability result to obtain a comparison sub-result includes: Obtain the degree of similarity between the explanatory sub-information and the preset interpretability result; Based on the degree of similarity, the comparison sub-results are obtained.
5. The method according to claim 4, characterized in that, The process of obtaining the similarity between the explanatory sub-information and the preset interpretability result includes: Based on the degree of contribution of each attribute feature in the interpretation sub-information to the partial discharge classification result, determine the ranking result of each target attribute feature in the interpretation sub-information; The similarity is obtained based on the sorting result and the target sorting result in the preset interpretability result.
6. A selection device for the interpretability method of a partial discharge classification model, characterized in that, The device includes: The acquisition module is used to acquire partial discharge classification models under different neural network architectures, and to interpret the various partial discharge classification models using multiple interpretability methods to obtain interpretation information corresponding to each interpretation method. The interpretation information is used to characterize the contribution of each attribute feature in the various partial discharge classification models to the partial discharge classification results. The comparison module is used to obtain the preset interpretability result corresponding to each of the partial discharge classification models, and for each interpretability method, compare the interpretation information corresponding to the interpretability method with the preset interpretability result corresponding to each of the partial discharge classification models to obtain the comparison result; The selection module is used to select the target interpretability method among the multiple interpretability methods based on the comparison results. The comparison module is used to: obtain the preset interpretability result corresponding to each partial discharge classification model when performing the process of: Feature extraction was performed on the ultra-high frequency partial discharge data under different fault types to obtain initial feature data with multiple attribute features; Based on the initial feature data under the different fault types, the correlation between each attribute feature and the fault type is obtained, and the attribute feature with a correlation greater than a preset threshold is taken as the target attribute feature, thereby obtaining at least one target attribute feature among the attribute features. For each target attribute feature, the feature data corresponding to the target attribute feature in the initial feature data is perturbed to obtain the perturbed feature data corresponding to the target attribute feature. The preset interpretability result corresponding to each partial discharge classification model is determined by combining the initial feature data and the perturbed feature data.
7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.
9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.