Electric shock fault identification method, device and equipment for power distribution network

Abstract

The embodiment of the invention provides a power distribution network electric shock fault identification method, device and equipment. The method comprises the steps that multi-modal data in a preset time period are acquired from a power distribution network, and the multi-modal data comprise various types of power distribution network parameters of the power distribution network; matrix conversion is carried out on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters; according to the characteristic matrix of the power distribution network parameters, determining attention weights between the power distribution network parameters and other power distribution network parameters; wherein the attention weight represents correlation between the power distribution network parameter and other power distribution network parameters; determining a fusion feature matrix of the power distribution network according to the attention weight and the feature matrix of the power distribution network parameters; and inputting the fusion feature matrix of the power distribution network into a preset identification model for identification to obtain an electric shock fault identification result of the power distribution network. The method is used for achieving the technical effect of improving the recognition accuracy of the electric shock fault of the power distribution network.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, device and equipment for identifying electric shock faults in a power distribution network. Background Art

[0002] During the operation of the power distribution network, electric shock faults may occur, wherein the electric shock faults may cause personal injury to the staff. Therefore, it is necessary to identify the electric shock faults in the power distribution network.

[0003] In the prior art, electric shock faults in the power distribution network can be identified based on single data of the power distribution network. However, this method cannot accurately identify electric shock faults in the power distribution network.

[0004] Furthermore, there is an urgent need for a solution that can effectively identify electric shock faults in the power distribution network. Summary of the invention

[0005] The embodiments of the present application provide a method, device and equipment for identifying electric shock faults in a power distribution network, so as to achieve the technical effect of improving the accuracy of identifying electric shock faults in a power distribution network.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying an electric shock fault in a power distribution network, comprising:

[0007] Acquire multimodal data within a preset time period from a power distribution network, wherein the multimodal data includes multiple types of power distribution network parameters of the power distribution network; and perform matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters;

[0008] Determine, according to the characteristic matrix of the power distribution network parameter, an attention weight between the power distribution network parameter and other power distribution network parameters; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters;

[0009] Determining a fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters;

[0010] The fused feature matrix of the power distribution network is input into a preset recognition model for recognition, so as to obtain an electric shock fault recognition result of the power distribution network.

[0011] In a possible implementation manner, determining the attention weight between the power distribution network parameter and other power distribution network parameters according to the characteristic matrix of the power distribution network parameter includes:

[0012] For each type of distribution network parameter, a characteristic matrix of the distribution network parameter is linearly transformed using different weight matrices to obtain a query characteristic matrix, a key characteristic matrix and a value characteristic matrix of the distribution network parameter; wherein the query characteristic matrix, the key characteristic matrix and the value characteristic matrix are characteristic matrices of the distribution network parameter after different linear transformations;

[0013] For each type of power distribution network parameter, the attention weight between the power distribution network parameter and other power distribution network parameters is determined according to the query feature matrix of the power distribution network parameter and the key feature matrix of other power distribution network parameters.

[0014] In a possible implementation manner, determining a fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters includes:

[0015] For each type of power distribution network parameter, the initial fusion feature of the power distribution network parameter is determined according to each attention weight between the power distribution network parameter and other power distribution network parameters, and the value feature matrix of each of the other power distribution network parameters; wherein the initial fusion feature of the power distribution network parameter represents the fusion parameter feature between the power distribution network parameter and each of the other power distribution network parameters;

[0016] The initial fusion features of the distribution network parameters are concatenated to obtain a fusion feature matrix of the distribution network.

[0017] In a possible implementation, the multimodal data includes:

[0018] Residual current signal of the distribution network, voltage fluctuation of the distribution network, load status of the distribution network, and environmental parameters of the distribution network.

[0019] In a possible implementation, the method further includes:

[0020] Obtaining a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario; wherein the first set of parameters to be trained includes a plurality of fused data, and the second set of parameters to be trained includes a plurality of fused data; the fused data represents fused data of multimodal data;

[0021] Repeat the following steps until a preset condition is met: randomly determine multiple scenarios, train the initial model according to the first set of parameters to be trained in each scenario, and obtain a trained initial model; train the trained initial model according to the second set of parameters to be trained in each scenario, and obtain a retrained initial model; and use the retrained initial model as the initial model;

[0022] Among them, the initial model after retraining obtained when the preset condition is met is the preset recognition model.

[0023] In a possible implementation manner, obtaining a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario includes:

[0024] Acquire first historical fusion data of the power distribution network in each scenario;

[0025] Using a pre-trained generator to perform data expansion on the first historical fusion data in the scenario to obtain a plurality of first simulated fusion data in the scenario; merging the plurality of first simulated fusion data in the scenario with the historical fusion data in the scenario to obtain each fusion data in the scenario;

[0026] Each fusion data in the scenario is split and processed to obtain a first set of parameters to be trained and a second set of parameters to be trained in the scenario.

[0027] In a possible implementation, the method further includes:

[0028] Acquire historical multimodal data of the power distribution network in each scenario, wherein the historical multimodal data has an electric shock event type; and perform feature fusion processing on the historical multimodal data to obtain first historical fusion data of the power distribution network;

[0029] Performing data expansion processing on the first historical fusion data based on the initial generator to obtain second historical fusion data;

[0030] The initial discriminator and the initial generator are trained using the first historical fusion data and the second historical fusion data to obtain the pre-trained generator corresponding to the initial generator.

[0031] In a possible implementation, the initial model is trained according to the first set of parameters to be trained in each scenario to obtain a trained initial model; and the trained initial model is trained according to the second set of parameters to be trained in each scenario to obtain a retrained initial model, including:

[0032] Using a plurality of fused data in the first parameter set to be trained in each scenario, the initial model is trained, and when it is determined that the value of the first loss function is minimized, the trained initial model is obtained;

[0033] The trained initial model is trained according to the multiple fused data in the second parameter set to be trained in each scenario, and when it is determined that the value of the obtained second loss function is minimum, the re-trained initial model is obtained.

[0034] In a possible implementation manner, the information of the power distribution network in different scenarios includes at least one of the following: load status, environmental parameters, and grounding point.

[0035] In a second aspect, an embodiment of the present application provides an electric shock fault identification device for a power distribution network, comprising: an acquisition module, configured to acquire multimodal data within a preset time period from the power distribution network, wherein the multimodal data includes multiple types of power distribution network parameters of the power distribution network; and performing matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters;

[0036] A first determination module is used to determine the attention weight between the power distribution network parameter and other power distribution network parameters according to the characteristic matrix of the power distribution network parameter; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters;

[0037] A second determination module, configured to determine a fusion characteristic matrix of the power distribution network according to the attention weight and the characteristic matrix of the power distribution network parameters;

[0038] The identification module is used to input the fusion feature matrix of the distribution network into a preset identification model for identification, so as to obtain the electric shock fault identification result of the distribution network.

[0039] In a third aspect, an embodiment of the present application provides an electric shock fault identification device for a power distribution network, comprising: a memory, a processor;

[0040] The memory stores computer-executable instructions;

[0041] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0042] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0043] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0044] The embodiments of the present application provide a method, device and equipment for identifying electric shock faults in a distribution network. By acquiring multimodal data including various types of distribution network parameters within a preset time period in the distribution network, the distribution network parameters are subjected to matrix conversion to obtain a feature matrix. The feature matrix is ​​then used to determine an attention weight matrix that characterizes the correlation between each distribution network parameter and other distribution network parameters. A fused feature matrix is ​​then determined based on the weight matrix and the feature matrix of the distribution network parameters. Finally, a preset recognition model is used to operate the fused feature matrix to obtain an identification result of an electric shock fault in the distribution network, thereby improving the recognition accuracy of electric shock faults in the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0046] Figure 1 A schematic diagram of a flow chart of a method for identifying electric shock faults in a power distribution network provided in this application;

[0047] Figure 2 A schematic diagram of the structure of the electric shock fault identification device for the power distribution network provided by the present application;

[0048] Figure 3 A schematic diagram of the structure of the electric shock fault identification device for the power distribution network provided in this application.

[0049] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0050] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0051] During the operation of the power distribution network, electric shock faults may occur. Among them, electric shock faults may cause personal injury to the staff. For example, there is residual current in the power distribution network. If the residual current is large, if the operator accidentally touches the residual current in the power distribution network during the operation, it may cause damage to the muscles, nerves and internal tissues of the body, thereby causing harm to the body. Therefore, it is necessary to identify the electric shock faults in the power distribution network to reduce the safety hazards caused by electric shock faults.

[0052] In the prior art, electric shock faults in the distribution network can be identified based on single data of the distribution network. The prior art performs function processing on the single data in the distribution network and uses the processed value obtained by the function processing to extract features. The extracted feature data is then used to train the recognition model, and the trained recognition model can be used to identify electric shock faults in the distribution network. Since the analysis data of the distribution network is single, the prior art cannot accurately identify electric shock faults in the distribution network.

[0053] Based on the above analysis, it can be seen that the prior art has a technical problem of low accuracy in identifying electric shock faults in power distribution networks.

[0054] The method, device and equipment for identifying electric shock faults in a distribution network provided in the present application obtain multimodal data including various types of distribution network parameters within a preset time period in the distribution network, perform matrix conversion on the distribution network parameters, and obtain a feature matrix; then use the feature matrix to determine an attention weight matrix that characterizes the correlation between each distribution network parameter and other distribution network parameters; then determine a fused feature matrix based on the weight matrix and the feature matrix of the distribution network parameters; finally, use a preset recognition model to operate on the fused feature matrix to obtain an electric shock fault identification result for the distribution network, thereby solving the above-mentioned technical problems.

[0055] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0056] Figure 1 A flow chart of a method for identifying electric shock faults in a power distribution network provided in this application, such as Figure 1 As shown, the method includes:

[0057] S101. Acquire multimodal data within a preset time period from a power distribution network, wherein the multimodal data includes various types of power distribution network parameters of the power distribution network; and perform matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters.

[0058] The multimodal data may be data obtained by preprocessing the original data. As an example, preprocessing the original data may include denoising the original data based on wavelet transform or Kalman filtering.

[0059] It may also include normalizing the data to be processed based on the following formula (1):

[0060] (1),

[0061] in, For the data after standardization, For the data to be processed, and are the mean and variance of the data respectively.

[0062] It may also include aligning the timestamps of the processed data using interpolation methods.

[0063] S102. Determine the attention weight between the power distribution network parameter and other power distribution network parameters according to the characteristic matrix of the power distribution network parameter; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters.

[0064] S103: Determine a fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters.

[0065] It should be noted that the fusion feature matrix not only retains the key information of each distribution network parameter, but also increases the interactive information between different distribution network parameters.

[0066] S104: Input the fused feature matrix of the power distribution network into a preset recognition model for recognition, and obtain the electric shock fault recognition result of the power distribution network.

[0067] The method for identifying electric shock faults in a distribution network provided in an embodiment of the present application obtains multimodal data including multiple types of distribution network parameters within a preset time period in the distribution network, performs matrix conversion on the distribution network parameters, and obtains a feature matrix; then uses the feature matrix to determine an attention weight matrix that characterizes the correlation between each distribution network parameter and other distribution network parameters; then determines a fused feature matrix based on the weight matrix and the feature matrix of the distribution network parameters; finally, uses a preset recognition model to operate on the fused feature matrix to obtain an electric shock fault recognition result for the distribution network, thereby improving the recognition accuracy of electric shock faults in the distribution network.

[0068] In some specific implementations, the multimodal data includes: a residual current signal of the power distribution network, a voltage fluctuation of the power distribution network, a load state of the power distribution network, and environmental parameters of the power distribution network.

[0069] It should be noted that the above multimodal data may be data in various forms including text, tables, pictures, etc.

[0070] As an example, the residual current signal may be textual record data of residual current fluctuations, the load status of the power distribution network may be load status data based on table statistics, and the environmental parameters of the power distribution network may be pictures taken of the power distribution network.

[0071] By deploying sensors and other collection devices in the distribution network, multimodal data of the distribution network can be captured at multiple key operating nodes of the distribution network.

[0072] In these implementations, multimodal data of the power distribution network is acquired, and the multimodal data reflects the situation of the power distribution network from multiple aspects, so as to improve the accuracy of subsequent data processing results.

[0073] In some of these embodiments, to ensure that the weight matrix fully exploits the correlation characteristics between different distribution network parameters, the attention weights between the distribution network parameters and other distribution network parameters are determined according to the characteristic matrix of the distribution network parameters, including:

[0074] First, for each type of distribution network parameters, different weight matrices are used to perform linear transformation on the characteristic matrix of the distribution network parameters to obtain the query characteristic matrix, key characteristic matrix and value characteristic matrix of the distribution network parameters; wherein the query characteristic matrix, key characteristic matrix and value characteristic matrix are characteristic matrices of the distribution network parameters after different linear transformations.

[0075] Second, for each type of distribution network parameter, the attention weights between the distribution network parameters and other distribution network parameters are determined according to the query feature matrix of the distribution network parameters and the key feature matrices of other distribution network parameters.

[0076] It should be noted that, for each power distribution network parameter, it may include a plurality of sampling data within a preset time period.

[0077] As an example, the preset time period is divided into a plurality of sub-time periods according to the step length T, a data is sampled in each sub-time period, and each sampled data constitutes a power distribution network parameter.

[0078] The distribution network parameters can be processed based on the self-attention mechanism. Specifically, in the first step, different weight matrices are used to perform linear transformation on the feature matrix of the distribution network parameters, and the query feature matrix of the distribution network parameters can be expressed by the following formula (2):

[0079] (2),

[0080] Among them, Qi W represents the query feature matrix of the i-th distribution network parameter; Q Represents the weight matrix of the query feature matrix, which is a d-dimensional matrix, W Q The dimension of X is smaller than the dimension of the characteristic matrix of the distribution network parameters to improve the computational efficiency; i The characteristic matrix representing the parameters of the distribution network.

[0081] The key characteristic matrix of the distribution network parameters can be expressed as follows:

[0082] (3),

[0083] Among them, K i W represents the key characteristic matrix of the i-th distribution network parameters; K Represents the weight matrix of the key feature matrix, which is a d-dimensional matrix.

[0084] The characteristic matrix of the distribution network parameters can be expressed as follows:

[0085] (4),

[0086] Among them, V i represents the characteristic matrix of the value of the i-th distribution network parameter; W V Represents the weight matrix of the value feature matrix, which is a d-dimensional matrix.

[0087] In the second step, the attention weight between the determined distribution network parameters and other distribution network parameters can be expressed as follows:

[0088] (5),

[0089] Among them, A ij represents the attention weight between the i-th distribution network parameter and the j-th distribution network parameter; SoftMax is a normalized exponential function; T is the time step of the preset time.

[0090] According to the feature matrix of attention weights and distribution network parameters, the fusion feature matrix of the distribution network is determined, including:

[0091] First, for each type of distribution network parameter, the initial fusion feature of the distribution network parameter is determined according to each attention weight between the distribution network parameter and other distribution network parameters, and the value feature matrix of each other distribution network parameter; wherein the initial fusion feature of the distribution network parameter represents the fusion parameter feature between the distribution network parameter and each other distribution network parameter;

[0092] Second, the initial fusion features of each distribution network parameter are spliced ​​to obtain the fusion feature matrix of the distribution network.

[0093] Specifically, the initial fusion characteristics of distribution network parameters can be expressed as follows:

[0094] (6),

[0095] Among them, F i represents the initial fusion features of the i-th distribution network parameter; M represents the total number of distribution parameter types.

[0096] The fusion characteristic matrix of the distribution network can be expressed as follows:

[0097] (7),

[0098] Among them, Z represents the fusion feature matrix; Concat represents the feature concatenation function; W O Represents the linear transformation matrix of the fused feature matrix.

[0099] In these embodiments, the distribution network parameters are processed based on the self-attention mechanism, and the resulting fused feature matrix not only retains the key information of each distribution network parameter, but also increases the interaction information between different distribution network parameters, thereby improving the accuracy of the feature data in describing the distribution network.

[0100] Regarding the training process of the preset recognition model in step S104, in some of these embodiments, the method further includes:

[0101] First, obtain a first set of parameters to be trained and a second set of parameters to be trained for the distribution network in each scenario; wherein the first set of parameters to be trained includes multiple sets of fused data, and the second set of parameters to be trained includes multiple sets of fused data; the fused data represents the fused data of the multimodal data.

[0102] Second, repeatedly perform the following steps until the preset conditions are met: randomly determine multiple scenarios, train the initial model according to the first set of parameters to be trained in each scenario, and obtain a trained initial model; train the trained initial model according to the second set of parameters to be trained in each scenario, and obtain a re-trained initial model; and use the re-trained initial model as the initial model.

[0103] Among them, the initial model after retraining obtained when the preset conditions are met is the preset recognition model.

[0104] It should be noted that the preset recognition model may be a model based on a meta-learning framework. The model based on the meta-learning framework defines multiple tasks for model training. In the process of modeling the power distribution network, a scene of the power distribution network may correspond to a task in the preset recognition model.

[0105] For multiple scenarios of the power distribution network, the information of the power distribution network in different scenarios includes at least one of the following: load status, environmental parameters, and grounding points. That is, different scenarios of the power distribution network have different load status, or environmental parameters, or grounding points.

[0106] In the first step, the fused data may include a characteristic matrix of distribution network parameters and electric shock fault identification results of the distribution network.

[0107] In the second step, when training the model, the input data is a fused feature matrix obtained by feature fusion of the feature matrix of the distribution network parameters in the fused data; the output is the predicted electric shock fault identification result of the distribution network; the electric shock fault identification result of the distribution network in the fused data is the true value.

[0108] As an example, the preset condition in the second step may be that the number of iterations of model training reaches a preset number of iterations.

[0109] As an example, the preset condition in the second step may also be receiving a stop instruction.

[0110] Specifically, according to the first set of parameters to be trained in each scenario, the initial model is trained to obtain a trained initial model; according to the second set of parameters to be trained in each scenario, the trained initial model is trained to obtain a retrained initial model, including:

[0111] First, multiple fused data in the first parameter set to be trained in each scenario are used to train the initial model, and when it is determined that the value of the obtained first loss function is the minimum, the trained initial model is obtained.

[0112] Second, the trained initial model is trained according to the multiple fused data in the second parameter set to be trained in each scenario, and when it is determined that the value of the obtained second loss function is the minimum, the re-trained initial model is obtained.

[0113] It should be noted that in the first step, multiple fusion features in the first set of parameters to be trained in each scenario are used to train the initial model, and the first set of parameters to be trained in each scenario is used to train the initial model to update the model parameters corresponding to the scenario in the model.

[0114] When the initial model is trained using multiple fused data in the first parameter set to be trained in each scenario, the value of the first loss function used is used to characterize the error between the true value and the predicted value, and the value of the first loss function is minimized through training. The first loss function can be expressed by the following formula (8):

[0115] (8),

[0116] in, represents a randomly determined scene n; represents the loss function value corresponding to scenario n; z is the fusion feature matrix of distribution network parameters; y is the electric shock fault identification result of the distribution network predicted by function f; is the model parameter corresponding to scene n; T test is the number of training parameters in the first parameter set to be trained.

[0117] For each scene, the model parameters corresponding to the scene in the model are updated by gradient descent. The parameters can be updated using the following formula (9):

[0118] (9),

[0119] in, Indicates the updated model parameters corresponding to the scene; Represents the learning rate in this scenario, which is used to control the step size of gradient descent.

[0120] In the second step, the trained initial model is trained based on the multiple fused data in the second parameter set to be trained under each scenario to update the model parameters corresponding to each scenario in the model. The second loss function used in the training can be expressed by the following formula (10):

[0121] (10),

[0122] in, A total loss function value of multiple second parameter sets to be trained representing multiple scenarios; Represents multiple randomly determined scenes.

[0123] By updating the model parameters corresponding to each scene in the model through stochastic gradient descent, the model parameters corresponding to each scene can be updated using the following formula (11):

[0124] (11),

[0125] in, Indicates the model parameters corresponding to each scenario; Represents the learning rate between multiple scenes.

[0126] In these embodiments, the model is trained by using the first set of parameters to be trained for any scenario to optimize the model parameters corresponding to the scenario; the model parameters corresponding to each scenario are optimized using the second set of parameters to be trained for each scenario, and the preset recognition model obtained by training can accurately identify the fault type of the distribution network and has the ability to quickly adapt to new scenarios.

[0127] In some of these embodiments, due to the scarcity of data samples of electric shock events in the power distribution network, the sample data of electric shock events in the power distribution network can be enriched by expanding the data set. Specifically, obtaining a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario includes:

[0128] First, obtain the first historical fusion data of the distribution network in each scenario.

[0129] Second, use a pre-trained generator to perform data expansion on the first historical fusion data in the scene to obtain multiple first simulated fusion data in the scene; merge the multiple first simulated fusion data in the scene and the historical fusion data in the scene to obtain each fusion data in the scene.

[0130] Third, each fusion data in the scene is split and processed to obtain a first set of parameters to be trained and a second set of parameters to be trained in the scene.

[0131] It should be noted that when using the generator to generate simulation data, the simulation fusion data can be generated based on the historical fusion data, by changing the scene information corresponding to the historical fusion data, or the electric shock fault identification result of the distribution network.

[0132] As an example, a historical fusion data is obtained, and based on the historical fusion data, a plurality of simulation fusion data are generated by changing the load state, or environmental parameters, or grounding points, or electric shock fault identification results corresponding to the historical fusion data.

[0133] The electric shock fault recognition model is trained using multiple historical fusion data and multiple simulated fusion data generated by the generator, which enriches the training samples and thus improves the recognition ability of the preset recognition model obtained by training.

[0134] Regarding the generator, the method also includes:

[0135] First, historical multimodal data of the distribution network in each scenario is obtained, wherein the historical multimodal data has an electric shock event type; and feature fusion processing is performed on the historical multimodal data to obtain first historical fusion data of the distribution network.

[0136] Second, based on the initial generator, data expansion processing is performed on the first historical fusion data to obtain second historical fusion data.

[0137] Third, use the first historical fusion data and the second historical fusion data to train the initial discriminator and the initial generator to obtain a pre-trained generator corresponding to the initial generator.

[0138] It should be noted that the feature fusion processing of the historical multimodal data in the first step includes extracting the feature matrix of the historical multimodal data, and the electric shock event type of the historical multimodal data can be the electric shock fault recognition result. The feature matrix of the historical multimodal data and the electric shock fault recognition result of the historical multimodal data together constitute the first historical fusion data.

[0139] The generator in the second step can be the generator in the conditional generative adversarial network.

[0140] The discriminator may be a discriminator in a generative conditional generative adversarial network. The generator is used to generate simulated data based on random noise and generation conditions; the discriminator is used to determine whether the input data is real data or non-real data. The generation conditions may include the load state of the power distribution network, environmental parameters, grounding points, and electric shock fault identification results. The generator generates simulated data by changing at least one of the conditions.

[0141] The generator generates simulated data which can be expressed by the following formula (12):

[0142] (12),

[0143] Among them, s g represents the generated simulated fusion data; G represents the generator; q represents random noise; C represents the generation condition; Represents a generator parameter.

[0144] The discriminator determines whether the input data is real data or non-real data, which can be expressed by the following formula (13):

[0145] (13),

[0146] Where d represents the probability that the discriminator judges that the input data is real data; D represents the discriminator; s represents the input data, including real data and simulated data; represents the discriminator parameters.

[0147] In the third step, the initial discriminator and the initial generator are trained using the first historical fusion data and the second historical fusion data to obtain a pre-trained generator corresponding to the initial generator. The goal of the discriminator is to maximize the probability of real samples while minimizing the probability of generated samples. Based on this goal, the loss function used by the discriminator can be expressed as follows:

[0148] (14),

[0149] Among them, L D Represents the loss function value of the discriminator; Express expectation; r Represents real data; Represents the distribution of real data; Represents the distribution of noise.

[0150] The goal of the generator is to generate samples that can deceive the discriminator. Based on this goal, the loss function used by the generator can be expressed as follows:

[0151] (15),

[0152] Among them, L G Represents the loss function value of the generator.

[0153] The generator and discriminator are trained through the adversarial optimization problem expressed in the following formula (16):

[0154] (16),

[0155] Among them, min represents the minimization function; max represents the maximization function.

[0156] In these embodiments, a training generator is used to generate simulated fusion data of multiple load states, environmental parameters, grounding points, and electric shock fault identification results of different distribution networks for training the distribution network electric shock fault identification model, thereby improving the model recognition capability of the distribution network electric shock fault identification model.

[0157] Figure 2 The structural diagram of the electric shock fault identification device for the power distribution network provided by the present application is as follows: Figure 2 As shown, the electric shock fault identification device 20 for the power distribution network provided in this embodiment includes:

[0158] The acquisition module 201 is used to acquire multimodal data within a preset time period from the power distribution network, wherein the multimodal data includes various types of power distribution network parameters of the power distribution network; and perform matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters.

[0159] The first determination module 202 is used to determine the attention weight between the power distribution network parameter and other power distribution network parameters according to the characteristic matrix of the power distribution network parameter; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters.

[0160] The second determination module 203 is used to determine the fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters.

[0161] The identification module 204 is used to input the fusion feature matrix of the power distribution network into a preset identification model for identification, so as to obtain the electric shock fault identification result of the power distribution network.

[0162] In some specific implementations, the first determination module 202 is further used to use different weight matrices to perform linear transformation processing on the characteristic matrix of the distribution network parameters for each type of distribution network parameters, so as to obtain a query characteristic matrix, a key characteristic matrix and a value characteristic matrix of the distribution network parameters; wherein the query characteristic matrix, the key characteristic matrix and the value characteristic matrix are characteristic matrices of the distribution network parameters after different linear transformations;

[0163] For each type of distribution network parameter, the attention weights between the distribution network parameter and other distribution network parameters are determined according to the query feature matrix of the distribution network parameter and the key feature matrix of other distribution network parameters.

[0164] In some embodiments of these embodiments, the second determination module 203 is further used to determine, for each type of power distribution network parameter, the initial fusion feature of the power distribution network parameter according to each attention weight between the power distribution network parameter and other power distribution network parameters, and the value feature matrix of each other power distribution network parameter; wherein the initial fusion feature of the power distribution network parameter represents the fusion parameter feature between the power distribution network parameter and each other power distribution network parameter;

[0165] The initial fusion features of each distribution network parameter are spliced ​​to obtain the fusion feature matrix of the distribution network.

[0166] In some of these embodiments, the multimodal data includes: a residual current signal of the power distribution network, a voltage fluctuation of the power distribution network, a load state of the power distribution network, and environmental parameters of the power distribution network.

[0167] In some embodiments of these embodiments, the device further includes a training module for obtaining a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario; wherein the first set of parameters to be trained includes a plurality of fused data, and the second set of parameters to be trained includes a plurality of fused data; the fused data represents fused data of the multimodal data;

[0168] Repeat the following steps until the preset conditions are met: randomly determine multiple scenarios, train the initial model according to the first set of parameters to be trained in each scenario, and obtain a trained initial model; train the trained initial model according to the second set of parameters to be trained in each scenario, and obtain a retrained initial model; and use the retrained initial model as the initial model;

[0169] Among them, the initial model after retraining obtained when the preset conditions are met is the preset recognition model.

[0170] In some of these embodiments, the training module is further configured to obtain first historical fusion data of the power distribution network in each scenario;

[0171] Using a pre-trained generator to perform data expansion on the first historical fusion data in the scene to obtain a plurality of first simulated fusion data in the scene; merging the plurality of first simulated fusion data in the scene and the historical fusion data in the scene to obtain each fusion data in the scene;

[0172] Each fusion data in the scene is split and processed to obtain a first set of parameters to be trained and a second set of parameters to be trained in the scene.

[0173] In some embodiments of these embodiments, the training module is further used to obtain historical multimodal data of the power distribution network in each scenario, wherein the historical multimodal data has an electric shock event type; and perform feature fusion processing on the historical multimodal data to obtain first historical fusion data of the power distribution network;

[0174] Performing data expansion processing on the first historical fusion data based on the initial generator to obtain second historical fusion data;

[0175] The initial discriminator and the initial generator are trained using the first historical fusion data and the second historical fusion data to obtain a pre-trained generator corresponding to the initial generator.

[0176] In some implementations of these implementations, the training module is further used to train the initial model using multiple fused data in the first parameter set to be trained in each scenario, and obtain the trained initial model when it is determined that the value of the obtained first loss function is the minimum;

[0177] The trained initial model is trained according to the multiple fused data in the second parameter set to be trained in each scenario, and when it is determined that the value of the obtained second loss function is minimum, the re-trained initial model is obtained.

[0178] In some of these embodiments, the information of the power distribution network in different scenarios includes at least one of the following: load status, environmental parameters, and grounding point.

[0179] The electric shock fault identification device for the power distribution network provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be described in detail here.

[0180] Figure 3 This is a schematic diagram of the structure of the electric shock fault identification device for the power distribution network provided by this application. Figure 3 As shown, the electric shock fault identification device 30 for the power distribution network provided in this embodiment includes: at least one processor 301 and a memory 302. Optionally, the device 30 also includes a communication component 303. The processor 301, the memory 302 and the communication component 303 are connected via a bus 304.

[0181] In a specific implementation process, at least one processor 301 executes the computer-executable instructions stored in the memory 302, so that at least one processor 301 executes the above method.

[0182] The specific implementation process of the processor 301 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0183] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0184] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0185] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0186] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0187] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0188] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0189] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0190] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0191] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0192] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0193] If the 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 invention, 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. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0194] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0195] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for identifying electric shock faults in a power distribution network, the method comprising: Acquire multimodal data within a preset time period from a power distribution network, wherein the multimodal data includes multiple types of power distribution network parameters of the power distribution network; and perform matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters; Determine, according to the characteristic matrix of the power distribution network parameter, an attention weight between the power distribution network parameter and other power distribution network parameters; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters; Determining a fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters; The fused feature matrix of the power distribution network is input into a preset recognition model for recognition, so as to obtain an electric shock fault recognition result of the power distribution network.

2. The method according to claim 1, characterized in that The determining, according to the characteristic matrix of the power distribution network parameter, the attention weight between the power distribution network parameter and other power distribution network parameters comprises: For each type of distribution network parameter, a characteristic matrix of the distribution network parameter is linearly transformed using different weight matrices to obtain a query characteristic matrix, a key characteristic matrix and a value characteristic matrix of the distribution network parameter; wherein the query characteristic matrix, the key characteristic matrix and the value characteristic matrix are characteristic matrices of the distribution network parameter after different linear transformations; For each type of power distribution network parameter, the attention weight between the power distribution network parameter and other power distribution network parameters is determined according to the query feature matrix of the power distribution network parameter and the key feature matrix of other power distribution network parameters.

3. The method according to claim 1, characterized in that The step of determining a fusion feature matrix of the power distribution network according to the attention weights and the feature matrix of the power distribution network parameters includes: For each type of power distribution network parameter, the initial fusion feature of the power distribution network parameter is determined according to each attention weight between the power distribution network parameter and other power distribution network parameters, and the value feature matrix of each of the other power distribution network parameters; wherein the initial fusion feature of the power distribution network parameter represents the fusion parameter feature between the power distribution network parameter and each of the other power distribution network parameters; The initial fusion features of the distribution network parameters are concatenated to obtain a fusion feature matrix of the distribution network.

4. The method according to claim 1, characterized in that The multimodal data includes: Residual current signal of the distribution network, voltage fluctuation of the distribution network, load status of the distribution network, and environmental parameters of the distribution network.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Obtaining a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario; wherein the first set of parameters to be trained includes a plurality of fused data, and the second set of parameters to be trained includes a plurality of fused data; the fused data represents fused data of multimodal data; Repeat the following steps until a preset condition is met: randomly determine multiple scenarios, train the initial model according to the first set of parameters to be trained in each scenario, and obtain a trained initial model; train the trained initial model according to the second set of parameters to be trained in each scenario, and obtain a retrained initial model; and use the retrained initial model as the initial model; Among them, the initial model after retraining obtained when the preset condition is met is the preset recognition model.

6. The method according to claim 5, characterized in that Acquiring a first set of parameters to be trained and a second set of parameters to be trained for the power distribution network in each scenario, including: Acquire first historical fusion data of the power distribution network in each scenario; Using a pre-trained generator to perform data expansion on the first historical fusion data in the scenario to obtain a plurality of first simulated fusion data in the scenario; merging the plurality of first simulated fusion data in the scenario with the historical fusion data in the scenario to obtain each fusion data in the scenario; Each fusion data in the scenario is split and processed to obtain a first set of parameters to be trained and a second set of parameters to be trained in the scenario.

7. The method according to claim 6, characterized in that The method further comprises: Acquire historical multimodal data of the power distribution network in each scenario, wherein the historical multimodal data has an electric shock event type; and perform feature fusion processing on the historical multimodal data to obtain first historical fusion data of the power distribution network; Performing data expansion processing on the first historical fusion data based on the initial generator to obtain second historical fusion data; The initial discriminator and the initial generator are trained using the first historical fusion data and the second historical fusion data to obtain the pre-trained generator corresponding to the initial generator.

8. The method according to claim 5, characterized in that The initial model is trained according to the first set of parameters to be trained in each scenario to obtain a trained initial model; According to the second set of parameters to be trained in each scenario, the trained initial model is trained to obtain a retrained initial model, including: Using a plurality of fused data in the first parameter set to be trained in each scenario, the initial model is trained, and when it is determined that the value of the obtained first loss function is minimum, the trained initial model is obtained; The trained initial model is trained according to the multiple fused data in the second parameter set to be trained in each scenario, and when it is determined that the value of the obtained second loss function is minimum, the re-trained initial model is obtained.

9. The method according to claim 5, characterized in that The information of the power distribution network in different scenarios includes at least one of the following: load status, environmental parameters, and grounding point.

10. A device for identifying electric shock faults in a power distribution network, characterized in that: include: An acquisition module, configured to acquire multimodal data within a preset time period from a power distribution network, wherein the multimodal data includes multiple types of power distribution network parameters of the power distribution network; and perform matrix conversion on the power distribution network parameters to obtain a characteristic matrix of the power distribution network parameters; A first determination module is used to determine the attention weight between the power distribution network parameter and other power distribution network parameters according to the characteristic matrix of the power distribution network parameter; wherein the attention weight represents the correlation between the power distribution network parameter and other power distribution network parameters; A second determination module, configured to determine a fusion characteristic matrix of the power distribution network according to the attention weight and the characteristic matrix of the power distribution network parameters; The identification module is used to input the fusion feature matrix of the distribution network into a preset identification model for identification, so as to obtain the electric shock fault identification result of the distribution network.

11. An electric shock fault identification device for a power distribution network, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 9 when executed by a processor.

13. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 9 when being executed by a processor.

Images