High-resistance grounding fault artificial intelligence identification method and system with interpretability

Abstract

The invention relates to an interpretable high-resistance grounding fault artificial intelligence identification method and system, and the method comprises the following steps: building a radiation type power distribution network model through simulation software, and collecting the zero-sequence current signals of an HIF signal and a non-HIF signal through the simulation software; performing multi-layer wavelet decomposition on the acquired zero-sequence current signal by using discrete wavelet transform; calculating a standard deviation of the detail coefficient of each level as a fault feature; constructing a fault feature data set; using the training set to train the constructed HIF recognition model of the high-resistance grounding fault, and establishing nonlinear mapping between fault features and recognition results; testing the trained HIF recognition model by using a test set, and then calculating an average value of the output of the test set as a prediction reference value of the HIF recognition model to a test set sample; and on the basis of the obtained prediction reference value, according to the SHAP attribution theory, local interpretation and global interpretation are performed on the decision basis for realizing fault recognition of the HIF recognition model.

Description

Technical Field

[0001] The present application relates to the technical field of relay protection for distribution networks, and more specifically, to an interpretable artificial intelligence identification method and system for high-resistance grounding faults. Background Art

[0002] As the link in the power system that is most closely connected to users, the distribution network provides electricity security for residents' lives and industrial production. In recent years, with the increase in the complexity of distribution lines, the probability of distribution network failures has also increased. Rapid identification, location and removal of distribution network failures is an important way to enhance power supply reliability and a challenge for the development of modern intelligent distribution networks.

[0003] At present, the main identification methods of high-resistance grounding faults in distribution networks include time domain method, frequency domain method, time-frequency domain method and artificial intelligence method. The time domain method usually uses the time domain characteristics such as nonlinear distortion and asymmetry of the fault signal as the basis for HIF identification, which is easy to analyze and derive, and has a clear physical meaning; the frequency domain method captures the low-frequency component and high-frequency component characteristics of the fault waveform, amplifies the fault frequency domain characteristics to achieve fault identification; the time-frequency domain method comprehensively utilizes the time domain and frequency domain characteristics of the fault signal to identify HIF, and has a strong ability to characterize features. The artificial intelligence method usually combines signal processing technology with artificial intelligence algorithms to form a HIF intelligent identification method. This method can realize adaptive learning of deep features in massive data and avoid the limitations of artificially extracted fault features on the experience of the prophet.

[0004] The HIF recognition method based on artificial intelligence can adaptively extract and identify fault features by learning the deep feature expression in massive data samples, avoiding the limitation of artificially extracted fault features by prophetic experience, and showing good adaptability and accuracy in dealing with complex and changeable fault features, providing new ideas for the reliable identification of HIF. However, the current HIF recognition method based on artificial intelligence usually produces a black box recognition model after offline training, and it is difficult to extract any information about the model behavior from its structure or weights. It is difficult to understand and explain the decision-making process of such models, resulting in a lack of trust and acceptance by users. Therefore, the ideal HIF recognition model should be able to track the decision-making basis of the model, make the decision-making process of the model more transparent, and enhance the interpretability of the HIF recognition process and the credibility of the recognition results.

[0005] Prior art, such as the Chinese patent with patent number "CN115171896A", discloses a system and method for predicting the long-term mortality risk of critically ill patients. The prediction system includes: a cloud database module for storing the physiological data of critically ill patients; a data processing module for acquiring the physiological data of the critically ill patients and preprocessing it to obtain input variables; a long-term mortality risk prediction module for inputting the input variables into the LSTM neural network model, predicting the long-term mortality risk of the critically ill patients, and interpreting the prediction results using the SHAP method. The present invention also provides a method for predicting the long-term mortality risk of critically ill patients using the above system. The present invention takes into account both the accuracy and interpretability of the prediction results, and has good application prospects.

[0006] The problems existing in the above-mentioned prior art are: using the SHAP method to interpret the LSTM neural network requires considering the special processing mechanism for sequential information in the time series contained in the internal structure of the LSTM neural network, and the need to process these complex time-related structures may face more difficulties; such as: the sequential dependency processing is highly complex, the model structure is highly complex, more computing resources are required, and the processing efficiency is low. Summary of the invention

[0007] In order to solve the above technical problems, the present invention proposes an interpretable artificial intelligence identification method and system for high-resistance grounding faults.

[0008] The technical solution of the present invention is as follows:

[0009] The present invention proposes an interpretable artificial intelligence identification method for high-resistance grounding faults, comprising the following steps:

[0010] Step S1, using simulation software to establish a radial distribution network model, and collecting zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software;

[0011] Step S2, using discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal; and calculating the standard deviation of the detail coefficient of each layer as the fault feature; constructing a fault feature data set, and dividing the data set into a training set and a test set;

[0012] Step S3, using the training set to train the constructed high-resistance grounding fault HIF identification model, and establishing a nonlinear mapping between fault features and identification results;

[0013] Step S4, using the test set to test the trained high-resistance grounding fault HIF identification model, and then calculating the average value of the output of the test set as the prediction benchmark value of the high-resistance grounding fault HIF identification model for the test set samples;

[0014] Step S5, based on the obtained prediction reference value and according to the SHAP attribution theory, a local explanation and a global explanation are made for the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification.

[0015] As a preferred implementation, the simulation software is used to establish a radial distribution network model, which includes: an infinite power source, a transformer, an overhead line, a cable line, a zero-sequence current transformer, an arc suppression coil, and a load module.

[0016] As a preferred implementation, the high-resistance grounding fault HIF identification model constructed by using training set training, wherein: the high-resistance grounding fault HIF identification model is constructed by using back-propagation BP neural network framework.

[0017] As a preferred implementation, the discrete wavelet transform is used to perform wavelet decomposition of several layers on the collected zero-sequence current signal; and the standard deviation of the detail coefficient of each layer is calculated as the fault feature; the wavelet decomposition formula is as follows:

[0018]

[0019] Among them, X is the waveform signal, cD r is the detail coefficient of the rth layer, cA r is the approximation coefficient of the rth layer, and R is the total number of decomposition layers;

[0020] The calculation formula of fault characteristics is:

[0021]

[0022] Among them, S r is the fault feature of the rth layer, k = 1, 2, 3, …, n.

[0023] As a preferred implementation, based on the obtained prediction reference value, according to the SHAP attribution theory, a local explanation and a global explanation are made for the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification; according to the SHAP attribution theory, an additive explanation model is constructed to fit the high-resistance grounding fault HIF identification model, and a feasible explanation scheme is provided for the prediction results of the high-resistance grounding fault HIF identification model. The specific expression formula is:

[0024]

[0025] Where f(x) is the high-resistance grounding fault HIF identification model, g(x) is the additive explanation model, n is the number of features of sample x, φ 0 is the prediction benchmark value of the high-resistance grounding fault HIF identification model for the sample, that is, the mean of the model's prediction results for all samples, φ j is the jth feature x of sample xj The Shapley value of is used to quantify the feature x j The impact on the model prediction results.

[0026] As a preferred implementation, the local interpretation of the high-resistance grounding fault HIF identification model is calculated according to the SHAP attribution theory, and the specific calculation formula is as follows:

[0027]

[0028] In the formula, {x 1 ,x 2 ,…,x n} represents the set of all features of n-dimensional sample x, and S is the set that does not contain feature x j The feature subset of |S| is the number of feature elements in set S. represents the weight, f x (S∪{x j}) and f x (S) respectively represent the features x j and does not contain feature x j The model prediction results in the case of feature x j The Shapley value of φ j is feature x j The mean of the marginal contributions in different feature subsets.

[0029] As a preferred implementation, the global interpretation of the HIF recognition model is calculated according to the SHAP attribution theory, and the specific calculation formula is as follows:

[0030]

[0031] In the formula, φ j (x i ) is the i-th sample x i The jth feature x j The Shapley value of Y j is feature x j The global Shapley value of , m is the total number of samples.

[0032] On the other hand, the present invention also provides an interpretable high-resistance grounding fault artificial intelligence identification system, comprising:

[0033] The data acquisition module uses simulation software to establish a radial distribution network model, and collects zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software;

[0034] The feature calculation module uses discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal; calculates the standard deviation of the detail coefficient of each layer as the fault feature; constructs a fault feature data set, and divides the data set into a training set and a test set;

[0035] Model training module, which uses the training set to train the constructed high-resistance grounding fault HIF identification model and establishes a nonlinear mapping between fault features and identification results;

[0036] The benchmark value determination module uses the test set to test the trained high-resistance grounding fault HIF identification model, and then calculates the average value of the output of the test set as the prediction benchmark value of the high-resistance grounding fault HIF identification model for the test set samples;

[0037] The model interpretation module, based on the obtained predicted benchmark value and according to the SHAP attribution theory, makes local and global interpretations on the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification.

[0038] On the other hand, the present invention further provides an electronic device having a computer program stored thereon, wherein when the computer program is executed by a processor, an interpretable artificial intelligence identification method for high-resistance grounding faults as described in any embodiment of the present invention is implemented.

[0039] On the other hand, the present invention also provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement an explainable high-resistance grounding fault artificial intelligence identification method as described in any embodiment of the present invention.

[0040] The present invention has the following beneficial effects:

[0041] 1. The SHAP method is used to interpret the BP neural network, quantifying the contribution of different input features to the model decision results from both global and local perspectives, thereby enhancing the persuasiveness of the recognition model output results.

[0042] 2. Increase the credibility of recognition results and make the model's decision-making process more transparent.

[0043] 3. It makes up for the shortcoming of existing AI-based HIF recognition methods that usually produce black-box recognition models after offline training, and solves the problem that such models are difficult to understand and explain their decision-making process, resulting in lack of trust and acceptance among users.

[0044] 4. From a physical perspective, the relationship between the characteristic values ​​and the high-frequency components contained in the HIF waveform can be used to better identify HIF.

[0045] 5. The key features that have the greatest impact on model decisions are different in different individual samples, but the features with higher global importance have an important impact on the prediction results of different individual samples. The joint effect of multiple features improves the accuracy of the HIF intelligent recognition model in predicting samples, making the model more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 This is a schematic diagram of the method flow of Example 1;

[0048] Figure 2 This is a simulation model diagram of a radial distribution network;

[0049] Figure 3 A local interpretation diagram for a high-resistance ground fault sample;

[0050] Figure 4 A local interpretation diagram for a non-high-resistance ground fault sample;

[0051] Figure 5 It is a graph of the global importance of features;

[0052] Figure 6 Schematic diagram of discrete wavelet transform multi-level decomposition. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0055] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0056] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0057] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0058] Embodiment 1:

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be combined with the specific embodiments of the present application and refer to the attached Figure 1 , clearly and completely describe the technical solution of the present invention.

[0060] To solve the problems in the prior art, the present invention provides an interpretable artificial intelligence identification method for high-resistance grounding faults, comprising the following steps:

[0061] Step S1, using simulation software to establish a radial distribution network model, and collecting zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software;

[0062] The radial distribution network model is established using EMTDC / PSCAD simulation software. Figure 2 As shown, it includes: infinite power supply, transformer, overhead line, cable line, zero-sequence current transformer, arc suppression coil and load module. The zero-sequence current signal of high-resistance grounding fault HIF signal and non-high-resistance grounding fault HIF signal is collected through the zero-sequence current transformer in the simulation software.

[0063] In this embodiment, the system operating frequency is 50 Hz, the sampling rate is 8 kHz, and the parameters of the cable line and the overhead line are shown in Table 1:

[0064] Table 1 Line parameters

[0065]

[0066] Table 2 describes the HIF or HIF interference events that occur at different fault locations (FP) and on different lines. Among them, the capacitor switching adopts a parallel three-phase capacitor model, the excitation inrush current is simulated by a single-phase transformer without load, the low-resistance grounding fault is simulated by a low-resistance model (5Ω-100Ω), the load switching adopts a three-phase asymmetric load model, and the HIF model is the Emanuel arc model. In addition, in order to conform to the actual engineering practice, the asynchronous closing of capacitor switching is added to the experiment to simulate the transient situation of non-fault. In this study, three-phase asynchronous closing means that phase A is connected to the system first, and phases B and C are connected to the system at the same time with the same delay. The initial fault angle is set to 0°, 30°, 60°, 90° and 120°.

[0067] Table 2 Line parameters

[0068]

[0069] Step S2, using discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal; and calculating the standard deviation of the detail coefficient of each layer as the fault feature; constructing a fault feature data set, and dividing the data set into a training set and a test set;

[0070] The zero-sequence current signal is decomposed into 6 layers of wavelets using discrete wavelet transform, and the standard deviation of the detail coefficients from the first to the sixth level is calculated as the fault feature. A fault feature data set is constructed, and the data set is divided into a training set and a test set for training and testing the high-resistance grounding fault HIF identification model.

[0071] Discrete wavelet transform DWT obtains multi-level detail coefficients and approximate coefficients of waveform signals through multi-layer low-pass and high-pass filters, such as Figure 6 As shown. Among them, LF represents a low-pass filter, HF represents a high-pass filter, cA represents an approximation coefficient, which represents the large-scale low-frequency information of the waveform signal, and cD represents a detail coefficient, which represents the small-scale high-frequency information of the waveform signal. Therefore, DWT can be used to decompose the waveform signal into multiple scales, which is expressed as follows:

[0072]

[0073] Among them, X is the waveform signal, cD r is the detail coefficient of the rth layer, cA r is the approximation coefficient of the rth layer, and R is the total number of decomposition layers. In order to reduce the amount of redundant information from the original waveform or its transformed format, feature extraction is required to find unique parameters that can characterize the important features of the waveform. The standard deviation of the signal, as an indicator reflecting the degree of discreteness of the signal distribution, can provide information about the level of change in the signal frequency distribution, as shown in the following formula:

[0074]

[0075] Among them, S r is the fault feature of the rth layer, k = 1, 2, 3, …, n.

[0076] Theoretically, the more decomposition levels of DWT, the more detailed the decomposed signal, and the more significant the extracted feature effect. However, considering that the sampling rate of the data collected in this embodiment is 8kHz, the maximum frequency of the signal is 4KHz according to the sampling theorem, then the 6-layer wavelet decomposition can obtain the standard deviation characteristics of the detail components representing the relatively high frequency bands of 2000-4000Hz and 1000-2000Hz, the characteristics representing the relatively medium frequency bands of 500-1000Hz and 250-500Hz, and the characteristics representing the relatively low frequency bands of 125-250Hz and 62.5-125Hz. HIF and its interference basically do not show any difference in the base frequency, so there is no need to further decompose the zero-sequence current signal at the existing sampling rate, so the standard deviation of the detail coefficients from the first to the sixth level is used as the fault feature.

[0077] Step S3, using the zero-sequence current data in the training set to train the HIF recognition model, and establish a nonlinear mapping between fault features and recognition results;

[0078] Step S4, using the test set to test the trained HIF recognition model, and then calculating the average value of the output of the test set as the prediction benchmark value of the HIF recognition model for the test set samples.

[0079] Step S5, based on the SHAP attribution theory, make local and global explanations for the decision basis of the HIF recognition model to realize fault recognition.

[0080] According to the SHAP attribution theory, an additive explanation model g(x) is constructed based on the Shapley value in cooperative game theory to fit the complex model f(x), providing a feasible explanation scheme for the prediction results of the complex model f(x), as shown in the following formula:

[0081]

[0082] Where n is the number of features of sample x, φ 0 is the model's prediction benchmark value for the sample, that is, the mean of the model's prediction results for all samples, φ j is the jth feature x of sample x j The Shapley value of is used to quantify the feature x j The impact on the model prediction results.

[0083] Therefore, the prediction result of the model for any sample can be expressed as the sum of the prediction benchmark value and the Shapley value of all features of the sample. The Shapley value of a feature is used to measure the importance of the feature to the model prediction result and reflects the contribution of the feature to the model prediction result. The specific calculation principle is as follows:

[0084]

[0085] In the formula, {x 1 ,x 2 ,…,x n} represents the set of all features of n-dimensional sample x, and S is the set that does not contain feature x j The feature subset of |S| is the number of feature elements in set S. represents the weight, f x (S∪{x j}) and f x (S) respectively represent the features x j and does not contain feature x j The model prediction results in the case of feature x j The Shapley value of φ j is feature x j The mean of the marginal contributions in different feature subsets.

[0086] In this embodiment, one sample with model prediction results of HIF and one sample with non-HIF are randomly selected from all test samples, and the SHAP attribution theory is used to explain how the HIF intelligent recognition model predicts these two samples. Figure 3 is the HIF sample prediction graph, Figure 4 This is the prediction graph for non-HIF samples. In the figure, black represents features with positive Shapley values, and gray represents features with negative Shapley values. The larger the color area, the larger the absolute value of the corresponding feature's Shapley value, which also means that the feature has a greater impact on the model's decision results. The predicted baseline value of the sample plus the Shapley value of each feature is the model output value, that is, the probability of predicting that the sample is a HIF sample. Figure 3 The Shapley value of feature D4 in this sample is positive 0.2909, indicating that it has a positive impact on the prediction result of the sample as HIF and has the greatest contribution.

[0087] For any given sample, the Shapley value of each feature reflects the degree of influence of the feature on the prediction result and the positive or negative influence. That is, the feature with a larger absolute value of the Shapley value has a greater impact on the model prediction result. The positive or negative Shapley value indicates that the feature will increase or decrease the output result of the model, which greatly increases the local interpretability of the complex model f(x). Based on this, if a feature x of all samples isj The absolute value of the Shapley value is taken as the average value to measure the global importance of the feature, which can be used to globally explain the complex model f(x). The global importance of the feature is measured as follows:

[0088]

[0089] In the formula, φ j (x i ) is the i-th sample x i The jth feature x j The Shapley value of Y j is feature x j The global Shapley value of , m is the total number of samples.

[0090] In this embodiment, the feature global importance ranking diagram is obtained by sorting the average absolute value of the feature Shapley value, such as Figure 5 As shown. The horizontal axis in the figure represents the average absolute value of the feature Shapley value, that is, the global Shapley value of the feature, which represents the average impact of the feature on the model prediction results. Different colors represent the average impact of the feature on different prediction results of the model. Black represents HIF, white represents non-HIF, and the vertical axis is the feature name. In short, the larger the global Shapley value of the feature, the greater the contribution of the corresponding feature to the model prediction results and the higher the importance. Therefore, features D4, D5 and D6 contribute more to the model prediction results and are more important.

[0091] From the perspective of physical meaning, the HIF waveform contains abundant high-frequency components, which leads to the values ​​of these features being higher than those under interference events. The higher the feature value, the higher the probability of HIF occurrence. Therefore, the model can use D4, D5 and D6, three components that express high-frequency features, to identify HIF. In summary, the physical correlation between each feature and the model identification results is basically consistent with the global analysis conclusion of SHAP attribution theory, indicating that the interpretation results of SHAP attribution theory are consistent with objective physical laws.

[0092] Combination Figure 3 , Figure 4 and Figure 5 It can be found that the key features that have the greatest impact on the model decision are not the same in different individual samples, but the features with higher global importance have an important impact on the prediction results of different individual samples. Due to the differences between different individual samples, it is easy to make wrong judgments if only a single feature is used to analyze and judge whether a sample is HIF. Under the joint action of multiple features, the accuracy of the HIF intelligent recognition model in predicting samples is greatly improved, making the model more reliable.

[0093] In summary, the present invention proposes an interpretable artificial intelligence identification method for high-resistance grounding faults, which quantifies the contribution of different input features to the model decision results from both global and local perspectives, enhances the persuasiveness of the output results of the identification model, increases the credibility of the identification results, and makes the decision-making process of the model more transparent.

[0094] Embodiment 2:

[0095] This embodiment provides an interpretable high-resistance grounding fault artificial intelligence identification system, including:

[0096] The data acquisition module uses simulation software to establish a radial distribution network model, and collects zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software;

[0097] The feature calculation module uses discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal; calculates the standard deviation of the detail coefficient of each layer as the fault feature; constructs a fault feature data set, and divides the data set into a training set and a test set;

[0098] Model training module, which uses the training set to train the constructed high-resistance grounding fault HIF identification model and establishes a nonlinear mapping between fault features and identification results;

[0099] The benchmark value determination module uses the test set to test the trained high-resistance grounding fault HIF identification model, and then calculates the average value of the output of the test set as the prediction benchmark value of the high-resistance grounding fault HIF identification model for the test set samples;

[0100] The model interpretation module, based on the obtained predicted benchmark value and according to the SHAP attribution theory, makes local and global interpretations on the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification.

[0101] Embodiment three:

[0102] This embodiment provides an electronic device having a computer program stored thereon. When the computer program is executed by a processor, an interpretable artificial intelligence identification method for high-resistance grounding faults as described in any embodiment of the present invention is implemented.

[0103] Embodiment 4:

[0104] This embodiment provides a computer-readable medium for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement an interpretable high-resistance grounding fault artificial intelligence identification method as described in any embodiment of the present invention.

[0105] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0106] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk and other media that can store program code. The above description is only an embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An interpretable artificial intelligence identification method for high-resistance grounding faults, characterized in that: The following steps are involved: Step S1, using simulation software to establish a radial distribution network model, and collecting zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software; Step S2, using discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal; and calculating the standard deviation of the detail coefficient of each layer as the fault feature; Construct a fault feature data set and divide the data set into a training set and a test set; Step S3, using the training set to train the constructed high-resistance grounding fault HIF identification model, and establishing a nonlinear mapping between fault features and identification results; Step S4, using the test set to test the trained high-resistance grounding fault HIF identification model, and then calculating the average value of the output of the test set as the prediction benchmark value of the high-resistance grounding fault HIF identification model for the test set samples; Step S5, based on the obtained prediction reference value and according to the SHAP attribution theory, a local explanation and a global explanation are made for the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification.

2. The interpretable artificial intelligence identification method for high-resistance grounding faults according to claim 1 is characterized in that: The simulation software is used to establish a radial distribution network model, which includes: an infinite power supply, a transformer, an overhead line, a cable line, a zero-sequence current transformer, an arc suppression coil and a load module.

3. The interpretable artificial intelligence identification method for high-resistance grounding fault according to claim 1 is characterized in that: The high-resistance grounding fault HIF identification model constructed by training with the training set, wherein: the high-resistance grounding fault HIF identification model is constructed by adopting the back-propagation BP neural network framework.

4. The interpretable artificial intelligence identification method for high-resistance grounding faults according to claim 1 is characterized in that: The discrete wavelet transform is used to perform wavelet decomposition of several layers on the collected zero-sequence current signal; and the standard deviation of the detail coefficient of each layer is calculated as the fault feature; the formula of the wavelet decomposition is as follows: Among them, X is the waveform signal, cD r is the detail coefficient of the rth layer, cA r is the approximation coefficient of the rth layer, and R is the total number of decomposition layers; The calculation formula of fault characteristics is: Among them, S r is the fault feature of the rth layer, k = 1, 2, 3, …, n.

5. The interpretable artificial intelligence identification method for high-resistance grounding faults according to claim 1 is characterized in that: As described above, based on the obtained predicted benchmark value, according to the SHAP attribution theory, the decision basis for the high-resistance grounding fault HIF identification model to achieve fault identification is explained locally and globally; According to the SHAP attribution theory, an additive explanatory model is constructed to fit the high-resistance grounding fault HIF identification model, which provides a feasible explanation scheme for the prediction results of the high-resistance grounding fault HIF identification model. The specific expression formula is: Where f(x) is the high-resistance grounding fault HIF identification model, g(x) is the additive explanatory model, n is the number of features of sample x, φ0 is the prediction benchmark value of the high-resistance grounding fault HIF identification model for the sample, that is, the mean of the model's prediction results for all samples, φ j is the jth feature x of sample x j The Shapley value of is used to quantify the feature x j The impact on the model prediction results.

6. The interpretable artificial intelligence identification method for high-resistance grounding faults according to claim 5 is characterized by: The local explanation of the high-resistance grounding fault HIF identification model calculated according to the SHAP attribution theory is as follows: In the formula, {x 1 ,x 2 ,…,x n } represents the set of all features of n-dimensional sample x, and S is the set that does not contain feature x j The feature subset of |S| is the number of feature elements in set S. represents the weight, f x (S∪{x j }) and f x (S) respectively represent the features x j and does not contain feature x j The model prediction results in the case of feature x j The Shapley value of φ j is feature x j The mean of the marginal contributions in different feature subsets.

7. The interpretable artificial intelligence identification method for high-resistance grounding faults according to claim 5 is characterized by: The global explanation of the HIF recognition model is calculated according to the SHAP attribution theory. The specific calculation formula is as follows: In the formula, φ j (x i ) is the i-th sample x i The jth feature x j The Shapley value of Y j is feature x j The global Shapley value of , m is the total number of samples.

8. An interpretable artificial intelligence identification system for high-resistance grounding faults, characterized in that: include: The data acquisition module uses simulation software to establish a radial distribution network model, and collects zero-sequence current signals of high-resistance grounding fault HIF signals and non-high-resistance grounding fault HIF signals through the simulation software; The feature calculation module uses discrete wavelet transform to perform wavelet decomposition of several layers on the collected zero-sequence current signal, and calculates the standard deviation of the detail coefficient of each layer as the fault feature; Construct a fault feature data set and divide the data set into a training set and a test set; Model training module, which uses the training set to train the constructed high-resistance grounding fault HIF identification model and establishes a nonlinear mapping between fault features and identification results; The reference value determination module uses the test set to test the trained high-resistance grounding fault HIF identification model, and then calculates the average value of the output of the test set as the prediction reference value of the high-resistance grounding fault HIF identification model for the test set samples; The model interpretation module, based on the obtained predicted benchmark value and according to the SHAP attribution theory, makes local and global interpretations on the decision basis of the high-resistance grounding fault HIF identification model to realize fault identification.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, an explainable artificial intelligence identification method for high-resistance grounding faults as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, an interpretable artificial intelligence identification method for high-resistance grounding faults as described in any one of claims 1 to 7 is implemented.

Images