Fault positioning model training method and device based on convolutional neural network, equipment and storage medium

Abstract

The invention discloses a fault positioning model training method and device based on a convolutional neural network, equipment and a storage medium. The method comprises the following steps: collecting current waveform data of a power distribution network through a magnetoelectric sensor; extracting transient characteristics in the current waveform data based on wavelet transform, and constructing a training set and a test set based on the transient characteristics; training a convolutional neural network based on the training set in combination with a normalized classifier to obtain a trained fault positioning model; determining the fault positioning accuracy of the fault positioning model based on the test set; and when the fault positioning accuracy meets an accuracy threshold, taking the fault positioning model as a target fault positioning model. According to the method, the data of the power distribution network are collected, and the model is trained, so that the fault positioning model with high fault positioning accuracy can be obtained.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of power distribution network fault handling technology, and in particular to a fault location model training method, device, equipment and storage medium based on convolutional neural network. Background Technology

[0002] With the advancement of dual-carbon goals, the construction of new power systems based on new energy sources is accelerating, and the structure of distribution networks is becoming increasingly complex. In addition, the characteristics of distribution networks, such as numerous feeders, small power supply range, and disordered switch distribution, will bring a certain impact to distribution network faults after the large-scale access of distributed power sources. This leads to a decrease in the accuracy of traditional methods in locating distribution network faults, making it difficult to meet the needs of fault handling.

[0003] To address this problem, researchers both domestically and internationally have conducted extensive studies on fault location in distribution networks. The main research methods include: impedance method, traveling wave method, S-signal injection method, and wide-area communication method. The impedance method calculates the line impedance using the voltage and current at the fault point, thus determining the fault location. However, this method has significant limitations, being affected by line branches, transition resistance, and other conditions, and is prone to false faults. The traveling wave method calculates the fault location based on the time it takes for voltage and current signals to reach the fault point. However, due to the complex structure of distribution networks and the numerous installation locations, this method struggles to extract useful information and requires substantial investment. The S-signal injection method determines the fault location by injecting a specific signal and analyzing its path. However, this method is susceptible to the influence of transition resistance and distributed capacitance, and suffers from low location efficiency and poor reliability. The wide-area communication method utilizes multi-point measurement information to determine the fault location, requiring fault indicators and other distribution equipment. However, this method has high communication requirements and is highly dependent on distribution equipment, making it difficult to simultaneously meet both reliability and sensitivity requirements.

[0004] Therefore, traditional fault location techniques are difficult to accurately locate faults in distribution networks. In particular, their applicability is significantly reduced in scenarios where a large number of distributed power sources are connected, resulting in a complex distribution network structure, and they can no longer meet the needs of practical applications. Summary of the Invention

[0005] This invention provides a fault location model training method, apparatus, device, and storage medium based on convolutional neural networks to solve the problem that existing methods cannot accurately locate faults in scenarios with complex power distribution network structures.

[0006] According to one aspect of the present invention, a method for training a fault localization model based on a convolutional neural network is provided, the method comprising:

[0007] Current waveform data of the power distribution network is collected using magnetoelectric sensors;

[0008] Transient features are extracted from the current waveform data based on wavelet transform, and training and testing sets are constructed based on the transient features.

[0009] Based on the training set and a normalized classifier, the convolutional neural network is trained to obtain a trained fault localization model.

[0010] The fault location accuracy of the fault location model is determined based on the test set.

[0011] When the fault location accuracy meets the accuracy threshold, the fault location model is used as the target fault location model.

[0012] According to another aspect of the present invention, a fault location model training device based on a convolutional neural network is provided, the device comprising:

[0013] The data acquisition module is used to acquire current waveform data of the power distribution network through a magnetoelectric sensor;

[0014] The extraction module is used to extract transient features from the current waveform data based on wavelet transform, and to construct training and testing sets based on the transient features;

[0015] The training module is used to train the convolutional neural network based on the training set and a normalized classifier to obtain a trained fault localization model.

[0016] The determination module is used to determine the fault location accuracy of the fault location model based on the test set;

[0017] The judgment module is used to select the fault location model as the target fault location model when the fault location accuracy meets the accuracy threshold.

[0018] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: at least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the fault location model training method based on convolutional neural networks as described in any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the fault location model training method based on a convolutional neural network as described in any embodiment of the present invention.

[0022] This invention discloses a method, apparatus, device, and storage medium for training a fault location model based on a convolutional neural network. The method includes: acquiring current waveform data of a power distribution network using a magnetoelectric sensor; extracting transient features from the current waveform data based on wavelet transform; constructing a training set and a test set based on the transient features; training the convolutional neural network using the training set and a normalized classifier to obtain a trained fault location model; determining the fault location accuracy of the fault location model based on the test set; and using the fault location accuracy as the target fault location model when it meets an accuracy threshold. This method, by acquiring data from the power distribution network and training the model, can obtain a fault location model with high fault location accuracy, solving the problem that existing methods cannot accurately locate faults in scenarios with complex power distribution network structures.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a fault location model training method based on a convolutional neural network, provided in Embodiment 1 of the present invention.

[0026] Figure 2 A flowchart illustrating a fault location model training method based on a convolutional neural network, provided in an embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a lightweight structure provided in an embodiment of the present invention;

[0028] Figure 4 A schematic diagram of a power distribution network structure is provided for an embodiment of the present invention;

[0029] Figure 5 A comparison chart of the accuracy of different fault location methods provided in an embodiment of the present invention;

[0030] Figure 6 This is a schematic diagram of the structure of a fault location model training device based on a convolutional neural network provided in Embodiment 2 of the present invention;

[0031] Figure 7 This is a schematic diagram of the structure of an electronic device using a fault location model training method based on a convolutional neural network, as described in an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention. It should be understood that the various steps described in the method embodiments of the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0033] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, any variations of the terms "comprising" and "having," etc., are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0036] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0037] Example 1

[0038] Figure 1 This is a flowchart illustrating a fault location model training method based on a convolutional neural network, as provided in Embodiment 1 of the present invention. This method is applicable to training models capable of locating faults in a power distribution network. The method can be executed by a fault location model training device based on a convolutional neural network, wherein the device can be implemented by software and / or hardware and is generally integrated on an electronic device. In this embodiment, the electronic device includes, but is not limited to, devices such as computers.

[0039] like Figure 1 As shown in Embodiment 1 of the present invention, a fault location model training method based on a convolutional neural network includes the following steps:

[0040] S110. Acquire current waveform data of the power distribution network through a magnetoelectric sensor.

[0041] Among them, magnetoelectric sensors can be energy conversion sensors that convert the measured physical quantity into an electrical signal based on the magnetoelectric coupling effect. Current waveform data can refer to continuous / discrete signals that reflect the change of current in a circuit over time.

[0042] In this embodiment, current waveform data in the power distribution network can be collected by a magnetoelectric sensor. When a fault occurs (such as single-phase grounding or short circuit), a non-sinusoidal transient zero-sequence current will be generated in the line. The magnetoelectric sensor can sense the alternating magnetic field generated by the current, convert the magnetic field change into a corresponding voltage signal, and thus obtain the complete transient current waveform.

[0043] Specifically, the sampling rate of the current data collected by the magnetoelectric sensor can be 12.8k. The current waveform data measured by the magnetoelectric sensor is a high sampling rate discrete sequence with time stamp. Based on the time stamp information, the current value of all nodes in the distribution network at a certain moment can be extracted. Then, transient features of the current waveform data of each node are extracted and the transient features of all nodes are integrated into a feature set for the identification of fault status in the distribution network.

[0044] S120. Extract transient features from the current waveform data based on wavelet transform, and construct training and testing sets based on the transient features.

[0045] Among them, wavelet transform can be a mathematical tool for simultaneously analyzing the time-frequency characteristics of a signal, while transient characteristics can refer to the sudden changes and non-stationary characteristics of current or voltage signals deviating from steady state at the moment a fault occurs.

[0046] In this embodiment, transient features in current waveform data can be extracted based on wavelet transform, and training and testing sets can be constructed based on the transient features.

[0047] In one embodiment, constructing the training set and test set based on the transient features includes: using the transient zero-sequence current wavelet decomposition coefficient features in the transient features as input features; normalizing the input features to obtain processed input features; and dividing the processed input features to obtain the training set and test set.

[0048] Among them, the transient zero-sequence current wavelet decomposition coefficient characteristics can refer to the set of approximate coefficients and detail coefficients at different scales obtained by decomposing the zero-sequence current waveform signal collected during the transient process of distribution network faults through discrete wavelet transform.

[0049] In this embodiment, the transient zero-sequence current wavelet decomposition coefficients in the transient features can be used as input features, and the input features can be normalized to obtain processed input features. The processed input features can be divided to obtain training and test sets.

[0050] For example, a high-sampling-rate magnetoelectric sensor can be used to acquire current waveform data. Wavelet transform can be used to extract transient features of the waveform. The transient zero-sequence current wavelet decomposition coefficients are then used as input to a lightweight convolutional neural network. The input data is then normalized and divided into test and training sets.

[0051] The wavelet decomposition coefficient characteristics of transient zero-sequence current can be normalized according to the following principle:

[0052] ;

[0053] In the formula, For normalized sample data, The original data is represented by 'min', which is the minimum value function, and 'max' is the maximum value function.

[0054] In one embodiment, the formula for the wavelet transform includes:

[0055] The wavelet transform formula is:

[0056] ;

[0057] in, The result of wavelet transform, The original discrete current waveform data, For wavelet basis, To control the scaling parameters of the wavelet basis, To control the displacement parameters of the wavelet basis;

[0058] The first one obtained after discrete wavelet transform The layer coefficient is:

[0059] ;

[0060] in, These are approximate coefficients. For detail coefficients, and These are low-pass and high-pass filters, respectively.

[0061] In this embodiment, features in the current waveform data can be extracted using wavelet transform.

[0062] S130. Based on the training set and a normalized classifier, the convolutional neural network is trained to obtain a trained fault location model.

[0063] The normalization classifier can be a Softmax classifier. The convolutional neural network can be a lightweight convolutional neural network.

[0064] In this embodiment, a convolutional neural network can be trained using a training set, and the training effect can be determined using a normalization classifier, thereby obtaining a well-trained fault location model.

[0065] In one embodiment, training a convolutional neural network based on the training set and a normalized classifier to obtain a trained fault localization model includes: inputting the training set into the convolutional neural network to obtain output features; inputting the output features into a normalized classifier to obtain the error between the output features and the actual labels; when the error is lower than an error threshold or the number of iterations reaches the maximum number of iterations, using the current convolutional neural network as the trained fault localization model; otherwise, updating the network parameters of the convolutional neural network based on backpropagation and continuing to train the updated convolutional neural network.

[0066] The output features can be the fault characteristics of the distribution network lines. The actual labels can be the actual classification results corresponding to the input values ​​of the convolutional neural network. The error threshold and the maximum number of iterations can be set according to the actual situation, and this embodiment does not limit them.

[0067] In this embodiment, the training set can be input into the convolutional neural network to obtain output features. The error between the output features and the actual labels can be determined by a normalization classifier. If the error is lower than the error threshold or the number of iterations reaches the maximum number of iterations, the current convolutional neural network can be used as a trained fault localization model. Otherwise, the network parameters of the convolutional neural network are updated based on backpropagation, and the updated convolutional neural network is trained until a convolutional neural network that meets the conditions is obtained.

[0068] For example, Figure 2 A flowchart illustrating a fault location model training method based on a convolutional neural network, as provided in an embodiment of the present invention, is shown below. Figure 2As shown, in this embodiment, the training set can be input into a lightweight convolutional neural network. The fault characteristics of the power distribution line are learned in a deep manner through forward propagation, the network parameters are updated through back propagation, and the learned features are input into a Softmax classifier for classification. The output is the fault section, thereby realizing fault location. By continuously iterating and optimizing the network parameters, a network model with better performance can be formed in the end.

[0069] In one embodiment, the convolutional neural network includes a lightweight structure, pooling layers, and fully connected layers. Inputting the training set into the convolutional neural network to obtain output features includes: inputting the training set into the lightweight structure to obtain a feature map; the lightweight structure includes convolution, identity mapping, and linear operations; inputting the feature map into the pooling layer to perform pooling loop operations to obtain pooled features; and inputting the pooled features into the fully connected layer to obtain output features.

[0070] The lightweight structure can include convolution, identity mapping and linear operations, and the convolution can be a small number of convolutions.

[0071] In this embodiment, the training set can be input into a lightweight structure to obtain a feature map, the feature map can be input into a pooling layer to perform pooling loop operations to obtain pooled features, and the pooled features can be input into a fully connected layer to obtain output features.

[0072] Specifically, convolutional neural networks are mostly used in image recognition. They are a typical feedforward neural network; during forward propagation, the input data undergoes convolution and pooling loop operations. When starting from the first... To the During layer convolution, the following operations are performed:

[0073] ;

[0074] in, for The convolution kernels used for image transmission in each layer cover all regions of the image during convolution operations. Each covered region is multiplied by the kernel and summed to form a new image for the next layer. For bias, This is the activation function.

[0075] When the activation function is the Sigmoid function:

[0076] ;

[0077] When the activation function is a rectified linear unit (ReLU) function:

[0078] ;

[0079] To prevent overfitting due to excessively large image features after convolution, a pooling operation is performed after convolution to reduce the image size. Pooling typically includes average pooling and max pooling. After convolution and pooling, the convolutional neural network finally passes through a fully connected layer to output feature values. When training the network parameters, the goal can be to minimize the loss function, as shown below:

[0080] ;

[0081] in, For the number of categories, For the actual reality category, i.e., the ideal output, For network output.

[0082] In one embodiment, inputting the training set into a lightweight structure to obtain a feature map includes: inputting the training set into a convolutional layer, performing convolution operations on the features in the training set to obtain a core feature map; processing the core feature map through linear operations to obtain a derived feature map; and concatenating the core feature map and the derived feature map through an identity mapping to obtain a final feature map.

[0083] The core feature map can be a low-dimensional, high-value feature map obtained by selectively extracting input features using a small number of convolutional units in a lightweight convolutional neural network. The derived feature map (Ghost feature map) can be a supplementary feature map generated based on the core feature map through low-cost linear expansion operations.

[0084] In this embodiment, the training set can be input into a convolutional layer to perform convolution operations on the features in the training set, resulting in a core feature map. This core feature map is then processed using linear operations to obtain a derived feature map. Finally, the core feature map and the derived feature map are concatenated using an identity mapping to obtain the final feature map. For example, Figure 3 A schematic diagram of a lightweight structure provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the main idea of ​​the lightweight convolutional network is to divide the convolution into two parts based on the relationships between feature maps, including three parts: a small number of convolutions, identity mapping, and linear operations. First, wavelet transform is used to extract transient features of the waveform. The wavelet decomposition coefficients of the transient zero-sequence current are used as input to the lightweight convolutional neural network. After general convolution, a small number of feature maps (i.e., core feature maps) are output. Then, the feature maps undergo linear transformation to form Ghost feature maps. Finally, the feature maps are concatenated for output. The difference between general convolution and a small number of convolutions can be compared:

[0085] Input features Dimensions The output of a typical convolution Dimensions Then output The formula is as follows:

[0086] ;

[0087] Where, the dimension of f is The kernel size is The number of convolution kernels is , For bias.

[0088] The size of a typical convolution is expressed as follows: The lightweight convolution formula is shown below:

[0089] ;

[0090] ;

[0091] in, For a small number of convolutional outputs, the dimension is , is input Feature maps formed by general convolution For the first step of the general convolution, the first... The feature map uses the first feature map The parameters of a linear transformation, This is an operation for feature identity mapping. Each feature is linearly transformed and output. There are a total of feature maps, with a total of Each feature map.

[0092] Assuming the size of each convolutional kernel is The speed ratio of lightweight convolution to regular convolution is as follows:

[0093] ;

[0094] Therefore, it can be concluded that the computational cost of lightweight convolutional networks is approximately reduced compared to regular convolutions. The fact that the speed is twice that of the previous data indicates that lightweight convolutional neural networks are faster and can significantly shorten the fault location time in power distribution networks.

[0095] S140. Determine the fault location accuracy of the fault location model based on the test set.

[0096] In this embodiment, the fault location model can be tested using a test set to obtain the fault location accuracy of the model for faults in the distribution network, thereby verifying the effectiveness of the model.

[0097] S150. When the fault location accuracy meets the accuracy threshold, the fault location model is used as the target fault location model.

[0098] The accuracy threshold can be set according to the actual situation, and this embodiment does not limit it.

[0099] In this embodiment, if the fault location accuracy meets the accuracy threshold, the fault location model can be used as the target fault location model; otherwise, the fault location model can continue to be trained.

[0100] This invention provides a fault location model training method based on a convolutional neural network, comprising: collecting current waveform data of a power distribution network using a magnetoelectric sensor; extracting transient features from the current waveform data based on wavelet transform; constructing a training set and a test set based on the transient features; training the convolutional neural network using the training set and a normalization classifier to obtain a trained fault location model; determining the fault location accuracy of the fault location model based on the test set; and using the fault location accuracy to obtain a target fault location model when the accuracy meets an accuracy threshold. This method, by collecting data from the power distribution network and training the model, can obtain a fault location model with high fault location accuracy, solving the problem that existing methods cannot accurately locate faults in scenarios with complex power distribution network structures.

[0101] Based on the above embodiments, modified embodiments of the above embodiments are proposed. It should be noted that, in order to keep the description brief, only the differences from the above embodiments are described in the modified embodiments.

[0102] In one embodiment, the expression for the normalized classification is:

[0103] ;

[0104] in, For the first Each sample belongs to category The probability, The output value of the normalization function. For input values.

[0105] In this embodiment, the Softmax function is a typical function for multi-class classification problems. If the input... The output labels are ,So The probability of being classified as category j is Softmax outputs a A dimensional vector, and this The elements of the dimensional vector are between 0 and 1, and their sum is 1. The output of the lightweight convolutional neural network is fed into a Softmax classifier for classification. That is, the output of the classifier is the fault segment, and "1" and "0" can be used to represent whether the fault has occurred in the corresponding segment.

[0106] Specifically, in this embodiment, fault location is determined by the faulty section. This embodiment can use a typical 33-node model of a distribution network, which has 32 sections. A lightweight convolutional neural network can be used to construct a mapping model between the input current data and the output sections, resulting in an intuitive one-to-one relationship. The Softmax classifier locates the fault by outputting the faulty section of the distribution network. Therefore, the Softmax classifier outputs 32-dimensional data. When a fault occurs in the a-th section, the a-th dimension of the Softmax classifier is 1, and the other dimensions are 0.

[0107] Compared with traditional distribution network fault location technology, the method of this invention utilizes the advantages of lightweight convolutional neural networks, such as small parameters, fast computing speed, and high accuracy, to learn the fault characteristics of synchronous phasor currents collected by magnetoelectric sensors. It can quickly and accurately locate faults for different fault types and different transition resistances, solving the problem that traditional fault location technology cannot be applied to complex distribution networks with a large number of distributed power sources.

[0108] Based on the technical solutions of the above embodiments, this invention provides several specific implementation methods.

[0109] As a specific implementation method of this embodiment, the following example is used to verify the effect of the present invention:

[0110] This embodiment allows for the construction of an IEEE 33-node distribution network model within Power Systems Computer Aided Design (PSCAD). Figure 4 A schematic diagram of a power distribution network structure is provided for an embodiment of the present invention, such as... Figure 4 As shown, magnetoelectric sensors can be installed at nodes 14, 22, and 26 to collect and measure zero-sequence synchronous phasor current data, and distributed power input can be provided from nodes 11, 18, 25, and 33.

[0111] Since there are 32 distribution network areas, the output dimension of the Softmax classifier is set to 32. The accuracy of the lightweight convolutional neural network training is calculated according to the following formula:

[0112] ;

[0113] in, For the accuracy of fault location. The number of samples tested; For the first The actual fault area of ​​each sample For the first Each sample network outputs a fault region, if and If they are equal, then ,otherwise .

[0114] Table 1 shows the simulation results of the lightweight convolutional neural network model. The results show that when the number of training iterations reaches 500, the model has a high accuracy in fault location.

[0115] Table 1 Test Results

[0116]

[0117] To verify that the method of this embodiment has a good localization effect on different transition resistors, experiments were conducted with different transition resistors and the number of iterations was set to 500. The results are shown in Table 2. The table shows that the method of this embodiment also has a good identification effect on high-resistance faults.

[0118] Table 2 Simulation results for different transition resistances

[0119]

[0120] Figure 5 This is a comparison chart of the accuracy of different fault location methods provided in an embodiment of the present invention, such as... Figure 5 As shown in the figure, in order to verify the superiority of the method proposed in this invention, the method of this invention is compared with the mainstream BP neural network, convolutional neural network and support vector machine. It can be seen from the figure that the method of this embodiment has a higher fault location accuracy.

[0121] Example 2

[0122] Figure 6 This is a schematic diagram of a fault location model training device based on a convolutional neural network provided in Embodiment 2 of the present invention. The device is applicable to training models that can locate faults in a power distribution network. The device can be implemented by software and / or hardware and is generally integrated into an electronic device.

[0123] like Figure 6 As shown, the device includes:

[0124] The acquisition module 210 is used to acquire current waveform data of the power distribution network through a magnetoelectric sensor;

[0125] Extraction module 220 is used to extract transient features from the current waveform data based on wavelet transform, and to construct training set and test set based on the transient features;

[0126] Training module 230 is used to train the convolutional neural network based on the training set and a normalization classifier to obtain a trained fault localization model.

[0127] The determination module 240 is used to determine the fault location accuracy of the fault location model based on the test set;

[0128] The judgment module 250 is used to use the fault location model as the target fault location model when the fault location accuracy meets the accuracy threshold.

[0129] This embodiment provides a fault location model training device based on a convolutional neural network, comprising: a data acquisition module for acquiring current waveform data of a power distribution network using a magnetoelectric sensor; an extraction module for extracting transient features from the current waveform data based on wavelet transform, and constructing a training set and a test set based on the transient features; a training module for training the convolutional neural network based on the training set and a normalized classifier to obtain a trained fault location model; a determination module for determining the fault location accuracy of the fault location model based on the test set; and a judgment module for selecting the fault location model as the target fault location model when the fault location accuracy meets an accuracy threshold. By acquiring data from the power distribution network and training the model, a fault location model with high fault location accuracy can be obtained, solving the problem that existing methods cannot accurately locate faults in scenarios with complex power distribution network structures.

[0130] Furthermore, the extraction module 220 includes:

[0131] The transient zero-sequence current wavelet decomposition coefficients in the transient features are used as input features;

[0132] The input features are normalized to obtain the processed input features;

[0133] The processed input features are divided to obtain a training set and a test set.

[0134] Furthermore, training module 230 includes:

[0135] The training set is input into a convolutional neural network to obtain the output features;

[0136] The output features are input into a normalized classifier to obtain the error between the output features and the actual labels;

[0137] When the error is lower than the error threshold or the number of iterations reaches the maximum number of iterations, the current convolutional neural network is used as the trained fault localization model; otherwise, the network parameters of the convolutional neural network are updated based on backpropagation, and the updated convolutional neural network is trained again.

[0138] Furthermore, the convolutional neural network includes a lightweight structure, pooling layers, and fully connected layers. The training set is input into the convolutional neural network to obtain output features, including:

[0139] The training set is input into a lightweight structure to obtain a feature map; the lightweight structure includes convolution, identity mapping, and linear operations.

[0140] The feature map is input into the pooling layer and a pooling loop operation is performed to obtain the pooled features.

[0141] The pooled features are input into a fully connected layer to obtain the output features.

[0142] Furthermore, the step of inputting the training set into a lightweight structure to obtain a feature map includes:

[0143] The training set is input into a convolutional layer, and the features in the training set are subjected to convolution operations to obtain the core feature map.

[0144] The core feature map is processed by linear operations to obtain a derived feature map;

[0145] The core feature map and the derived feature map are concatenated by identity mapping to obtain the final feature map.

[0146] Furthermore, the expression for the normalized classification is:

[0147] ;

[0148] in, For the first Each sample belongs to category The probability, The output value of the normalization function. For input values.

[0149] Furthermore, the formula for the wavelet transform includes:

[0150] The wavelet transform formula is:

[0151] ;

[0152] in, The result of wavelet transform, The original discrete current waveform data, For wavelet basis, To control the scaling parameters of the wavelet basis, To control the displacement parameters of the wavelet basis;

[0153] The first one obtained after discrete wavelet transform The layer coefficient is:

[0154] ;

[0155] in, These are approximate coefficients. For detail coefficients, and These are low-pass and high-pass filters, respectively.

[0156] The above-mentioned fault location model training device based on convolutional neural networks can execute the fault location model training method based on convolutional neural networks provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0157] Example 3

[0158] Figure 7 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0159] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0160] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0161] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a fault location model training method based on a convolutional neural network.

[0162] In some embodiments, the convolutional neural network-based fault location model training method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the convolutional neural network-based fault location model training method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the convolutional neural network-based fault location model training method by any other suitable means (e.g., by means of firmware).

[0163] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0164] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0165] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0166] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0167] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0168] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0169] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0170] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for training a fault location model based on a convolutional neural network, characterized in that, The method comprises: Collecting current waveform data of a power distribution network through a magneto-electric sensor; Extracting transient characteristics in the current waveform data based on wavelet transform, and constructing a training set and a test set based on the transient characteristics; Training a convolutional neural network based on the training set combined with a normalization classifier to obtain a trained fault location model; Determining a fault location accuracy of the fault location model based on the test set; When the fault location accuracy meets an accuracy threshold, regarding the fault location model as a target fault location model.

2. The method of claim 1, wherein, The constructing of the training set and the test set based on the transient characteristics comprises: Taking transient zero sequence current wavelet decomposition coefficient characteristics in the transient characteristics as input characteristics; Performing normalization processing on the input characteristics to obtain processed input characteristics; Dividing the processed input characteristics to obtain the training set and the test set.

3. The method of claim 1, wherein, The training of the convolutional neural network based on the training set combined with the normalization classifier to obtain the trained fault location model comprises: Inputting the training set into the convolutional neural network to obtain output characteristics; Inputting the output characteristics into the normalization classifier to obtain errors of the output characteristics and actual labels; When the errors are lower than an error threshold or the number of iterations reaches a maximum number of iterations, regarding the current convolutional neural network as the trained fault location model, otherwise updating network parameters of the convolutional neural network based on back propagation and continuing to train the updated convolutional neural network.

4. The method of claim 3, wherein, The convolutional neural network comprises a lightweight structure, a pooling layer and a full connection layer, and the inputting of the training set into the convolutional neural network to obtain the output characteristics comprises: Inputting the training set into the lightweight structure to obtain feature maps; the lightweight structure comprises convolution, identity mapping and linear operation; Inputting the feature maps into the pooling layer to perform pooling circulation operation to obtain pooled features; Inputting the pooled features into the full connection layer to obtain the output characteristics.

5. The method of claim 4, wherein, The inputting of the training set into the lightweight structure to obtain the feature maps comprises: Inputting the training set into convolution to perform convolution operation on characteristics in the training set to obtain core feature maps; Performing processing on the core feature maps through linear operation to obtain derived feature maps; Splicing the core feature maps and the derived feature maps through identity mapping to obtain final feature maps.

6. The method of claim 1, wherein, An expression of the normalization classification is: ； wherein, is the probability that the th sample belongs to the class , is the output value of the normalization function, is the input value.

7. The method of claim 1, wherein, A formula of the wavelet transform comprises: A wavelet change formula is: ； wherein, is a result of a wavelet transform, is original discrete current waveform data, is a wavelet basis, is a scale parameter controlling the wavelet basis, is a shift parameter controlling the wavelet basis; The first layer coefficients obtained after the discrete wavelet transformation are: The layer coefficients are: ； wherein is an approximation coefficient, is a detail coefficient, and are low-pass and high-pass filters, respectively. 8.A device for training a fault location model based on a convolutional neural network, characterized in that, The device comprises: A collection module configured to collect current waveform data of a power distribution network through a magneto-electric sensor; An extraction module configured to extract transient characteristics in the current waveform data based on wavelet transform, and construct a training set and a test set based on the transient characteristics; A training module configured to train a convolutional neural network based on the training set combined with a normalization classifier to obtain a trained fault location model; A determination module configured to determine a fault location accuracy of the fault location model based on the test set; A judgment module configured to, when the fault location accuracy meets an accuracy threshold, regard the fault location model as a target fault location model.

9. An electronic device, comprising: The device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the method for training a fault location model based on a convolutional neural network according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to implement the method for training a fault location model based on a convolutional neural network according to any one of claims 1-7 when executed.

Images