Single-phase ground fault locating method and system based on adaptive attention neural network
Patent Information
- Application Number
- CN202611307391.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-09-25
AI Technical Summary
[0006]针对现有单相接地故障定位方法存在人工特征设计复杂、故障特征提取能力有限、复杂工况下定位精度不足以及模型泛化能力较差等问题，本发明提供了一种基于自适应注意力神经网络的单相接地故障定位方法及系统，通过对故障波形进行预处理、波形分段、零序特征构建、自适应注意力特征提取以及多特征融合分析，实现单相接地故障区段的准确识别，提高故障定位精度和工程应用可靠性
本发明通过对故障波形进行预处理、分段以及零序特征构建，建立零序电场与零序电流联合表征模型；利用自适应注意机制自动确定故障波形中的关键时间片段，实现故障关键信息的有效提取；结合三相电流不平衡度、方差、零序电流均方根值以及零序电场与零序电流相似度等辅助特征，实现深度特征与数值特征融合；利用分类模型完成故障区间内部节点和故障区间外部节点判别，并结合配电网拓扑结构实现单相接地故障定位。
Smart Images
Figure CN122815093A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of power distribution network fault diagnosis and artificial intelligence technology, and in particular to a method and system for locating single-phase grounding faults based on an adaptive attention neural network. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Single-phase grounding faults are a common type of fault in power distribution networks. If the location of the fault cannot be determined promptly and accurately after it occurs, it will increase the time for fault investigation and repair, reduce power supply reliability, and in severe cases, may even lead to equipment damage and economic losses. Therefore, achieving rapid and accurate location of single-phase grounding faults is of great engineering significance.
[0004] Currently, existing methods for locating single-phase grounding faults mainly include fault segment identification methods based on electrical quantity characteristics such as zero-sequence current and zero-sequence voltage, fault identification methods based on time-domain, frequency-domain, or time-frequency-domain feature extraction, and methods using artificial intelligence models for fault classification. While these methods have improved the single-phase grounding fault location capability to some extent, they still suffer from drawbacks such as reliance on manual experience in feature design, complex feature extraction processes, and insufficient adaptability to different operating environments and fault conditions. Especially in cases of high-resistance grounding, weak fault current, or strong on-site noise, fault features are easily masked, thus affecting the accuracy of fault location.
[0005] With the development of deep learning technology, using neural networks to automatically identify fault waveforms has become a feasible approach. However, existing deep learning methods often directly borrow image recognition network structures without fully integrating the temporal characteristics and physical features of distribution network fault waveforms. Furthermore, the region in an actual fault waveform that truly reflects fault information usually only occupies a portion of the overall waveform, and a large amount of redundant data weakens the model's ability to extract key fault features. In addition, models trained solely on simulation data are ill-suited to the complex working conditions of the field. Summary of the Invention
[0006] To address the problems of existing single-phase grounding fault location methods, such as complex manual feature design, limited fault feature extraction capabilities, insufficient location accuracy under complex operating conditions, and poor model generalization ability, this invention provides a single-phase grounding fault location method and system based on an adaptive attention neural network. By preprocessing the fault waveform, segmenting the waveform, constructing zero-sequence features, extracting adaptive attention features, and performing multi-feature fusion analysis, the method achieves accurate identification of single-phase grounding fault sections, improving fault location accuracy and reliability in engineering applications.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for locating single-phase grounding faults based on an adaptive attention neural network.
[0008] A single-phase ground fault location method based on an adaptive attention neural network includes: The fault waveform data of each line node during the fault occurrence period are acquired, and then standardized preprocessing, waveform segmentation and zero-sequence synthesis are performed in sequence to obtain the zero-sequence electric field data and zero-sequence current data of each segment, and the numerical characteristic information of each segment is recorded; the numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. The zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment are input into an adaptive attention neural network. This adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the corresponding zero-sequence electric field features, zero-sequence current features, and numerical feature information based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result for each line node based on the auxiliary features. Based on the fault location classification results of each line node, and combined with the distribution network topology, the section where the single-phase grounding fault is located is determined.
[0009] Furthermore, the method for acquiring fault waveform data of each line node during the fault occurrence period, and sequentially performing standardization preprocessing, waveform segmentation, and zero-sequence synthesis to obtain zero-sequence electric field data and zero-sequence current data for each segment includes: The fault waveform data is processed by uniform sampling rate, load current removal, and DC component removal to obtain preprocessed fault waveform data. The preprocessed fault waveform data is segmented, and each segment contains three-phase electric field data and three-phase current data; Zero-sequence electric field data is obtained by performing zero-sequence synthesis on the three-phase electric field data in each segment; zero-sequence current data is obtained by performing zero-sequence synthesis on the three-phase current data in each segment.
[0010] Furthermore, the adaptive attention module is used to perform feature embedding, position encoding, and attention processing on the zero-sequence electric field features, and to determine the location of the critical fault segment based on the attention probability of each segment; including: the adaptive attention module performs feature embedding on the zero-sequence electric field features to obtain electric field feature embedding; introduces time position encoding to distinguish the time position of different segments in the fault waveform data; adds the electric field feature embedding and the time position encoding and performs normalization processing to obtain fused features; calculates the attention matrix based on the fused features, and after normalization, extracts the main diagonal elements to obtain the specific vector corresponding to each segment, and performs softmax processing to obtain the attention probability vector of each segment; based on the attention probability vector of each segment, the segment with the highest attention probability is determined, which is the location of the critical fault segment.
[0011] Furthermore, the auxiliary feature fusion module is used to obtain auxiliary features based on the data output by the feature selection module; including: Calculate the sum of the root mean square values of the three-phase currents based on their root mean square values; calculate the three-phase current asymmetry based on the sum of their root mean square values. Select the maximum value among the root mean square values of the three-phase currents and calculate the variance of the three-phase currents; Calculate the zero-sequence current scaling value based on the root mean square value of the zero-sequence current; Based on the zero-sequence electric field characteristics and the zero-sequence current characteristics, calculate the similarity between the zero-sequence electric field characteristics and the zero-sequence current characteristics; The auxiliary features include: three-phase current asymmetry, three-phase current variance, zero-sequence current scaling value, and similarity between zero-sequence electric field features and zero-sequence current features.
[0012] Furthermore, the fully connected classification module is used to determine the fault location classification result of each line node based on auxiliary features, including: the fully connected classification module outputs the probability that the corresponding line node is located within the fault interval and the probability that it is located outside the fault interval based on the auxiliary features. If the probability that it is located within the fault interval is greater than the probability that it is located outside the fault interval, then the node is determined to be located within the fault interval; otherwise, the node is determined to be located outside the fault interval.
[0013] Furthermore, the adaptive attention neural network updates its network parameters using a joint loss function during training, and the joint loss function is expressed by the following formula:
[0014]
[0015]
[0016] Where Loss represents the joint loss function; Used to constrain the network to focus on key fault segments in the fault waveform; Used to constrain the node classification results output by the network; Indicates training rounds, This indicates the location of the critical fault segment predicted by the model. Indicates the location label of a critical fault segment. Represents the cross-entropy function; This represents the node classification result predicted by the model. This represents the actual category label of the node.
[0017] A second aspect of the present invention provides a single-phase ground fault location system based on an adaptive attention neural network.
[0018] A single-phase ground fault location system based on an adaptive attention neural network includes: The data processing unit is used to acquire fault waveform data of each line node during the fault occurrence, and sequentially perform standardization preprocessing, waveform segmentation and zero-sequence synthesis to obtain zero-sequence electric field data and zero-sequence current data of each segment, and record the numerical characteristic information within each segment; the numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. An adaptive attention neural network unit is used to input the zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment into the adaptive attention neural network. The adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the zero-sequence electric field features, zero-sequence current features, and numerical feature information of the corresponding segment based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result of each line node based on the auxiliary features. The fault location unit is used to determine the section where a single-phase grounding fault is located based on the fault location classification results of each line node and the topological connection relationship of the distribution network.
[0019] A third aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the single-phase ground fault location method based on an adaptive attention neural network as described in the first aspect above.
[0020] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps of the single-phase ground fault location method based on an adaptive attention neural network as described in the first aspect above.
[0021] The fifth aspect of the present invention provides a computer program product or computer program.
[0022] This invention provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the single-phase ground fault location method based on an adaptive attention neural network as described in the first aspect above.
[0023] Compared with the prior art, the beneficial effects of the present invention are: This invention establishes a joint characterization model of zero-sequence electric field and zero-sequence current by preprocessing, segmenting, and constructing zero-sequence features on the fault waveform; it automatically identifies key time segments in the fault waveform using an adaptive attention mechanism to effectively extract key fault information; it integrates deep features and numerical features by combining auxiliary features such as three-phase current imbalance, variance, root mean square value of zero-sequence current, and similarity between zero-sequence electric field and zero-sequence current; it uses a classification model to distinguish nodes inside and outside the fault interval, and combines the distribution network topology to locate single-phase grounding faults.
[0024] This invention eliminates the need for manually designing complex fault features, enabling the direct learning of effective features from fault data. Furthermore, it utilizes joint training with simulated and real-world data to enhance the model's generalization ability and robustness. Even under conditions of high-resistance grounding and unclear fault features, it can still achieve accurate fault location, offering advantages such as high positioning accuracy, high computational efficiency, and strong engineering adaptability. Attached Figure Description
[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0026] Figure 1This is a flowchart illustrating a single-phase grounding fault location method based on an adaptive attention neural network, as shown in an embodiment of the present invention. Figure 2 This is a schematic diagram of the single-phase grounding fault location system architecture shown in an embodiment of the present invention; Figure 3 This is an overall architecture diagram of an adaptive attention mechanism neural network for classifying fault waveforms, as shown in an embodiment of the present invention. Figure 4 This is a schematic diagram of the adaptive attention mechanism structure shown in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating auxiliary feature construction and fully connected classification in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the determination of fault zones based on node classification results, as shown in an embodiment of the present invention. Figure 7 This is a structural diagram of a single-phase grounding fault location system based on an adaptive attention neural network, as shown in an embodiment of the present invention. Figure 8 This is a structural diagram of a computer device shown in an embodiment of the present invention. Detailed Implementation
[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0028] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0029] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0030] As described in the background section, in order to solve the technical problem of how to automatically extract key features from fault waveforms and improve the positioning accuracy and generalization ability of the model in complex operating environments, this invention provides a single-phase grounding fault location method and system based on an adaptive attention neural network. The following describes the solution of this invention in detail through several embodiments.
[0031] Figure 1 This is a flowchart illustrating the single-phase grounding fault location method based on an adaptive attention mechanism neural network in Embodiment 1 of the present invention; see also... Figure 1 The method includes: The fault waveform data of each line node during the fault occurrence period are acquired, and then standardized preprocessing, waveform segmentation and zero-sequence synthesis are performed in sequence to obtain the zero-sequence electric field data and zero-sequence current data of each segment, and the numerical characteristic information of each segment is recorded; the numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. The zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment are input into an adaptive attention neural network. This adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the corresponding zero-sequence electric field features, zero-sequence current features, and numerical feature information based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result for each line node based on the auxiliary features. Based on the fault location classification results of each line node, and combined with the distribution network topology, the section where the single-phase grounding fault is located is determined.
[0032] This invention achieves accurate identification of single-phase grounding fault sections by preprocessing the fault waveform, segmenting the waveform, constructing zero-sequence features, extracting adaptive attention features, and performing multi-feature fusion analysis, thereby improving fault location accuracy and reliability in engineering applications.
[0033] Figure 2 This is a schematic diagram of the single-phase grounding fault location system architecture shown in an embodiment of the present invention; Figure 2 In the diagram, 1 to 9 represent collection units 1 to 9, respectively. The single-phase grounding fault location system for the distribution network includes a fault indicator and a distribution automation master station. The fault indicator includes a data acquisition unit and a collection unit. The data acquisition unit is installed on the three-phase lines of the distribution network and is used to trigger waveform recording when a single-phase grounding fault occurs, acquiring three-phase electric field data and three-phase current data during the fault occurrence period. The collection unit is used to synchronize and summarize the data acquired by the data acquisition unit and upload the fault waveform data of the corresponding line node to the distribution automation master station.
[0034] When a single-phase ground fault occurs in a distribution network, fault indicators located at different line nodes will collect different three-phase electric field waveforms and three-phase current waveforms. For nodes located on the power supply side of the fault point, the fault waveform characteristics differ from those of nodes located on the load side of the fault point or on non-faulty branches.
[0035] Therefore, this embodiment transforms the single-phase ground fault location problem into a node classification problem, that is, determining whether the node where each fault indicator is located is inside or outside the fault interval. Figure 3 This is an overall architecture diagram of an adaptive attention mechanism neural network for classifying fault waveforms, as shown in an embodiment of the present invention; see reference. Figure 3 .
[0036] It should be noted that the output of the adaptive attention neural network is not the direct location of the fault point, but rather the classification result of the fault interval for the corresponding node. After obtaining the classification results of multiple nodes, the distribution automation master station also needs to combine the distribution network topology connections to determine the adjacent node segments where the fault interval is located.
[0037] The fault waveform data uploaded by each line node includes three-phase electric field data and three-phase current data. Since the sampling rates of different fault indicators, on-site load conditions, and environmental interference may vary, this embodiment first performs standardization preprocessing on the fault waveform data to improve the consistency and comparability of the model input data. This standardization preprocessing includes uniform sampling rate processing, removal of load current, and removal of DC components.
[0038] In uniform sampling rate processing, let the original discrete fault waveform be... Perform a Fourier transform on it to obtain the frequency domain sequence. , where n represents the time index and k represents the frequency index, its expression is:
[0039] Furthermore, for frequency domain sequences Perform frequency domain interpolation to obtain the interpolated frequency domain sequence. .
[0040]
[0041] Its function is to convert frequency domain data at different sampling rates to a uniform number of frequency points. Then, for... Perform an inverse Fourier transform to obtain the interpolated time-domain signal. Its expression is:
[0042] Where M represents the number of frequency points after interpolation.
[0043] Then, the interpolated time-domain signal is sampled at equal intervals according to the preset downsampling factor D to obtain the downsampled fault waveform. Its expression is:
[0044] Through the above processing, fault waveforms acquired at different sampling rates can be uniformly converted into standard waveform data with the same sampling rate and the same data length.
[0045] In the process of removing load current, the waveform during the normal operating cycle before the fault occurs is selected as the load waveform reference. Let the load waveform be... ,but:
[0046] Based on the aforementioned load waveform reference, the downsampled fault waveform is periodically deducted for load components to obtain the deloaded fault waveform. Its expression is:
[0047] By removing load current processing, the impact of different load levels on fault feature extraction can be reduced.
[0048] In the process of removing the DC component, the fault waveform after load removal... High-pass filtering is performed to remove DC bias and low-frequency drift components, resulting in pre-processed three-phase electric field waveforms and three-phase current waveforms.
[0049] After standardization preprocessing, the preprocessed three-phase electric field waveforms and three-phase current waveforms are segmented. During segmentation, the three-phase electric field data and three-phase current data adopt the same segmentation rules to ensure that the electric field information and current information within the same segment correspond in time. Each fault waveform is divided into multiple waveform segments, and each waveform segment contains three-phase electric field data and three-phase current data. Preferably, the three-phase electric field data and three-phase current data are each segmented into 14 time segments, and each time segment contains three-phase data and a preset number of sampling points.
[0050] Furthermore, zero-sequence electric field data in each waveform segment is synthesized to obtain zero-sequence electric field data; zero-sequence current data in each waveform segment is synthesized to obtain zero-sequence current data.
[0051] Zero-sequence synthesis can highlight the common changing characteristics of the three-phase system when a single-phase ground fault occurs, and reduce the impact of redundant information in the original three-phase data on model learning.
[0052] Simultaneously, the root mean square values of the three-phase current and the root mean square value of the zero-sequence current within each segment are recorded as numerical information for subsequent auxiliary feature construction. This numerical information is used to characterize the changes in current amplitude within different time segments and, together with subsequent deep features, is used for node classification.
[0053] The zero-sequence electric field data, zero-sequence current data, and numerical feature information are input into the trained adaptive attention mechanism neural network model. Figure 4 This is a schematic diagram of the adaptive attention mechanism structure shown in an embodiment of the present invention; see reference. Figure 4 The adaptive attention mechanism neural network model includes a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module.
[0054] The convolutional feature extraction module is used to perform multi-layer convolutional feature extraction on the segmented zero-sequence electric field data and zero-sequence current data, respectively. The multi-layer convolutional structure includes several convolutional layers and activation function layers. The feature maps output from the shallow convolutional layers are processed by the activation function and then input into the next convolutional layer. Preferably, the activation function uses... The function, whose expression is:
[0055] Here, α is the slope parameter of the negative half-axis. Through multi-layer convolution processing, the zero-sequence electric field characteristics and zero-sequence current characteristics can be obtained separately.
[0056] like Figure 4 As shown, the adaptive attention module is used to identify key fault segments requiring focused attention from multiple waveform segments. First, the zero-sequence electric field characteristics are... Inputting the fully connected layer yields the electric field feature embedding. Its expression is:
[0057] To enable the model to distinguish the temporal positions of different segments in the fault waveform, this embodiment introduces position encoding. Position encoding is constructed using sine and cosine functions, and its expression is:
[0058] in, Indicates the first pos The fault waveform segment is in the 2nd i Location encoding values in each dimension Indicates the first pos The fault waveform segment is in the 2nd i +1 positional encoding values in the dimension; Indicates the segment position, and i represents the dimension index. Indicates the location encoding dimension.
[0059] The electric field feature embedding is added to the time location code and then normalized to obtain the fused feature after incorporating location information. Its expression is:
[0060] Then, based on the fusion characteristics Calculate the attention matrix Its expression is:
[0061] The attention matrix is processed using softmax to obtain the normalized attention matrix. Then extract The main diagonal elements are used to obtain the specific vectors corresponding to each segment. Further After performing softmax processing, the attention probability vector SDFSAM for each segment is obtained, and its expression is:
[0062] In SDFSAM, each element corresponds to the attention probability of a fault waveform segment, and the sum of the attention probabilities of all segments is 1. The higher the attention probability, the more likely the segment is to contain fault features that contribute to fault interval discrimination. Therefore, the segment with the highest attention probability is identified as the critical fault segment.
[0063] Based on the key fault segment location output by the adaptive attention module, the feature selection module selects the zero-sequence electric field feature, zero-sequence current feature, three-phase current root mean square value, and zero-sequence current root mean square value of the corresponding segment as input for subsequent auxiliary feature fusion and classification judgment.
[0064] like Figure 5 As shown, the auxiliary feature fusion module is used to construct the three-phase current asymmetry, three-phase current variance, zero-sequence current scaling value, and similarity between zero-sequence electric field features and zero-sequence current features.
[0065] First, based on the root mean square value of the three-phase current Calculate their sum:
[0066] The degree of three-phase current asymmetry is expressed as: Furthermore, calculate the maximum value among the root mean square values of the three-phase currents. :
[0067] The variance of the three-phase current is expressed as:
[0068] For the root mean square value of zero-sequence current After scaling, the result is obtained Its expression is:
[0069] Simultaneously, the zero-sequence electric field characteristics are calculated. With zero-sequence current characteristics Similarity between them:
[0070] The three-phase current asymmetry, three-phase current variance, zero-sequence current scaling value, and similarity between the zero-sequence electric field and zero-sequence current are input into the fully connected classification module. The fully connected classification module outputs the probability that the corresponding line node is located within the fault interval and the probability that it is located outside the fault interval. If the probability of being located within the fault interval is greater than the probability of being located outside the fault interval, the node is determined to be located within the fault interval; otherwise, the node is determined to be located outside the fault interval.
[0071] like Figure 6 As shown, after acquiring the classification results of multiple line nodes, the distribution automation master station determines the section where a single-phase ground fault occurs by combining the distribution network topology. Specifically, the classification results of adjacent nodes are analyzed sequentially along the power supply direction of the feeder. When a node closer to the power source is determined to be within the fault zone (Inside the Fault, ITF), while its adjacent node farther from the power source is determined to be outside the fault zone (Outside the Fault, OTF), the line segment between these two adjacent nodes is determined to be the section where the single-phase ground fault occurs. For example, when nodes 1 and 2 are determined to be within the fault zone, and node 3 is determined to be outside the fault zone, the line segment between nodes 2 and 3 is determined to be the section where the single-phase ground fault occurs.
[0072] In this embodiment, the adaptive attention neural network model is jointly trained using simulated fault data and actual fault acquisition data. Simulated fault data provides typical fault characteristics under different fault conditions, enabling the model to learn effective feature extraction methods from fault waveforms. Actual fault acquisition data provides waveform characteristics under conditions of on-site noise, equipment sampling errors, and complex operating conditions, allowing the model to learn to ignore noise interference and improve its engineering adaptability. Through joint training with simulated and actual data, the model can learn common fault characteristics from both types of data, improving its generalization ability and robustness.
[0073] During training, the adaptive attention mechanism neural network model is trained using mini-batch gradient descent, and the model parameters are updated using an optimizer. The model loss function includes a position loss function and a classification loss function, with the joint loss function being:
[0074] in, This is used to constrain the model to focus on key fault segments in the fault waveform. Used to constrain the node classification results output by the model.
[0075] The location loss function is expressed as:
[0076] in, Indicates training rounds, This indicates the location of the critical fault segment predicted by the model. Indicates the location label of a critical fault segment. This represents the cross-entropy function. By setting a position loss that decays with each training epoch, some segmentation position information can be provided to the model in the early stages of training, while the position constraint is reduced in the later stages of training, enabling the model to adaptively focus on the correct critical fault segments.
[0077] The classification loss function is expressed as:
[0078] in, This represents the node classification result predicted by the model. This represents the true classification label of the node. By minimizing the joint loss function, the parameters of the adaptive attention mechanism neural network model are updated to obtain the trained fault interval discrimination model.
[0079] The above combination Figure 1 The single-phase grounding fault location method based on adaptive attention neural network provided in the embodiments of the present invention has been described in detail. Next, the single-phase grounding fault location system based on adaptive attention neural network provided in the embodiments of the present invention will be described in conjunction with the accompanying drawings.
[0080] Figure 7 This is a schematic diagram of the structure of a single-phase ground fault location system based on an adaptive attention neural network, as shown in an embodiment of the present invention. Figure 7 The system described in this invention includes: The data processing unit is used to acquire fault waveform data of each line node during the fault occurrence, and sequentially perform standardization preprocessing, waveform segmentation and zero-sequence synthesis to obtain zero-sequence electric field data and zero-sequence current data of each segment, and record the numerical characteristic information within each segment; the numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. An adaptive attention neural network unit is used to input the zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment into the adaptive attention neural network. The adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the zero-sequence electric field features, zero-sequence current features, and numerical feature information of the corresponding segment based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result of each line node based on the auxiliary features. The fault location unit is used to determine the section where a single-phase grounding fault is located based on the fault location classification results of each line node and the topological connection relationship of the distribution network.
[0081] In some embodiments, the data processing unit is specifically used for: performing uniform sampling rate processing, load current removal processing, and DC component removal processing on the fault waveform data to obtain preprocessed fault waveform data; performing segmented processing on the preprocessed fault waveform data, with each segment containing three-phase electric field data and three-phase current data; performing zero-sequence synthesis on the three-phase electric field data in each segment to obtain zero-sequence electric field data; and performing zero-sequence synthesis on the three-phase current data in each segment to obtain zero-sequence current data.
[0082] In some embodiments, the adaptive attention module embeds zero-sequence electric field features to obtain electric field feature embeddings; introduces time position coding to distinguish the time positions of different segments in the fault waveform data; adds the electric field feature embeddings and time position codings and performs normalization processing to obtain fused features; calculates the attention matrix based on the fused features, and after normalization, extracts the main diagonal elements to obtain the specific vectors corresponding to each segment, and performs softmax processing to obtain the attention probability vectors of each segment; based on the attention probability vectors of each segment, the segment with the highest attention probability is determined, which is the location of the critical fault segment.
[0083] In some embodiments, the auxiliary feature fusion module is used to calculate the sum of the root mean square values of the three-phase currents based on the root mean square values of the three-phase currents; calculate the three-phase current asymmetry based on the sum of the root mean square values of the three-phase currents; select the maximum value among the root mean square values of the three-phase currents to calculate the three-phase current variance; calculate the zero-sequence current scaling value based on the root mean square value of the zero-sequence current; and calculate the similarity between the zero-sequence electric field feature and the zero-sequence current feature based on the zero-sequence electric field feature and the zero-sequence current feature. The auxiliary features include: three-phase current asymmetry, three-phase current variance, zero-sequence current scaling value, and the similarity between the zero-sequence electric field feature and the zero-sequence current feature.
[0084] In some embodiments, the fully connected classification module outputs the probability that the corresponding line node is located within the fault interval and the probability that it is located outside the fault interval based on auxiliary features. If the probability that it is located within the fault interval is greater than the probability that it is located outside the fault interval, then the node is determined to be located within the fault interval; otherwise, the node is determined to be located outside the fault interval.
[0085] In some embodiments, the adaptive attention neural network updates network parameters using a joint loss function during training, and the joint loss function is expressed by the following formula:
[0086]
[0087]
[0088] Where Loss represents the joint loss function; Used to constrain the network to focus on key fault segments in the fault waveform; Used to constrain the node classification results output by the network; Indicates training rounds, This indicates the location of the critical fault segment predicted by the model. Indicates the location label of a critical fault segment. Represents the cross-entropy function; This represents the node classification result predicted by the model. This represents the actual category label of the node.
[0089] According to embodiments of the present invention, the single-phase ground fault location system based on an adaptive attention neural network can correspond to the execution of the method described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the single-phase ground fault location system based on the adaptive attention neural network are respectively for implementing Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.
[0090] See Figure 8The diagram shows the structure of a computer device, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium are connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory. The computer-readable storage medium stores computer programs, including program instructions, and the processor executes the program instructions stored in the computer-readable storage medium. The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device, adapted to implement one or more instructions, specifically adapted to load and execute one or more instructions to implement the corresponding steps in the embodiment of the single-phase ground fault location method based on an adaptive attention neural network.
[0091] This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device.
[0092] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.
[0093] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above embodiment of the single-phase grounding fault location method based on adaptive attention neural network.
[0094] This embodiment provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps in the above-described embodiment of the single-phase ground fault location method based on an adaptive attention neural network.
[0095] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0096] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0099] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A single-phase ground fault location method based on an adaptive attention neural network, characterized in that, include: The fault waveform data of each line node during the fault occurrence period are acquired, and then standardized preprocessing, waveform segmentation and zero-sequence synthesis are performed in sequence to obtain the zero-sequence electric field data and zero-sequence current data of each segment, and the numerical characteristic information of each segment is recorded; the numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. The zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment are input into an adaptive attention neural network. This adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the corresponding zero-sequence electric field features, zero-sequence current features, and numerical feature information based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result for each line node based on the auxiliary features. Based on the fault location classification results of each line node, and combined with the distribution network topology, the section where the single-phase grounding fault is located is determined.
2. The single-phase grounding fault location method based on adaptive attention neural network according to claim 1, characterized in that, The method involves acquiring fault waveform data from each line node during the fault occurrence period, performing standardized preprocessing, waveform segmentation, and zero-sequence synthesis sequentially to obtain zero-sequence electric field data and zero-sequence current data for each segment; the method includes: The fault waveform data is processed by uniform sampling rate, load current removal, and DC component removal to obtain preprocessed fault waveform data. The preprocessed fault waveform data is segmented, and each segment contains three-phase electric field data and three-phase current data; Zero-sequence electric field data is obtained by performing zero-sequence synthesis on the three-phase electric field data in each segment; zero-sequence current data is obtained by performing zero-sequence synthesis on the three-phase current data in each segment.
3. The single-phase grounding fault location method based on adaptive attention neural network according to claim 1, characterized in that, The adaptive attention module is used to perform feature embedding, position encoding, and attention processing on the zero-sequence electric field features, and to determine the location of the critical fault segment based on the attention probability of each segment. This includes: the adaptive attention module embedding the zero-sequence electric field features to obtain electric field feature embeddings; introducing time position encoding to distinguish the time positions of different segments in the fault waveform data; adding the electric field feature embeddings and time position encodings and performing normalization processing to obtain fused features; calculating the attention matrix based on the fused features, and after normalization, extracting the main diagonal elements to obtain the specific vectors corresponding to each segment, and performing softmax processing to obtain the attention probability vectors of each segment; and determining the segment with the highest attention probability based on the attention probability vectors of each segment, which is the location of the critical fault segment.
4. The single-phase grounding fault location method based on adaptive attention neural network according to claim 1, characterized in that, The auxiliary feature fusion module is used to obtain auxiliary features based on the data output by the feature selection module; including: Calculate the sum of the root mean square values of the three-phase currents based on their root mean square values; calculate the three-phase current asymmetry based on the sum of their root mean square values. Select the maximum value among the root mean square values of the three-phase currents and calculate the variance of the three-phase currents; Calculate the zero-sequence current scaling value based on the root mean square value of the zero-sequence current; Based on the zero-sequence electric field characteristics and the zero-sequence current characteristics, calculate the similarity between the zero-sequence electric field characteristics and the zero-sequence current characteristics; The auxiliary features include: three-phase current asymmetry, three-phase current variance, zero-sequence current scaling value, and similarity between zero-sequence electric field features and zero-sequence current features.
5. The single-phase grounding fault location method based on adaptive attention neural network according to claim 1, characterized in that, The fully connected classification module is used to determine the fault location classification result of each line node based on auxiliary features, including: the fully connected classification module outputs the probability that the corresponding line node is located within the fault interval and the probability that it is located outside the fault interval based on the auxiliary features. If the probability that it is located within the fault interval is greater than the probability that it is located outside the fault interval, then the node is determined to be located within the fault interval; otherwise, the node is determined to be located outside the fault interval.
6. The single-phase grounding fault location method based on adaptive attention neural network according to claim 1, characterized in that, The adaptive attention neural network updates its network parameters during training using a joint loss function, which is expressed by the following formula: Where Loss represents the joint loss function; Used to constrain the network to focus on key fault segments in the fault waveform; Used to constrain the node classification results output by the network; Indicates training rounds, This indicates the location of the critical fault segment predicted by the model. Indicates the location label of a critical fault segment. Represents the cross-entropy function; This represents the node classification result predicted by the model. This represents the actual category label of the node.
7. A single-phase ground fault location system based on an adaptive attention neural network, characterized in that, include: The data processing unit is used to acquire fault waveform data of each line node during the fault occurrence, and sequentially perform standardization preprocessing, waveform segmentation and zero-sequence synthesis to obtain zero-sequence electric field data and zero-sequence current data of each segment, and record the numerical characteristic information within each segment. The numerical characteristic information includes the root mean square value of the three-phase current and the root mean square value of the zero-sequence current. An adaptive attention neural network unit is used to input the zero-sequence electric field data, zero-sequence current data, and numerical feature information of each segment into the adaptive attention neural network. The adaptive attention neural network includes: a convolutional feature extraction module, an adaptive attention module, a feature selection module, an auxiliary feature fusion module, and a fully connected classification module. The convolutional feature extraction module extracts features from the zero-sequence electric field data and zero-sequence current data to obtain zero-sequence electric field features and zero-sequence current features. The adaptive attention module performs feature embedding, position encoding, and attention processing on the zero-sequence electric field features, determining the location of key fault segments based on the attention probability of each segment. The feature selection module selects the zero-sequence electric field features, zero-sequence current features, and numerical feature information of the corresponding segment based on the location of the key fault segments. The auxiliary feature fusion module obtains auxiliary features based on the data output by the feature selection module. The fully connected classification module determines the fault location classification result of each line node based on the auxiliary features. The fault location unit is used to determine the section where a single-phase grounding fault is located based on the fault location classification results of each line node and the topological connection relationship of the distribution network.
8. A computer device, characterized in that, A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the single-phase ground fault location method based on an adaptive attention neural network as described in any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the steps of the single-phase ground fault location method based on an adaptive attention neural network as described in any one of claims 1-6.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the single-phase ground fault location method based on an adaptive attention neural network as described in any one of claims 1-6.