A method, system and storage medium for single-phase ground fault traveling wave section positioning in an active power distribution network

Abstract

The present application relates to the technical field of power distribution network fault diagnosis, and discloses a method and system for single-phase ground fault traveling wave section positioning in an active power distribution network and a storage medium, wherein signals during the half cycle before the fault starts to the two cycles after the fault are intercepted as signals to be analyzed, Karenbauer transformation is used to transform three-phase current phase components into current and traveling wave components, and the current and traveling wave components are superimposed to make a data set. A ConvNext-RFCBAM model is constructed, and the original module in the ConvNext model is replaced with an RFCBAM module in which the depth 2d convolution is replaced. The trained ConvNext-RFCBAM model is used to determine single-phase ground faults and perform section positioning on the single-phase ground faults. Compared with the prior art, the present application solves the problem of insufficient position feature perception caused by traditional convolution kernel parameter sharing, so that each receptive field sliding window obtains exclusive attention weight, avoiding the defect of traditional spatial attention weight sharing across sliding windows. At the same time, the accuracy and anti-interference ability of single-phase ground fault section positioning in the active power distribution network can be improved.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault diagnosis technology, and in particular to a method, system and storage medium for locating traveling wave sections of single-phase grounding faults in active power distribution networks. Background Technology

[0002] In my country, medium- and low-voltage active distribution networks widely adopt an operation mode where the neutral point is ungrounded and grounded via an arc suppression coil or a small resistor. These distribution networks are connected to a large number of distributed power sources such as photovoltaics and wind power, becoming complex networks with multiple power sources. When a single-phase ground fault occurs in the distribution network, a traveling wave signal is generated at the fault point and propagates along the line. The amplitude, wavefront arrival time, and propagation speed of the traveling wave are intrinsically related to the fault section. However, the inverter-type output characteristics of distributed power sources can change the generation and propagation patterns of the fault traveling wave, causing the traveling wave signal to exhibit nonlinear and multi-source superposition characteristics. At the same time, the multi-branch and topologically variable characteristics of the distribution network can also cause the traveling wave to be reflected, further increasing the difficulty of section location.

[0003] Traditional traveling wave (TW) location methods for single-phase-to-ground fault traveling wave sections in active distribution networks have several limitations. Single-ended TW methods are susceptible to wave velocity estimation errors and reflection / reflection interference, resulting in low location accuracy. Double-ended TW methods require stringent data synchronization and necessitate high-precision measurement equipment at both ends of the line, leading to high engineering implementation costs. Existing TW section location methods are ill-suited to effectively address the challenges of multiple power sources, varied topologies, and complex TW characteristics in active distribution networks, and their location accuracy and robustness fail to meet practical engineering requirements.

[0004] Artificial Intelligence (AI) is an interdisciplinary field that integrates machine learning, deep learning, and other technologies. It can learn from massive amounts of data to uncover hidden patterns and achieve accurate modeling and solutions to complex nonlinear problems. Within the AI ​​framework, deep learning models, with their powerful feature self-learning capabilities, have become an effective means of analyzing and locating fault characteristics in power systems. Models such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Graph Neural Networks (GNN) can efficiently extract time-domain, frequency-domain, and topological correlation features of traveling waves, respectively. For the analysis of traveling wave signals, AI methods do not require complex mechanistic models; they can directly learn the mapping relationship between fault sections and traveling wave characteristics from the voltage and current time-series data of the traveling wave, improving the model's robustness under complex operating conditions.

[0005] In recent years, the application of artificial intelligence (AI) technology in power system fault location has gradually increased. However, research on the location of traveling wave sections of single-phase grounding faults in active distribution networks still has shortcomings: existing models are mostly designed for traditional passive distribution networks and do not fully consider the impact of distributed generation on traveling wave characteristics. Therefore, it is urgent to use AI methods to construct a single-phase grounding fault traveling wave section location model adapted to the characteristics of active distribution networks, which solves the problems of low location accuracy and poor adaptability of traditional methods and enables rapid and accurate location of fault sections. Summary of the Invention

[0006] Purpose of the invention: In view of the problems pointed out in the background art, the present invention provides a method, system and storage medium for locating traveling wave sections of single-phase grounding faults in active distribution networks, which can improve the accuracy and anti-interference ability of locating traveling wave sections of single-phase grounding faults in active distribution networks.

[0007] Technical solution: This invention provides a method for locating the traveling wave section of a single-phase ground fault in an active power distribution network, comprising:

[0008] Collect the three-phase current of each line, and extract the signal from the first half cycle of the fault to the two cycles after the fault as the signal to be analyzed.

[0009] The Karenbauer transform converts the intercepted three-phase current phase components into currents. and Traveling wave components, maximum values ​​after normalization, superimposed with normalized values and Dataset is created by constructing feature maps from current traveling wave components;

[0010] ConvNext-RFCBAM model is constructed, which replaces the 2D convolution in the original module with RFCBAM module based on the ConvNext model, and the network model is trained using dataset.

[0011] The trained ConvNext-RFCBAM model is used to determine single-phase grounding faults and to locate the fault sections.

[0012] Furthermore, before collecting the three-phase current of each line, zero-sequence voltage and three-phase current data of the line are collected from the target monitoring area. If the zero-sequence voltage exceeds the preset threshold or its variation exceeds the set range, the single-phase grounding fault section location process is initiated; otherwise, monitoring of the line continues until the location conditions are met. Detecting that the zero-sequence voltage exceeds the preset threshold or its variation exceeds the specified range means that the zero-sequence voltage U0 > the activation value U0. set , or the zero-sequence voltage mutation △U0 > the set value ε.

[0013] Furthermore, the Karenbauer transform converts the phase components of the three-phase current into currents. and The traveling wave component, using the Karenbauer transform formula, is as follows:

[0014] ;

[0015] in, , and These are the currents of phases a, b, and c, respectively. , and They are zero modulus, and Current traveling wave component.

[0016] Will and The formula for normalizing the maximum value of the traveling wave component is as follows:

[0017]

[0018] This represents the magnitude of the electrical quantity at the current moment in the time-domain image after normalization. This electrical quantity is either zero-sequence voltage or zero-sequence current. This represents the actual amplitude of the electrical quantity at the current moment in the time-domain image; This represents the maximum amplitude of the electrical quantity recorded within the time window T.

[0019] Furthermore, the ConvNext-RFCBAM model includes: a 2D convolutional layer, 3 ConvNeXt-RFCBAM Block sub-modules, a downsampling layer, 3 ConvNeXt-RFCBAM Block sub-modules, a downsampling layer, 9 ConvNeXt-RFCBAM Block sub-modules, a downsampling layer, 3 ConvNeXt Block-RFCBAM sub-modules, a pooling layer, and a fully connected layer.

[0020] Furthermore, the ConvNeXt Block-RFCBAM submodule consists of an RFCBAM module, a normalization layer, a convolutional layer, a GELU activation function, and a regularization layer. The RFCBAM module takes the spatial features of the receptive field as the core optimization direction, reconstructs the attention mechanism on the basis of the classic CBAM, and simultaneously integrates channel attention and spatial attention, focusing on the independent weight allocation of features at each location within the receptive field, so that each receptive field sliding window obtains its own attention weight.

[0021] Furthermore, the training steps of the ConvNext-RFCBAM model include:

[0022] Obtain the image dataset, which includes seven different labeled lines. and Time-domain image of the traveling wave component of the current;

[0023] The image dataset was used as the training set, and the following labels were used: "Line normal", "Single-phase ground fault occurred 0 to 1 km from the first section of the feeder", "Single-phase ground fault occurred 1 to 2 km from the first section of the feeder", "Single-phase ground fault occurred 2 to 3 km from the first section of the feeder", "Single-phase ground fault occurred 3 to 4 km from the first section of the feeder", "Single-phase ground fault occurred 4 to 5 km from the first section of the feeder", and "Single-phase ground fault occurred 5 to 6 km from the first section of the feeder".

[0024] The ConvNext-RFCBAM model was trained with parameter optimization using the training dataset. Gradient descent and other optimization algorithms were used to minimize the prediction error. After multiple iterations and verifications, the optimal configuration parameters were finally determined.

[0025] Based on the optimal configuration parameters of the ConvNext-RFCBAM model, a trained single-phase ground fault traveling wave segment location model is obtained.

[0026] Furthermore, the starting value is set to 0.1 times the rated voltage, ΔU0 is the current zero-sequence voltage minus the previous zero-sequence voltage, and ε takes a value of 0~0.1V.

[0027] This invention also discloses a system for locating traveling wave sections of single-phase ground faults in active power distribution networks, comprising a data acquisition unit, an analysis unit, and a fault location unit:

[0028] The dataset acquisition unit is used to collect the three-phase current of each line and extract the signal from the first half-cycle before the fault starts to the two cycles after the fault as the signal to be analyzed; the Karenbauer transform converts the extracted three-phase current phase components into current. and Traveling wave components, maximum values ​​after normalization, superimposed with normalized values and Dataset is created by constructing feature maps from current traveling wave components;

[0029] The model building module is used to build the ConvNext-RFCBAM model, which replaces the 2d depth convolution in the original module with the RFCBAM module based on the ConvNext model.

[0030] The model training module is used to train the constructed ConvNext-RFCBAM model based on the dataset obtained by the dataset acquisition unit;

[0031] The fault location unit is used to identify feature images using a pre-trained ConvNext-RFCBAM model, and finally outputs the section location conclusion of the single-phase grounding fault of the line.

[0032] Also includes:

[0033] The data acquisition unit is used to collect the zero-sequence and three-phase electrical parameters of the lines in the active power distribution network of the monitoring area in real time;

[0034] The analysis unit monitors the collected zero-sequence voltage signal in real time. When the zero-sequence voltage exceeds the preset threshold or a significant change occurs, the fault location process is triggered; otherwise, the monitoring of the line continues for cyclic detection.

[0035] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed, implements the above-described method for locating traveling wave sections of single-phase grounding faults in active power distribution networks.

[0036] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0037] 1. The model training process of this invention uses a sample library based on image features, which contains traveling wave electrical feature maps of the system under normal operating and fault conditions. These feature maps are obtained by analyzing current... and The composite image generated by standardizing and feature fusion of the time-domain waveform of the traveling wave component ensures that the trained fault identification model has excellent noise suppression performance.

[0038] 2. The single-phase grounding fault location model of this invention is based on the ConvNext-RFCBAM neural network, which improves the accuracy of fault location compared with the original convolutional neural networks (VGG16 network, AlexNet network, ResNet network, DenseNet network and ConvNext network). Furthermore, the fault detection using the single-phase grounding fault location model still has a high detection accuracy in noisy environments, indicating that this method has good anti-interference ability. In addition, this invention improves the 2D convolution in the original module of the ConvNext network, so that the model focuses on the independent weight allocation of features at each location within the receptive field, which solves the problem of insufficient location feature perception caused by the sharing of traditional convolution kernel parameters. This allows each receptive field sliding window to obtain its own attention weight, avoiding the defect of traditional spatial attention weight sharing across sliding windows. Attached Figure Description

[0039] Figure 1The diagram shown is a flowchart of a method for locating a single-phase grounding fault section in an active power distribution network according to an embodiment of the present invention.

[0040] Figure 2 The image shown is an example of normalized traveling wave image data after a single-phase grounding with a transition resistance of 0.01 ohms occurred at a distance of 1 to 2 km, according to an embodiment of the present invention.

[0041] Figure 3 The image shown is an example of normalized traveling wave image data after a single-phase grounding with a 100-ohm transition resistance occurred at a distance of 1 to 2 km, according to an embodiment of the present invention.

[0042] Figure 4 The image shown is an example of image data normalized without single-phase grounding traveling wave in one embodiment of the present invention;

[0043] Figure 5 The diagram shown is a ConvNext-RFCBAM network structure diagram in one embodiment of the present invention;

[0044] Figure 6 The figure shown is a simulation model diagram of a single-phase grounding fault section location in an active power distribution network according to one embodiment of the present invention.

[0045] Figure 7 The diagram shown illustrates the change in loss value during the training process of the ConvNext-RFCBAM network in one embodiment of the present invention.

[0046] Figure 8 The diagram shown illustrates the accuracy variation during the training process of the ConvNext-RFCBAM network in one embodiment of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0048] Example 1

[0049] like Figure 1 As shown in the figure, this invention provides a method for locating the traveling wave section of a single-phase ground fault in an active distribution network, including the following steps:

[0050] S1: Real-time acquisition of zero-sequence voltage and three-phase current of power distribution lines within the monitoring area.

[0051] In this embodiment, step S1 acquires the zero-sequence voltage and three-phase current of the line under test in the area under test, and utilizes the sensing unit (including a three-phase current detector and a zero-sequence voltage detector) integrated into the microprocessor protection system to achieve real-time data acquisition. As an important component of the smart grid, the microprocessor protection system integrates advanced technologies such as digital processing technology, intelligent algorithms, and automated control, and possesses advantages such as accurate measurement, stable operation, and adaptive decision-making. Since this device is existing technology and not an innovation of this invention, it will not be described in detail here.

[0052] S2: Perform threshold judgment on the acquired zero-sequence voltage signal. When the zero-sequence voltage amplitude exceeds the preset threshold or fluctuates significantly, execute the fault location process. If the trigger condition is not met, switch the monitoring object and re-execute the parameter acquisition and judgment process until the fault location condition is met.

[0053] In this embodiment, if the zero-sequence voltage of the line under test is greater than the starting value, or the sudden change in zero-sequence voltage is greater than the set value, then the successive grounding fault detection step is entered; otherwise, the line continues to be monitored and the above steps are repeated until the single-phase grounding fault section location step is entered, specifically including:

[0054] Based on the zero-sequence voltage U0 obtained in real time in step S1, it is determined whether the line under test needs to enter the single-phase ground fault section for location; if so, proceed to the successive ground fault detection steps S3-S4; if not, select another line as the line under test and return to step S1.

[0055] The judgment criterion is: zero-sequence voltage U0 > starting value U set Or, the zero-sequence voltage mutation ΔU0 > the set value ε;

[0056] Specifically, U set The value is 0.1 times the rated voltage, ΔU0 is the current zero-sequence voltage minus the previous zero-sequence voltage, and ε is 0-0.1V.

[0057] The fault location process specifically includes:

[0058] S3: Collect the phase current of each line, and extract the signal from the first half cycle of the fault to the two cycles after the fault as the signal to be analyzed; Karenbauer transform converts the phase components of the three-phase current into current. and Traveling wave component; and The traveling wave components are normalized to their maximum values; the normalized currents are then superimposed. and The traveling wave components form a feature map, which is then used to create a dataset.

[0059] In this embodiment, step S3 involves feature extraction and data conversion of the zero-sequence and three-phase electrical parameters of the monitored line to generate a feature map for analysis. The specific processing flow is as follows:

[0060] S31: The Karenbauer transformation converts the phase components of a three-phase current into a current. and The traveling wave component is represented by the Karenbauer transform formula as follows:

[0061]

[0062] in, , and These are the currents of phases a, b, and c, respectively. , and They are zero modulus, and Current traveling wave component.

[0063] Set a time window T, and based on the time window T and the zero-sequence current and zero-sequence voltage of the circuit under test, fuse the zero-sequence voltage and zero-sequence current into a time-domain image.

[0064] Specifically, such as Figure 2-4 As shown, the data acquisition period is configured to 1 microsecond. Based on the zero-sequence and three-phase electrical parameters acquired by the protection system, digital signal processing technology is used to construct a time-domain waveform diagram of the zero-sequence electrical characteristics according to the preset sampling period and frequency specifications. This waveform diagram uses a time-amplitude coordinate system and visually reflects the dynamic changes of the electrical parameters through continuous curves.

[0065] S32: Normalize the time-domain image of the traveling wave current to obtain the normalized time-domain image of the traveling wave current.

[0066] Specifically, the normalization formula is as follows:

[0067]

[0068] In this process, set This represents the magnitude of the electrical quantity at the current moment in the time-domain image after normalization. This electrical quantity can be zero-sequence voltage or zero-sequence current. This represents the actual amplitude of the electrical quantity at the current moment in the time-domain image; while This represents the maximum amplitude of the electrical quantity recorded within the time window T. This normalization process effectively converts electrical quantity data of different magnitudes into a unified scale, facilitating subsequent analysis and processing.

[0069] S33: Yes Current traveling wave and The time-domain images of the current traveling wave normalized are superimposed to obtain preprocessed image data, such as... Figure 2 , Figure 3 and Figure 4 The images shown are, respectively, the preprocessed images of the traveling wave current after a single-phase ground fault with a 0.01-ohm transition resistance occurring at 1 to 2 km, and the normalized images of the traveling wave current after a single-phase ground fault with a 0.01-ohm transition resistance occurring at 1 to 2 km.

[0070] S4: Input the processed feature map into the pre-trained intelligent positioning model and output the section location result of the single-phase grounding fault of the line.

[0071] In this embodiment, the training steps for the single-phase ground fault section location model in step S4 are as follows:

[0072] Obtain the image dataset, which includes time-domain images of current traveling waves of seven different labeled lines after preprocessing.

[0073] The image dataset was used as the training set for the single-phase ground fault section location model, and the following labels were used: "Line normal", "Single-phase ground fault occurred 0 to 1 km from the first section of the feeder", "Single-phase ground fault occurred 1 to 2 km from the first section of the feeder", "Single-phase ground fault occurred 2 to 3 km from the first section of the feeder", "Single-phase ground fault occurred 3 to 4 km from the first section of the feeder", "Single-phase ground fault occurred 4 to 5 km from the first section of the feeder", and "Single-phase ground fault occurred 5 to 6 km from the first section of the feeder".

[0074] Based on the training set, a single-phase ground fault location model is trained to determine its optimal parameters. Based on these optimal parameters, the trained single-phase ground fault location model is obtained.

[0075] Specifically, such as Figure 5 As shown, the structure of the single-phase ground fault location model, namely the ConvNext-RFCBAM model, includes:

[0076] The system consists of a 2D convolutional layer, three ConvNeXt-RFCBAM Block sub-modules, a downsampling layer, three ConvNeXt-RFCBAM Block sub-modules, another downsampling layer, nine ConvNeXt Block-RFCBAM sub-modules, another downsampling layer, three ConvNeXt-RFCBAM Block sub-modules, a pooling layer, and a fully connected layer, all connected sequentially. The RFCBAM module focuses on optimizing the spatial features of the receptive field, reconstructing the attention mechanism based on classic CBAM and simultaneously fusing channel attention and spatial attention. It focuses on the independent weight allocation of features at each location within the receptive field, solving the problem of insufficient location feature perception caused by parameter sharing in traditional convolutional kernels. This allows each receptive field sliding window to obtain its own attention weights, avoiding the defect of traditional spatial attention weight sharing across sliding windows.

[0077] Specifically, the ConvNeXt-RFCBAM Block submodule consists of RFCBAM layers, normalization layers, convolutional layers, GELU activation functions, and regularization layers.

[0078] In this embodiment, the hardware and software environment for model training and verification of this invention patent is: 13th Gen Intelli5-13600KF processor, NVIDIA 4070 (12G), and 32G of memory.

[0079] Example 2:

[0080] This embodiment provides a system for locating traveling wave sections of single-phase ground faults in active distribution networks, including:

[0081] The system includes a data acquisition unit, an analysis unit, a dataset acquisition unit, a model building module, a model training module, and a fault location unit.

[0082] The data acquisition unit is responsible for collecting the zero-sequence electrical parameters of each transmission line within the monitoring area in real time.

[0083] The analysis unit monitors the collected zero-sequence voltage signal in real time. When the zero-sequence voltage exceeds the preset threshold or a significant change occurs, the fault location process is triggered; otherwise, the monitoring of the line continues for cyclic detection.

[0084] The dataset acquisition unit is used to collect the three-phase current of each line and extract the signal from the first half-cycle before the fault starts to the two cycles after the fault as the signal to be analyzed; the Karenbauer transform converts the extracted three-phase current phase components into current. and Traveling wave components, maximum values ​​after normalization, superimposed with normalized values and Data sets are created by constructing feature maps from the traveling wave components of current.

[0085] The model building module is used to build the ConvNext-RFCBAM model, which replaces the 2d depth convolution in the original module with the RFCBAM module based on the ConvNext model.

[0086] The model training module is used to train the constructed ConvNext-RFCBAM model based on the dataset obtained by the dataset acquisition unit.

[0087] The fault location unit performs signal processing and data conversion on the collected zero-sequence electrical parameters to generate a feature image that can be analyzed. It then uses a pre-trained ConvNext-RFCBAM model to identify the feature image and finally outputs the location conclusion of the single-phase grounding fault in the collector line.

[0088] Example 3:

[0089] This embodiment provides a computer-readable storage medium storing a computer program that, when executed, implements the method for locating traveling wave sections of single-phase grounding faults in active power distribution networks as described in Embodiment 1.

[0090] Those skilled in the art should understand that this technical solution can be implemented in various ways, including but not limited to methods, system architectures, or program products. Specifically, it can be implemented on a pure hardware platform, a pure software solution, or a hybrid implementation combining hardware and software. Furthermore, this technical solution can also be embodied in a computer-readable storage medium containing executable program code, with the medium encompassing various storage media such as disks, optical discs, and flash memory.

[0091] Simulation experiment:

[0092] Based on the method for locating traveling wave sections of single-phase ground faults in active distribution networks provided in Example 1, this example provides a calculation case, which is described below:

[0093] This embodiment utilizes PSCAD / EMTDC to simulate the system and build a simulation model. A simplified diagram of its simulation structure is shown below. Figure 6The diagram shows a simulation model of the distribution network after incorporating Distributed Generation (DG). This simulation model divides a 6km distribution line into 1km, 2km, 2km, and 1km sections, with the last 1km being a cable and the remaining sections being overhead lines. The simulation system uses 10kV / 50Hz power supply parameters and includes four distributed power generation units (wind power and photovoltaic power generation models). The arc suppression coil is configured with an inductance of 0.3H. The simulation platform is set to a 1MHz sampling frequency, with a phase-A ground fault occurring on the feeder at 0.1 seconds. The neural network training and verification uses a 0.14s time window. The line parameters are shown in Table 1.

[0094] Table 1 Line Parameters

[0095]

[0096] During the parameter optimization process of the ConvNext-RFCBAM model, the trend of training error with the number of iterations is as follows: Figure 7 As shown, the model error exhibits a monotonically decreasing characteristic and eventually converges to near zero. The model performance evaluation results are as follows: Figure 8 As shown, after 20 training cycles, the classification accuracy reached about 95%, demonstrating excellent recognition ability.

[0097] Validating the ConvNext-RFCBAM model requires building a test set. The method for building the test set is as follows:

[0098] Multiple sets of current data were obtained through simulation, and... and The superimposed waveform of the traveling current wave. The composition of the training set is shown in Table 2, with a total of 792 combinations. The training set, test set, and validation set are divided into a ratio of 5:3:2.

[0099] Define the accuracy of the dataset. To determine the ratio of the number of correct samples to the total number of samples.

[0100] Table 2 Grounding fault parameters of the test dataset

[0101]

[0102] Validation set experiment

[0103] Using 20% ​​of the dataset, the accuracy of each interval was validated. A higher accuracy for an interval indicates that the single-phase ground fault location model has a high accuracy rate in identifying single-phase ground faults occurring in that interval. Table 3 shows the location accuracy of the six intervals for different models in the validation set.

[0104] Table 3. Ground fault location accuracy of the validation dataset

[0105]

[0106] As shown in Table 3, among all the models compared, ConvNext-RFCBAM significantly outperformed other models in all six localization intervals (0-1 km, 1-2 km, 2-3 km, 3-4 km, 4-5 km, and 5-6 km). Specifically, ConvNext-RFCBAM achieved accuracies of 94.05%, 96.42%, 94.83%, 94.51%, 95.02%, and 94.67% in each interval, all exceeding 94%, with the highest reaching 96.42%. In contrast, other models such as VGG16, AlexNet, ResNet, DenseNet, and the original ConvNext model generally fluctuated between 83% and 95% in each interval, with a significant decrease in some intervals. ConvNext-RFCBAM not only maintained high accuracy across all ranges, but also consistently outperformed the original ConvNext model without the RFCBAM module. For example, it improved accuracy by approximately 0.63 percentage points in the 0-1 km range and by approximately 0.58 percentage points in the 1-2 km range. This demonstrates that introducing the RFCBAM attention mechanism effectively enhances the model's ability to extract fault traveling wave features, improves localization robustness and generalization ability at different fault distances, and thus verifies the superiority of the proposed improved structure in single-phase ground fault location tasks.

[0107] It should be noted that although this article describes specific implementation schemes in detail with reference to the accompanying drawings, these embodiments are for illustrative purposes only and do not constitute a limitation on the technical solution. Within the core concept scope of this technical solution, those skilled in the art can make various modifications and improvements, and these derivative solutions should all fall within the protection scope of this technical solution.

Claims

1. A method for locating traveling wave sections of single-phase ground faults in active distribution networks, characterized in that, include: Collect the three-phase current of each line, and extract the signal from the first half cycle of the fault to the two cycles after the fault as the signal to be analyzed. The Karenbauer transform converts the intercepted three-phase current phase components into currents. and Traveling wave components, maximum values ​​after normalization, superimposed with normalized values and Dataset is created by constructing feature maps from current traveling wave components; ConvNext-RFCBAM model is constructed, which replaces the 2D convolution in the original module with RFCBAM module based on the ConvNext model, and the network model is trained using dataset. The trained ConvNext-RFCBAM model is used to determine single-phase grounding faults and to locate the fault sections.

2. The method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to claim 1, characterized in that, Before collecting the three-phase current of each line, zero-sequence voltage and three-phase current data of the lines are collected from the target monitoring area. If the zero-sequence voltage exceeds the preset threshold or its variation exceeds the set range, the single-phase grounding fault section location process is initiated; otherwise, monitoring of the line continues until the location conditions are met. Detecting that the zero-sequence voltage exceeds the preset threshold or its variation exceeds the specified range means that the zero-sequence voltage U0 > the activation value U0. set , or the zero-sequence voltage mutation △U0 > the set value ε.

3. The method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to claim 1, characterized in that, The Karenbauer transform converts the phase components of a three-phase current into currents. and The traveling wave component, using the Karenbauer transform formula, is as follows: ； in, , and These are the currents of phases a, b, and c, respectively. , and They are zero modulus, and Current traveling wave component.

4. The method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to claim 1, characterized in that, The ConvNext-RFCBAM model includes: a sequentially connected 2D convolutional layer, a 3-layer ConvNeXt-RFCBAMBlock submodule, a downsampling layer, a 3-layer ConvNeXt-RFCBAM Block submodule, a downsampling layer, a 9-layer ConvNeXt-RFCBAM Block submodule, a downsampling layer, a 3-layer ConvNeXt Block-RFCBAM submodule, a pooling layer, and a fully connected layer.

5. The method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to claim 4, characterized in that, The ConvNeXt Block-RFCBAM submodule consists of an RFCBAM module, a normalization layer, a convolutional layer, a GELU activation function, and a regularization layer. The RFCBAM module focuses on the spatial features of the receptive field as the core optimization direction. It reconstructs the attention mechanism based on the classic CBAM, and simultaneously integrates channel attention and spatial attention. It focuses on the independent weight allocation of features at each location within the receptive field, allowing each receptive field sliding window to obtain its own exclusive attention weight.

6. The method for locating traveling wave sections of single-phase ground faults in active distribution networks according to claim 1, characterized in that, The training steps of the ConvNext-RFCBAM model include: Obtain the image dataset, which includes seven different labeled lines. and Time-domain image of the traveling wave component of the current; The image dataset was used as the training set, and the following labels were used: "Line normal", "Single-phase ground fault occurred 0 to 1 km from the first section of the feeder", "Single-phase ground fault occurred 1 to 2 km from the first section of the feeder", "Single-phase ground fault occurred 2 to 3 km from the first section of the feeder", "Single-phase ground fault occurred 3 to 4 km from the first section of the feeder", "Single-phase ground fault occurred 4 to 5 km from the first section of the feeder", and "Single-phase ground fault occurred 5 to 6 km from the first section of the feeder". The ConvNext-RFCBAM model was trained with parameter optimization using the training dataset. Gradient descent and other optimization algorithms were used to minimize the prediction error. After multiple iterations and verifications, the optimal configuration parameters were finally determined. Based on the optimal configuration parameters of the ConvNext-RFCBAM model, a trained single-phase ground fault traveling wave segment location model is obtained.

7. The method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to claim 2, characterized in that, The starting value is set to 0.1 times the rated voltage, ΔU0 is the current zero-sequence voltage minus the previous zero-sequence voltage, and ε takes a value of 0~0.1V.

8. A system based on the method for locating traveling wave sections of single-phase grounding faults in active distribution networks according to any one of claims 1 to 7, characterized in that, It includes a data acquisition unit, an analysis unit, and a fault location unit: The dataset acquisition unit is used to collect the three-phase current of each line and extract the signal from the first half-cycle before the fault starts to the two cycles after the fault as the signal to be analyzed; the Karenbauer transform converts the extracted three-phase current phase components into current. and Traveling wave components, maximum values ​​after normalization, superimposed with normalized values and Dataset is created by constructing feature maps from current traveling wave components; The model building module is used to build the ConvNext-RFCBAM model, which replaces the 2d depth convolution in the original module with the RFCBAM module based on the ConvNext model. The model training module is used to train the constructed ConvNext-RFCBAM model based on the dataset obtained by the dataset acquisition unit; The fault location unit is used to identify feature images using a pre-trained ConvNext-RFCBAM model, and finally outputs the section location conclusion of the single-phase grounding fault of the line.

9. The traveling wave section location system for single-phase grounding faults in active distribution networks according to claim 8, characterized in that, Also includes: The data acquisition unit is used to collect the zero-sequence and three-phase electrical parameters of the lines in the active power distribution network of the monitoring area in real time; The analysis unit monitors the collected zero-sequence voltage signal in real time. When the zero-sequence voltage exceeds the preset threshold or a significant change occurs, the fault location process is triggered; otherwise, the monitoring of the line continues for cyclic detection.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed, implements the method for locating traveling wave sections of single-phase grounding faults in active power distribution networks as described in any one of claims 1-7.

Images