Fault judgment method, device and equipment of power distribution network feeder section and storage medium

Abstract

The application relates to a power distribution network feeder section fault judgment method, device and equipment and a storage medium. The method comprises the following steps: acquiring zero sequence currents at both ends of a target feeder section in a power distribution network, and performing normalization processing on the zero sequence currents at both ends of the target feeder section to obtain normalized first and second target currents; the first and second target currents are input into a pre-trained target neural network model for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; the first and second feature vectors are spliced through a feature fusion layer of the target neural network model to obtain a fusion feature vector; and the fusion feature vector is input into a preset linear classifier for classification and discrimination to determine a fault discrimination result of the target feeder section. The application can improve the accuracy of power distribution network fault section positioning, and reduce the influence of noise interference and signal transformation sensitivity.

Description

Technical Field

[0001] This application relates to the field of power grid fault identification technology, and in particular to a fault judgment method, device, equipment and storage medium for a distribution network feeder section. Background Technology

[0002] Currently, fault location in power distribution networks is becoming increasingly difficult, especially in resonant grounding systems. The compensation effect of the arc suppression coil causes the fault current amplitude to decrease significantly and its characteristics to become weak, making it difficult for traditional methods to effectively identify fault sections.

[0003] Traditional fault location methods mainly include methods based on original signals and methods based on signal transformation. Methods based on original signals directly utilize the difference between the original waveforms of zero-sequence voltage and zero-sequence current for location. However, due to the prevalence of electromagnetic interference and equipment noise in power distribution networks, the key characteristics of the current at both ends of the fault point are easily interfered with by noise, resulting in significant deviations in the location results. Methods based on signal transformation convert the time-domain signal to the frequency domain or time-frequency domain to enhance feature extraction. However, the strong transient nature of zero-sequence current affects feature extraction, causing the peak position of the transformed features to shift and the energy distribution to be uneven, resulting in poor location performance under different fault conditions. Summary of the Invention

[0004] To address or partially address the problems existing in related technologies, this application provides a method, apparatus, equipment, and storage medium for fault diagnosis of feeder sections in a distribution network, which can improve the accuracy of fault location in the distribution network and reduce the impact of noise interference and signal transformation sensitivity.

[0005] The first aspect of this application provides a fault judgment method for a feeder section of a distribution network, comprising: acquiring the zero-sequence current at both ends of a target feeder section in the distribution network, and normalizing the zero-sequence current at both ends of the target feeder section to obtain a normalized first target current and a second target current; inputting the first target current and the second target current into a first feature extraction network and a second feature extraction network of a pre-trained target neural network model respectively for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector; and inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault judgment result of the target feeder section.

[0006] In conjunction with the first aspect, in one possible implementation of the first aspect, the normalization of the first zero-sequence current and the second zero-sequence current includes: normalizing the first zero-sequence current and the second zero-sequence current using the following formula:

[0007]

[0008] in, The value of the current at the i-th sampling point in the waveform of the zero-sequence current; The dataset consists of the waveforms of the zero-sequence current; The minimum current value in the dataset consisting of the waveforms of the zero-sequence current; The maximum current value in the dataset consisting of the waveforms of the zero-sequence current; This represents the current value at the i-th sampling point in the normalized zero-sequence current waveform.

[0009] In conjunction with the first aspect, in one possible implementation of the first aspect, the training process of the target neural network model includes: acquiring historical fault data; the historical fault data includes fault zero-sequence currents collected under different operating conditions; labeling the fault zero-sequence current samples as upstream fault current samples and / or downstream fault current samples according to the positional relationship between the collection location of the fault zero-sequence current and the two ends of the fault feeder segment, and constructing a pre-training dataset based on the upstream fault current samples and / or downstream fault current samples; inputting the pre-training dataset into a preset convolutional neural network model, the preset convolutional neural network model including a classification output link; training the preset convolutional neural network model using the pre-training dataset until the classification output link can distinguish between upstream fault current samples and downstream fault current samples, thereby obtaining a target neural network model including the first feature extraction network and the second feature extraction network.

[0010] In conjunction with the first aspect, in one possible implementation of the first aspect, the step of concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector includes: the feature fusion layer performing a vector concatenation operation on the first feature vector and the second feature vector, connecting the first feature vector and the second feature vector end-to-end in a preset order to obtain the fused feature vector; wherein the dimension of the fused feature vector is the sum of the dimensions of the first feature vector and the second feature vector.

[0011] In conjunction with the first aspect, in one possible implementation of the first aspect, the step of inputting the first target current and the second target current into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model for feature extraction to obtain the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current includes: inputting the target current into the convolutional layer of the target neural network model, performing a sliding convolution operation on the adjustment vector through the convolutional layer to obtain the local transient features corresponding to the target current, and generating a first feature map corresponding to the local transient features; inputting the preliminary feature map into the pooling layer of the target neural network model, performing dimensionality reduction processing on the first feature map according to the max pooling strategy to obtain the second feature map; and performing a flattening operation on the second feature map to obtain the feature vector corresponding to the target current.

[0012] In conjunction with the first aspect, in one possible implementation of the first aspect, the step of inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault discrimination result of the target feeder segment includes: inputting the fused feature vector into the fully connected layer of the preset linear classifier, performing nonlinear mapping on the fused feature vector through the fully connected layer to obtain a target feature vector; inputting the target feature vector into the output layer of the preset linear classifier, processing the target feature vector through an activation function, and outputting the classification probability value corresponding to the target feature vector; the classification probability value includes a first probability value corresponding to the faulty segment and a second probability value corresponding to the normal segment.

[0013] In conjunction with the first aspect, in one possible implementation of the first aspect, the method further includes: if only one target feeder segment in the distribution network has a classification result of being a fault segment, then the feeder segment is determined as a target fault segment; if at least two target feeder segments in the distribution network have a classification result of being fault segments, then the first probability values ​​of the target feeder segments are compared, and the target feeder segment corresponding to the largest first probability value is determined as the target fault segment; if no target feeder segment in the distribution network has a classification result of being a fault segment, then a warning message is generated.

[0014] A second aspect of this application provides a fault detection device for a feeder section of a distribution network, comprising: an acquisition module for acquiring the zero-sequence current at both ends of a target feeder section in the distribution network and normalizing the zero-sequence current at both ends of the target feeder section to obtain a normalized first target current and a second target current; a processing module for inputting the first target current and the second target current into a first feature extraction network and a second feature extraction network of a pre-trained target neural network model for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; a fusion module for concatenating the first feature vector and the second feature vector through a feature fusion layer of the target neural network model to obtain a fused feature vector; and a classification module for inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault detection result of the target feeder section.

[0015] A third aspect of this application provides an electronic device, comprising:

[0016] Processor; and

[0017] A memory that stores executable code, which, when executed by the processor, causes the processor to perform the method described above.

[0018] A fourth aspect of this application provides a computer-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to perform the method described above.

[0019] The technical solution provided in this application may include the following beneficial effects:

[0020] This application discloses a fault judgment method and related apparatus for a distribution network feeder section, comprising: acquiring the zero-sequence current at both ends of a target feeder section in the distribution network, and normalizing the zero-sequence current at both ends of the target feeder section to obtain a normalized first target current and a second target current; inputting the first target current and the second target current into the first feature extraction network and the second feature extraction network of a pre-trained target neural network model respectively for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector; and inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault judgment result of the target feeder section. In this method, by acquiring the zero-sequence current at both ends of the target feeder section, normalizing it, extracting and fusing features using a pre-trained target neural network model, and finally classifying and discriminating through a linear classifier, the fault section can be accurately judged, thereby improving the accuracy of fault section location in the distribution network and reducing the impact of noise interference and signal transformation sensitivity. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart illustrating the fault judgment method for a distribution network feeder section as shown in the embodiments of this application;

[0023] Figure 2 This is a schematic diagram of the structure of a fault detection device for a distribution network feeder section shown in an embodiment of this application;

[0024] Figure 3 This is a schematic diagram of the structure of an electronic device shown in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0026] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.

[0028] Currently, fault location in power distribution networks is becoming increasingly difficult, especially in resonant grounding systems. The compensation effect of the arc suppression coil causes the fault current amplitude to decrease significantly and its characteristics to become weak, making it difficult for traditional methods to effectively identify fault sections.

[0029] Traditional fault location methods mainly include methods based on original signals and methods based on signal transformation. Methods based on original signals directly utilize the difference between the original waveforms of zero-sequence voltage and zero-sequence current for location. However, due to the prevalence of electromagnetic interference and equipment noise in power distribution networks, the key characteristics of the current at both ends of the fault point are easily interfered with by noise, resulting in significant deviations in the location results. Methods based on signal transformation convert the time-domain signal to the frequency domain or time-frequency domain to enhance feature extraction. However, the strong transient nature of zero-sequence current affects feature extraction, causing the peak position of the transformed features to shift and the energy distribution to be uneven, resulting in poor location performance under different fault conditions.

[0030] To address the aforementioned issues, this application provides a method for fault diagnosis in a distribution network feeder section, which can improve the accuracy of fault location in the distribution network and reduce the impact of noise interference and signal transformation sensitivity.

[0031] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0032] Figure 1 This is a flowchart illustrating the fault diagnosis method for a distribution network feeder section as shown in the embodiments of this application.

[0033] See Figure 1 A fault diagnosis method for a distribution network feeder section includes:

[0034] S110: Obtain the zero-sequence current at both ends of the target feeder section in the distribution network, and normalize the zero-sequence current at both ends of the target feeder section to obtain the normalized first target current and second target current.

[0035] Specifically, feeder terminal units (FTUs) can be set at both ends of the target feeder section. The zero-sequence current at both ends of the target feeder section can be collected through the feeder terminal units. Zero-sequence current refers to the three-phase current vector sum being zero under ideal conditions. When a single-phase ground fault occurs, the three-phase currents lose balance, and their vector sum is no longer zero. The current component generated at this time is the zero-sequence current. After acquiring the zero-sequence current, it is normalized. The current value of each sampling point can be divided by the maximum current value in the waveform, thereby mapping all current values ​​to the interval [0,1] to obtain the first target current and the second target current.

[0036] S120: Input the first target current and the second target current into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model respectively to extract features, and obtain the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current.

[0037] Specifically, the target neural network model can extract deep features related to the fault from the zero-sequence current waveform. It can include different layers, such as feature extraction networks and feature fusion layers. After feature extraction by the first feature extraction network and the second feature extraction network, the corresponding feature vector can be obtained.

[0038] S130: The first feature vector and the second feature vector are concatenated through the feature fusion layer of the target neural network model to obtain the fused feature vector.

[0039] Specifically, the feature fusion layer can combine multiple feature vectors into a more comprehensive fused feature vector. For example, it can add corresponding elements of the first and second feature vectors to generate a fused feature vector with the same dimension as the original feature vector.

[0040] S140: Input the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault discrimination result of the target feeder section.

[0041] Specifically, after obtaining the fused feature vector, it can be input into a linear classifier. The linear classifier can output the value corresponding to the fused feature vector. If the value is greater than a preset threshold, the fault identification result of the target feeder segment corresponding to the fused feature vector is a fault segment; otherwise, it is a normal segment.

[0042] For example, in a distribution network, if a single-phase ground fault occurs in a feeder section, the zero-sequence current waveform at the time of the fault can be acquired in real time by feeder terminal units (FTUs) deployed at both ends of the feeder section. Then, these two raw zero-sequence current waveforms are normalized to obtain a first target current and a second target current. These two target currents are then input into a target neural network model. The first feature extraction network receives the first target current and performs multi-level feature learning on the waveform through its internal convolutional and pooling layers, capturing the local transient features of the waveform and converting it into a high-dimensional first feature vector. The second feature extraction network processes the second target current in the same way, generating the corresponding... The second feature vector is then concatenated by the feature fusion layer to obtain a fused feature vector. Finally, the fused feature vector is input into a linear classifier, which can determine whether the feeder section is in a fault state or a normal state based on the pattern of the fused feature vector. For example, the linear classifier outputs a classification probability value. If the probability value indicates a high probability of a fault (e.g., greater than 0.8), the feeder section is determined to be a fault section; if the probability value indicates a high probability of a normal state, the section is determined to be a normal section. This effectively solves the problem of accurately locating the fault location of a single-phase ground fault in a resonant distribution network due to weak features, and improves the operational stability of the distribution network.

[0043] This application discloses a fault judgment method for a distribution network feeder section, comprising: acquiring the zero-sequence current at both ends of a target feeder section in the distribution network, and normalizing the zero-sequence current at both ends of the target feeder section to obtain a normalized first target current and a second target current; inputting the first target current and the second target current into the first feature extraction network and the second feature extraction network of a pre-trained target neural network model respectively for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector; and inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault judgment result of the target feeder section. In this method, by acquiring the zero-sequence current at both ends of the target feeder section, normalizing it, extracting and fusing features using a pre-trained target neural network model, and finally classifying and discriminating using a linear classifier, the fault section can be accurately judged, thereby improving the accuracy of fault section location in the distribution network and reducing the impact of noise interference and signal transformation sensitivity.

[0044] In one possible implementation, the first zero-sequence current and the second zero-sequence current are normalized by: normalizing the first zero-sequence current and the second zero-sequence current using the following formula:

[0045]

[0046] in, The current value at the i-th sampling point in the waveform of the zero-sequence current; A dataset consisting of the waveforms of zero-sequence current; The minimum current value in the dataset consisting of the waveforms of the zero-sequence current; The maximum current value in the dataset consisting of the waveforms of the zero-sequence current; This represents the current value at the i-th sampling point in the normalized zero-sequence current waveform.

[0047] In one possible implementation, the training process of the target neural network model includes: acquiring historical fault data; the historical fault data includes fault zero-sequence currents collected under different operating conditions; labeling fault zero-sequence current samples as upstream fault current samples and / or downstream fault current samples according to the positional relationship between the collection location of the fault zero-sequence current and the two ends of the fault feeder section, and constructing a pre-training dataset based on the upstream fault current samples and / or downstream fault current samples; inputting the pre-training dataset into a preset convolutional neural network model, the preset convolutional neural network model including a classification output link; training the preset convolutional neural network model using the pre-training dataset until the classification output link can distinguish between upstream fault current samples and downstream fault current samples, thereby obtaining a target neural network model including a first feature extraction network and a second feature extraction network.

[0048] Specifically, historical fault data refers to relevant data generated from historical fault events in the distribution network. This data can include zero-sequence fault currents collected under different operating conditions. If the fault occurs on the power source side (upstream) of the feeder section, the collected zero-sequence current exhibits upstream fault characteristics and is used as an upstream fault current sample. If the fault occurs on the load side (downstream) of the feeder section, it exhibits downstream fault characteristics and is used as a downstream fault current sample. A pre-trained dataset is then constructed based on the current samples. A pre-defined convolutional neural network model can classify the fault direction based on the extracted features. Through repeated iterative training, the model continuously adjusts its internal parameters until the classification output link can accurately distinguish between upstream and downstream fault current samples. The trained convolutional neural network model includes a first feature extraction network and a second feature extraction network for processing the zero-sequence current at both ends of the feeder section, so that the target neural network model can extract the classified feature vector from the zero-sequence current at both ends of the feeder section more accurately and effectively.

[0049] For example, multiple sets of zero-sequence current data can be extracted from historical fault data. If the fault point is located on the power supply side of the feeder section, the zero-sequence current sample is labeled as an upstream fault; if the fault point is located on the load side of the feeder section, it is labeled as a downstream fault. For instance, for a feeder section AB, if the fault occurs before point A, the zero-sequence current samples collected from points A and B are both labeled as upstream fault samples; if the fault occurs after point B, they are all labeled as downstream fault samples. Then, a pre-training dataset is constructed, and then the pre-training dataset is input into a convolutional neural network model. This convolutional neural network model can contain three convolutional layers, each followed by a ReLU activation function, a max pooling layer, two fully connected layers, and a Softmax classification output layer. After the pre-training dataset is input into this convolutional neural network model, it can be trained using the Adam optimizer and the cross-entropy loss function. When the model's classification output link achieves a preset threshold in distinguishing between upstream and downstream fault current samples, the convolutional neural network model can be used as the target neural network model.

[0050] In one possible implementation, a fused feature vector is obtained by concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model. This includes: the feature fusion layer performing a vector concatenation operation on the first feature vector and the second feature vector, connecting the first feature vector and the second feature vector end to end in a preset order to obtain the fused feature vector; wherein the dimension of the fused feature vector is the sum of the dimensions of the first feature vector and the second feature vector.

[0051] In one possible implementation, the first target current and the second target current are respectively input into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current. This includes: inputting the target current into the convolutional layer of the target neural network model, performing a sliding convolution operation on the adjustment vector through the convolutional layer to obtain the local transient features corresponding to the target current, and generating a first feature map corresponding to the local transient features; inputting the preliminary feature map into the pooling layer of the target neural network model, performing dimensionality reduction processing on the first feature map according to the max pooling strategy to obtain the second feature map; and performing a flattening operation on the second feature map to obtain the feature vector corresponding to the target current.

[0052] Specifically, the convolutional layer can contain multiple convolutional kernels. These kernels can perform sliding convolution operations on the target current. Sliding convolution refers to the kernel sliding across the target current with a preset step size, performing element-wise multiplication and summation operations with the local region of the target current at each position to obtain the local transient features corresponding to the target current. These local transient features are the rapid change patterns of the current signal in a short period of time, such as sudden current changes, high-frequency oscillations, or specific waveform distortions during a fault. The local transient features extracted by the convolutional layer can effectively capture instantaneous and local signal anomalies in the target current and construct a first feature map. The first feature map is then input into a pooling layer to perform pooling operations, reducing the dimension of the feature map to obtain a second feature map. Finally, the second feature map is flattened to convert the multidimensional feature map into a one-dimensional vector, obtaining the feature vector corresponding to the target current. This feature vector includes all the extracted local transient features in the target current and can be used as subsequent input to the fully connected layer.

[0053] In one possible implementation, the fused feature vector is input into a preset linear classifier for classification and discrimination to determine the fault discrimination result of the target feeder segment. This includes: inputting the fused feature vector into the fully connected layer of the preset linear classifier, performing nonlinear mapping on the fused feature vector through the fully connected layer to obtain a target feature vector; inputting the target feature vector into the output layer of the preset linear classifier, processing the target feature vector through an activation function, and outputting the classification probability value corresponding to the target feature vector; the classification probability value includes a first probability value corresponding to the faulty segment and a second probability value corresponding to the normal segment.

[0054] Specifically, after obtaining the fused feature vector, it can be input into a pre-defined linear classifier composed of a multilayer perceptron (MLP). This classifier first includes a fully connected layer, which can consist of several neurons, each connected to all dimensions of the fused feature vector. After this fully connected layer, the fused feature vector is non-linearly mapped using the ReLU function to obtain the target feature vector. Subsequently, the target feature vector is input into the output layer of the classifier. This output layer can use the Sigmoid activation function to map the input target feature vector to a value between 0 and 1. This value is the first probability value corresponding to the faulty section. Correspondingly, the second probability value corresponding to the normal section can be obtained by subtracting the first probability value from 1. This improves the accuracy and robustness of fault identification, especially when facing different operating conditions of the distribution network, enabling more accurate identification of faulty sections.

[0055] In one possible implementation, the method further includes: if only one target feeder segment in the distribution network is classified as a fault segment, then the feeder segment is identified as a target fault segment; if at least two target feeder segments in the distribution network are classified as fault segments, then the first probability values ​​of the target feeder segments are compared, and the target feeder segment corresponding to the largest first probability value is identified as the target fault segment; if no target feeder segment in the distribution network is classified as a fault segment, then a warning message is generated.

[0056] This application discloses a fault judgment method for a distribution network feeder section, comprising: acquiring the zero-sequence current at both ends of a target feeder section in the distribution network, and normalizing the zero-sequence current at both ends of the target feeder section to obtain a normalized first target current and a second target current; inputting the first target current and the second target current into the first feature extraction network and the second feature extraction network of a pre-trained target neural network model respectively for feature extraction to obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector; and inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault judgment result of the target feeder section. In this method, by acquiring the zero-sequence current at both ends of the target feeder section, normalizing it, extracting and fusing features using a pre-trained target neural network model, and finally classifying and discriminating using a linear classifier, the fault section can be accurately judged, thereby improving the accuracy of fault section location in the distribution network and reducing the impact of noise interference and signal transformation sensitivity.

[0057] Corresponding to the aforementioned application function implementation method embodiments, this application also provides a fault judgment device, electronic equipment, and corresponding embodiments for a distribution network feeder section.

[0058] The method of this invention was experimentally verified in the scenario of locating a single-phase ground fault section in a distribution network. Method A (2D-ConvNet location method based on three-phase current) and Method B (1D-ConvNet location method based on transient zero-sequence current single-channel splicing) were selected as comparison benchmarks. Combining the computational efficiency data in Table 1 and the location accuracy data in Table 2, it can be seen that: in terms of computational performance, the floating-point operation quantity of the method of this invention is only 0.015M and the number of parameters is 0.006M, which is much lower than that of Method A. The execution time of a single sample is 1.065ms. While ensuring accuracy, it achieves lightweight and fast inference, which is more suitable for on-site deployment in distribution network projects. In terms of location accuracy, under standard fault conditions, the accuracy, precision, and recall rate of the method of this invention all reach 100%, which is better than Method A and Method B. It effectively overcomes the problem that the fault current characteristics are weak due to the compensation of the arc suppression coil in the distribution network, and that traditional methods are prone to misjudgment and missed judgment, thus achieving accurate location of the fault section.

[0059] Table 1 Comparison of computational efficiency of neural network models

[0060]

[0061] Table 2. Positioning performance of the method of the present invention and the comparative method

[0062]

[0063] Figure 2 This is a schematic diagram of the structure of a fault diagnosis device for a distribution network feeder section shown in an embodiment of this application.

[0064] See Figure 2 A fault diagnosis device 200 for a distribution network feeder section includes:

[0065] The acquisition module 210 is used to acquire the zero-sequence current at both ends of the target feeder section in the distribution network, and to normalize the zero-sequence current at both ends of the target feeder section to obtain the normalized first target current and second target current.

[0066] The processing module 220 is used to input the first target current and the second target current into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model respectively to extract features, so as to obtain the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current.

[0067] The fusion module 230 is used to concatenate the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector.

[0068] The classification module 240 is used to input the fused feature vector into a preset linear classifier for classification and discrimination, and to determine the fault discrimination result of the target feeder section.

[0069] In one possible implementation, the acquisition module 210 is further configured to normalize the first zero-sequence current and the second zero-sequence current using the following formula:

[0070]

[0071] in, The current value at the i-th sampling point in the waveform of the zero-sequence current; A dataset consisting of the waveforms of zero-sequence current; The minimum current value in the dataset consisting of the waveforms of the zero-sequence current; The maximum current value in the dataset consisting of the waveforms of the zero-sequence current; This represents the current value at the i-th sampling point in the normalized zero-sequence current waveform.

[0072] In one possible implementation, the processing module 220 is further configured to acquire historical fault data; the historical fault data includes fault zero-sequence currents collected under different operating conditions; based on the positional relationship between the fault zero-sequence current collection location and the two ends of the fault feeder section, the fault zero-sequence current samples are labeled as upstream fault current samples and / or downstream fault current samples, and a pre-training dataset is constructed based on the upstream fault current samples and / or downstream fault current samples; the pre-training dataset is input into a preset convolutional neural network model, the preset convolutional neural network model including a classification output link; the preset convolutional neural network model is trained using the pre-training dataset until the classification output link can distinguish between upstream fault current samples and downstream fault current samples, thereby obtaining a target neural network model including a first feature extraction network and a second feature extraction network.

[0073] In one possible implementation, the fusion module 230 is further configured to perform a vector concatenation operation on the first feature vector and the second feature vector by connecting the first feature vector and the second feature vector in a preset order to obtain a fused feature vector; wherein the dimension of the fused feature vector is the sum of the dimensions of the first feature vector and the second feature vector.

[0074] In one possible implementation, the processing module 220 is further configured to input the target current into the convolutional layer of the target neural network model, perform sliding convolution operation on the adjustment vector through the convolutional layer to obtain the local transient features corresponding to the target current, and generate a first feature map corresponding to the local transient features; input the preliminary feature map into the pooling layer of the target neural network model, perform dimensionality reduction processing on the first feature map according to the max pooling strategy to obtain a second feature map; and perform flattening operation on the second feature map to obtain the feature vector corresponding to the target current.

[0075] In one possible implementation, the classification module 240 is further configured to input the fused feature vector into the fully connected layer of a preset linear classifier, perform nonlinear mapping on the fused feature vector through the fully connected layer to obtain a target feature vector; input the target feature vector into the output layer of the preset linear classifier, process the target feature vector through an activation function, and output the classification probability value corresponding to the target feature vector; the classification probability value includes a first probability value corresponding to the faulty segment and a second probability value corresponding to the normal segment.

[0076] In one possible implementation, the classification module 240 is further configured to: if only one target feeder segment in the distribution network is classified as a fault segment, then determine the feeder segment as a target fault segment; if at least two target feeder segments in the distribution network are classified as fault segments, then compare the first probability values ​​of the target feeder segments and determine the target feeder segment corresponding to the largest first probability value as the target fault segment; if no target feeder segment in the distribution network is classified as a fault segment, then generate a warning message.

[0077] This application discloses a fault diagnosis device for a distribution network feeder section, comprising: an acquisition module for acquiring the zero-sequence currents at both ends of a target feeder section in the distribution network and normalizing the zero-sequence currents at both ends of the target feeder section to obtain a normalized first target current and a second target current; a processing module for inputting the first target current and the second target current into a pre-trained target neural network model's first feature extraction network and second feature extraction network, respectively, to extract features and obtain a first feature vector corresponding to the first target current and a second feature vector corresponding to the second target current; a fusion module for concatenating the first feature vector and the second feature vector through a feature fusion layer of the target neural network model to obtain a fused feature vector; and a classification module for inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault diagnosis result of the target feeder section. In this method, by acquiring the zero-sequence currents at both ends of the target feeder section, normalizing them, extracting and fusing features using a pre-trained target neural network model, and finally classifying and discriminating using a linear classifier, the fault section can be accurately determined, improving the accuracy of fault section location in the distribution network and reducing the impact of noise interference and signal transformation sensitivity.

[0078] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated further here.

[0079] This application also provides an electronic device. Figure 3 This is a schematic diagram of the hardware structure of an embodiment of the electronic device of this application. The electronic device includes a memory 320 and at least one processor 310. The memory 320 is electrically connected to the at least one processor 310. The memory 320 stores instructions. The at least one processor 310 calls the instructions in the memory 320 to cause the electronic device to execute the fault judgment method for the feeder section of the distribution network according to any of the foregoing embodiments of this application.

[0080] Specifically, the processor 310 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0081] Memory 320 may include a mass storage device for data or instructions. For example, and not limitingly, memory 320 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 320 may include removable or non-removable (or fixed) media. Where appropriate, memory 320 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 320 is non-volatile solid-state memory. In a particular embodiment, memory 320 includes read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0082] In one example, the control device may also include a communication interface 330 and a bus 340. The processor 310, memory 320, and communication interface 330 are connected via the bus 340 and communicate with each other.

[0083] The communication interface 330 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0084] Bus 340 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 340 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0085] Furthermore, in conjunction with the fault diagnosis method for distribution network feeder sections in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores executable code, which, when executed by a processor, implements any of the fault diagnosis methods for distribution network feeder sections in the above embodiments.

[0086] This application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0087] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0088] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0089] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for fault diagnosis in a feeder section of a distribution network, characterized in that, include: The zero-sequence current at both ends of the target feeder section in the distribution network is obtained, and the zero-sequence current at both ends of the target feeder section is normalized to obtain the normalized first target current and second target current. The first target current and the second target current are respectively input into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model to extract features, thereby obtaining the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current. The first feature vector and the second feature vector are concatenated through the feature fusion layer of the target neural network model to obtain a fused feature vector; The fused feature vector is input into a preset linear classifier for classification and discrimination to determine the fault discrimination result of the target feeder section.

2. The method according to claim 1, characterized in that, The normalization process for the first zero-sequence current and the second zero-sequence current includes: The first zero-sequence current and the second zero-sequence current are normalized using the following formula: , in, The value of the current at the i-th sampling point in the waveform of the zero-sequence current; The dataset consists of the waveforms of the zero-sequence current; The minimum current value in the dataset consisting of the waveforms of the zero-sequence current; The maximum current value in the dataset consisting of the waveforms of the zero-sequence current; This represents the current value at the i-th sampling point in the normalized zero-sequence current waveform.

3. The method according to claim 1, characterized in that, The training process of the target neural network model includes: Acquire historical fault data; the historical fault data includes fault zero-sequence current collected under different operating conditions; Based on the positional relationship between the sampling location of the fault zero-sequence current and the two ends of the fault feeder section, the fault zero-sequence current samples are labeled as upstream fault current samples and / or downstream fault current samples, and a pre-training dataset is constructed based on the upstream fault current samples and / or downstream fault current samples. The pre-trained dataset is input into a preset convolutional neural network model, which includes a classification output path. The pre-trained dataset is used to train the preset convolutional neural network model until the classification output link can distinguish between upstream fault current samples and downstream fault current samples, thereby obtaining a target neural network model including the first feature extraction network and the second feature extraction network.

4. The method according to claim 1, characterized in that, The step of concatenating the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain the fused feature vector includes: The feature fusion layer performs a vector concatenation operation on the first feature vector and the second feature vector, connecting the first feature vector and the second feature vector end to end in a preset order to obtain the fused feature vector; wherein the dimension of the fused feature vector is the sum of the dimensions of the first feature vector and the second feature vector.

5. The method according to claim 1, characterized in that, The step of inputting the first target current and the second target current into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model for feature extraction, respectively, to obtain the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current, includes: The target current is input into the convolutional layer of the target neural network model, and the adjustment vector is subjected to sliding convolution operation through the convolutional layer to obtain the local transient features corresponding to the target current, and a first feature map corresponding to the local transient features is generated. The preliminary feature map is input into the pooling layer of the target neural network model, and the first feature map is dimensionality reduced according to the max pooling strategy to obtain the second feature map; The second feature map is flattened to obtain the feature vector corresponding to the target current.

6. The method according to claim 1 or 5, characterized in that, The step of inputting the fused feature vector into a preset linear classifier for classification and discrimination to determine the fault discrimination result of the target feeder section includes: The fused feature vector is input into the fully connected layer of the preset linear classifier, and the fused feature vector is non-linearly mapped through the fully connected layer to obtain the target feature vector; The target feature vector is input to the output layer of the preset linear classifier, and the target feature vector is processed by the activation function to output the classification probability value corresponding to the target feature vector; the classification probability value includes a first probability value corresponding to the faulty segment and a second probability value corresponding to the normal segment.

7. The method according to claim 7, characterized in that, The method further includes: If only one target feeder section in the distribution network is classified as a fault section, then the feeder section is identified as the target fault section. If at least two target feeder sections in the distribution network are classified as fault sections, then the first probability values ​​of the target feeder sections are compared, and the target feeder section corresponding to the largest first probability value is determined as the target fault section. If the classification result of the target feeder section in the distribution network is not a fault section, a warning message is generated.

8. A fault diagnosis device for a feeder section of a power distribution network, characterized in that, include: The acquisition module is used to acquire the zero-sequence current at both ends of the target feeder section in the distribution network, and to normalize the zero-sequence current at both ends of the target feeder section to obtain the normalized first target current and second target current. The processing module is used to input the first target current and the second target current into the first feature extraction network and the second feature extraction network of the pre-trained target neural network model respectively for feature extraction, so as to obtain the first feature vector corresponding to the first target current and the second feature vector corresponding to the second target current; The fusion module is used to concatenate the first feature vector and the second feature vector through the feature fusion layer of the target neural network model to obtain a fused feature vector; The classification module is used to input the fused feature vector into a preset linear classifier for classification and discrimination, and to determine the fault discrimination result of the target feeder section.

9. An electronic device, characterized in that, include: processor; as well as A memory having executable code stored thereon, which, when executed by the processor, causes the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores executable code that, when executed by a processor of an electronic device, causes the processor to perform the method as described in any one of claims 1-7.

Images