Fault positioning method, system and device based on synchronous waveform data processing and storage medium

Abstract

The invention discloses a fault positioning method, system and device based on synchronous waveform data processing and a storage medium, and the method comprises the steps: obtaining the synchronous waveform data of each node of a power distribution network, and constructing a power distribution network space-time adjacency matrix based on the synchronous waveform data and the topological relation of the power distribution network; presetting a power distribution network fault positioning model, taking the space-time adjacency matrix as input data of the power distribution network fault positioning model, and preprocessing the input data; and based on the preprocessed data, positioning a power distribution network fault line through a power distribution network fault positioning model. According to the invention, node space-time information collected by the synchronous waveform measurement device can be synchronously processed; a learnable mask matrix is introduced to carry out weighted optimization on the adjacency relation between nodes, and the feature extraction capability and the information aggregation depth are improved in combination with skip connection and an attention mechanism, so that the adaptability and robustness of the model under the condition of sparse measurement points are effectively enhanced, and the accurate judgment of the fault position is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network fault location, and particularly relates to a fault location method, system and device based on synchronous waveform data processing and a storage medium. BACKGROUND

[0002] As an important part of the power system, the fault location accuracy of the power distribution network directly affects the power supply reliability. It is of great practical significance to realize the rapid and accurate positioning of short-circuit faults in the power distribution network. Traditional positioning technologies mainly include impedance method, traveling wave method and matrix method, etc. However, in the context of the gradual access of distributed energy resources (DER), the structure of the power distribution network is becoming increasingly complex, and the traditional methods face significant challenges in extracting fault characteristic signals and designing criteria, making it difficult to adapt to the changing conditions under the active power distribution environment, and the accuracy has also decreased. In order to overcome these problems, the phasor measurement unit (PMU) technology has been widely introduced into the field of power distribution network fault detection in recent years. PMU has high time synchronization accuracy and can synchronously collect phasor information of voltage and current at multiple nodes, thereby effectively capturing the dynamic response at the fault instant. The research around PMU is constantly developing, and a series of fault location methods such as phasor mutation trend analysis and power direction discrimination have been proposed, which has become one of the hot directions in this field. In addition, with the continuous improvement of the digitalization and automation level of the power distribution network, the accumulation of massive operation data provides support for the application of artificial intelligence technology. Intelligent algorithms represented by deep learning can automatically extract potential feature patterns from historical data, realizing efficient fault diagnosis and positioning.

[0003] Several studies have explored combining PMU measurement data with neural networks, graph learning, and other models to improve the accuracy and generalization ability of fault identification. However, it is worth noting that most current PMU-based intelligent methods still rely on low-frequency phasor information and lack the ability to model short-term, high-frequency waveform features. Furthermore, they fail to fully consider the current direction changes and fault characteristic uncertainties caused by high-proportion distributed source injection, resulting in limited feature extraction capabilities and susceptibility to interference in fault location. In recent years, with the development of Waveform Measurement Units (WMUs), researchers can obtain higher-precision and more complete voltage and current waveform data. Compared to the steady-state phasors provided by PMUs, WMU waveform information is richer and can reflect finer-grained changes in operating states. Fault location research based on this type of waveform data helps to achieve faster and more accurate identification and segment location of faults in complex active distribution networks, and improves the model's adaptability to scenarios with high distributed source penetration, possessing significant research and engineering application value. Therefore, there is an urgent need for a fault location method that can accurately extract fault feature information in complex active distribution networks. Summary of the Invention

[0004] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a fault location method, system, device, and storage medium based on synchronous waveform data processing to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide a fault location method based on synchronous waveform data processing, comprising: acquiring synchronous waveform data of each node in the distribution network, and constructing a spatiotemporal adjacency matrix of the distribution network based on the synchronous waveform data and the topology of the distribution network;

[0007] A pre-defined distribution network fault location model is established, and the spatiotemporal adjacency matrix is ​​used as the input data for the distribution network fault location model. The input data is then pre-processed.

[0008] Based on the preprocessed data, the faulty lines in the distribution network are located using the distribution network fault location model.

[0009] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, the construction of the distribution network spatiotemporal adjacency matrix includes:

[0010] The distribution network is modeled as an undirected graph structure, where nodes represent substations, load nodes, and distributed power source access points; edges represent electrical connections.

[0011] Synchronous waveform measurement devices are deployed at key nodes to collect three-phase voltage and current waveform data, forming node features. The measurement value of each node at any time is represented as a vector, and the traveling wave data collected by the node are stacked. All node features are combined into a three-dimensional matrix as the data input for the graph structure.

[0012] Based on the connection relationship of distribution network nodes and the continuity of the same node in the time dimension, a spatial adjacency matrix and a temporal adjacency matrix are constructed. The spatial adjacency matrix and the temporal adjacency matrix are then fused to generate a spatiotemporal adjacency matrix, thereby synchronously modeling the spatiotemporal information of the distribution network.

[0013] The beneficial effect of this preferred technical solution is that by constructing a spatiotemporal adjacency matrix and designing a spatiotemporal fusion graph convolutional structure, it is possible to simultaneously extract the evolution information of nodes in the time dimension and spatial topology, thereby reflecting the fault occurrence and propagation characteristics more comprehensively and accurately, and improving the positioning accuracy.

[0014] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, the method includes: using the spatiotemporal adjacency matrix as input data for the distribution network fault location model, and preprocessing the input data, which includes: inputting the spatiotemporal adjacency matrix as input data into the fully connected layer of the distribution network fault location model for dimensionality upscaling, and inputting it into the batch normalization layer for batch normalization.

[0015] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, the method further includes: adjusting the adjacency strength of the distribution network nodes using a learnable mask matrix based on the normalized input data to obtain the connection weights between different nodes, and synchronously aggregating the temporal and spatial information of the distribution network.

[0016] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, it further includes: introducing a skip connection and attention mechanism in the graph convolution extraction of the spatiotemporal signal, and using global max pooling + average pooling to obtain key fault features.

[0017] The beneficial effects of this preferred technical solution are that it introduces a joint design of skip connections and attention mechanism into the graph convolutional network structure. Skip connections allow the model to retain both low-order and high-order feature representations at the same time, expanding the receptive field. The attention mechanism enables the model to adaptively assign higher weights to key neighbors when feature aggregation, thereby improving the accuracy and robustness of fault feature representation and enhancing the model's adaptability to complex working conditions.

[0018] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, the faulty line in the distribution network is located based on the preprocessed data using a distribution network fault location model. This includes: using a multi-scale graph convolutional structure to cover the spatial range of the sparse graph, retaining feature representations of different levels through skip connections, and using an additive attention mechanism to weightedly fuse the convolutional outputs of graphs of different orders to generate a node embedding representation for identifying the fault area, represented as follows:

[0019]

[0020] Among them, F (i) It is the graph convolution output of the i-th layer, α i These are the importance weights of the output of each layer learned by the attention mechanism, satisfying ∑ i α i =1.

[0021] As a preferred embodiment of the fault location method based on synchronous waveform data processing described in this invention, it further includes: segmenting the waveform sequence into graphs using a sliding window approach, extracting features from several time periods in parallel, and assuming the total number of sampling points is T, the sliding window size is s, and the step size is a, the total number of spatiotemporal graph convolutional module groups k generated is calculated as follows:

[0022]

[0023] K spatiotemporal feature modules are constructed in parallel, and their output features after attention fusion are spliced ​​together to express the spatiotemporal fusion features of the distribution network, enabling spatial perception and localization of fault impacts under sparse graph data.

[0024] Secondly, the present invention provides a fault location system based on synchronous waveform data processing, comprising: a spatiotemporal adjacency matrix construction module, used to acquire synchronous waveform data of each node in the distribution network, and construct a spatiotemporal adjacency matrix of the distribution network based on the synchronous waveform data and the topology of the distribution network;

[0025] The data preprocessing module is used to preset the distribution network fault location model, use the spatiotemporal adjacency matrix as the input data of the distribution network fault location model, and preprocess the input data.

[0026] The fault location module is used to locate faulty lines in the distribution network based on the preprocessed data and the distribution network fault location model.

[0027] Thirdly, the present invention provides an electronic device, comprising:

[0028] Memory and processor;

[0029] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the fault location method based on synchronous waveform data processing.

[0030] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the fault location method based on synchronous waveform data processing.

[0031] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a spatiotemporal adjacency matrix and designing a spatiotemporal fusion graph convolutional structure, this invention can simultaneously extract the evolutionary information of nodes in the temporal dimension and spatial topology, thereby reflecting the fault occurrence and propagation characteristics more comprehensively and accurately, and improving the localization accuracy. Furthermore, by introducing a learnable mask matrix to adjust the weighted basic adjacency relationships, the model can automatically learn the true physical influence between nodes through data training, achieving a more reasonable and effective feature aggregation and propagation process. In the graph convolutional network structure, a joint design of skip connections and attention mechanisms is introduced. Skip connections allow the model to retain both low-order and high-order feature representations simultaneously, expanding the receptive field. The attention mechanism enables the model to adaptively assign higher weights to key neighbors during feature aggregation, thereby improving the accuracy and robustness of fault feature representation and enhancing the model's adaptability to complex working conditions. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0033] Figure 1 This is a schematic diagram illustrating a fault location method, system, device, and storage medium based on synchronous waveform data processing according to an embodiment of the present invention.

[0034] Figure 2 This is a schematic flowchart illustrating a fault location method, system, device, and storage medium based on synchronous waveform data processing according to an embodiment of the present invention.

[0035] Figure 3 This is a schematic diagram of the spatiotemporal adjacency matrix of a fault location method, system, device and storage medium based on synchronous waveform data processing according to an embodiment of the present invention.

[0036] Figure 4This is a simulated power distribution network topology diagram illustrating a fault location method, system, device, and storage medium based on synchronous waveform data processing according to an embodiment of the present invention. Detailed Implementation

[0037] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0038] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides a fault location method based on synchronous waveform data processing, including:

[0039] S100: Obtain the synchronization waveform data of each node in the distribution network, and construct the spatiotemporal adjacency matrix of the distribution network based on the synchronization waveform data and the topology of the distribution network.

[0040] S200: Preset distribution network fault location model, use the spatiotemporal adjacency matrix as input data for the distribution network fault location model, and preprocess the input data;

[0041] S300: Based on the preprocessed data, the faulty lines in the distribution network are located using the distribution network fault location model.

[0042] It should be noted that traditional graph convolution methods can only model the spatial connectivity relationships in the power grid topology, making it difficult to capture the dynamic changes in faults over time. However, actual fault processes exhibit significant temporal abrupt changes. Therefore, this application constructs a spatiotemporal adjacency matrix and designs a spatiotemporal fusion graph convolution structure, which can simultaneously extract the evolutionary information of nodes in the temporal dimension and spatial topology, thereby more comprehensively and accurately reflecting the fault occurrence and propagation characteristics and improving location accuracy. Traditional adjacency matrices only reflect whether there is a connection between nodes, failing to distinguish the strength of the connection and making it difficult to reflect the physical coupling differences caused by factors such as line impedance, distance, and voltage level in actual distribution networks. To overcome this problem, this application introduces a learnable mask matrix to weight and adjust the basic adjacency relationships, enabling the model to automatically learn the true physical influence between nodes through data training, achieving a more reasonable and effective feature aggregation and propagation process. To improve the model's location capability in situations where the distribution of measurement points is sparse or some nodes are missing, this application introduces a joint design of skip connections and attention mechanisms in the graph convolution network structure. Skip-layer connections allow the model to retain both low-order and high-order feature representations, expanding the receptive field; attention mechanisms enable the model to adaptively assign higher weights to key neighbors during feature aggregation, thereby improving the accuracy and robustness of fault feature representation and enhancing the model's adaptability to complex working conditions.

[0043] In this embodiment of the application, the step S100 of constructing the spatiotemporal adjacency matrix of the distribution network includes:

[0044] The distribution network is modeled as an undirected graph structure, where nodes represent substations, load nodes, and distributed power source access points; edges represent electrical connections.

[0045] Synchronous waveform measurement devices are deployed at key nodes to collect three-phase voltage and current waveform data, forming node features. The measurement value of each node at any time is represented as a vector, and the traveling wave data collected by the node are stacked. All node features are combined into a three-dimensional matrix as the data input for the graph structure.

[0046] Based on the connection relationship of distribution network nodes and the continuity of the same node in the time dimension, a spatial adjacency matrix and a temporal adjacency matrix are constructed. The spatial adjacency matrix and the temporal adjacency matrix are then fused to generate a spatiotemporal adjacency matrix, thereby synchronously modeling the spatiotemporal information of the distribution network.

[0047] Specifically, the spatiotemporal graph structure data of the distribution network is constructed, and the distribution network is modeled as an undirected graph structure, where nodes represent substations, load nodes, distributed power supply access points, etc., and edges represent electrical connection relationships. WMUs are configured on selected key nodes to collect their three-phase voltage and current waveform data.

[0048] Using the three-phase voltage and current waveform data measured at the nodes as node features, a graph-structured data structure reflecting the distribution network's operating status is constructed. The data measured at a node at a certain moment can be represented by a vector A. n,t express.

[0049] A n,t =[U a U b U c ,I a ,I b ,I c ,]

[0050] Where n is the node number; t is the sampling time; U a This is the voltage value of phase A at node I; a This represents the downstream current value of phase A at node A.

[0051] Matrix A contains only waveform feature data at all nodes of the distribution network. However, given the fluctuations in the output of distributed power sources and the complex and ever-changing topology of the new distribution network, relying solely on this waveform feature data cannot guarantee the accuracy of the data, thereby reducing the accuracy of subsequent fault location methods.

[0052] Stacking the traveling wave data acquired at this node can be represented as a 6×T matrix A. n A n =[A n,1 A n,2 ,...A n,t ,...,A n,T ]

[0053] Where T is the length of the time window.

[0054] In an optional embodiment, the topology of the distribution network, i.e., the connection relationships between nodes, is used as the modeling basis; a spatial adjacency matrix A is constructed based on the physical connection relationships of the power grid. S ∈R N×N Where N is the total number of nodes. If node i is connected to node j, then A S(i,j) =1 otherwise 0. This adjacency matrix clarifies the physical connections of each node in the power grid and can serve as a path structure for information aggregation in graph neural networks, used to extract spatial features between nodes and their neighbors, such as the geographical distribution patterns of voltage and current.

[0055] Traditional graph neural networks (GNNs) extract features only from the spatial structure at a single moment, failing to fully explore the dynamic evolution of node states over time. Furthermore, common spatiotemporal graph neural networks (ST-GNNs) employ alternating graph convolution and temporal convolution to extract features. This approach suffers from low information transfer efficiency, weak temporal modeling capabilities, and significant information loss during the fusion of spatiotemporal information.

[0056] In an optional embodiment, this application proposes constructing a unified spatiotemporal adjacency matrix A. ST This is to achieve synchronous modeling and joint propagation of spatial structure and temporal evolution characteristics. The matrix consists of two parts: one is the aforementioned spatial adjacency matrix A. S Another type is the time adjacency matrix A. T A T It is an identity matrix that represents the continuity of a node's state in the time dimension. That is, in the time dimension, there is information connection between the same node in adjacent time steps. It is used to establish the convolutional propagation path in the node's time series.

[0057] A S and A T By combining according to specific rules, the final spatiotemporal adjacency matrix A can be formed. ST ∈R sN×sN , where s is the number of time steps considered. For example, when s = 4, A ST It can be viewed as a set of 4 main diagonal pieces (A). S ) and several diagonal pieces (for A) T A block-like matrix structure composed of ) such as Figure 1 As shown. Block A on the main diagonal. S Indicates the spatial connectivity within each time step, with A on the secondary diagonal. T This connects the states of the same node across different time steps, enabling feature propagation across time steps.

[0058] This construction method naturally integrates the physical topology information of the power grid and the state changes of nodes over time in the graph convolution operation, effectively improving the model's sensitivity to the dynamic evolution of faults and its feature extraction capability, thereby enhancing the accuracy and robustness of fault location. After constructing a spatiotemporal graph structure that reflects the spatial connection relationships and temporal evolution process between nodes in the distribution network, this structure is used as the input to the graph neural network model. The spatiotemporal graph convolution method can extract spatiotemporal fusion features that reflect the location of faults in the distribution network from this spatiotemporal graph structure. After a fault occurs, the impact gradually extends from the connection between the faulty node and the edge, and the range continues to expand. Different fault conditions produce different impacts, and graph convolution can fully map these impacts by aggregating information from neighboring nodes to update features.

[0059] Therefore, this application will, based on the constructed spatiotemporal graph structure, introduce a spatiotemporal fusion graph convolution mechanism, enabling the model to simultaneously perceive the spatial characteristics of its neighboring nodes and its own temporal change trends at each node, thereby accurately capturing the differences in node behavior and improving the ability to identify fault areas.

[0060] Graph convolution is a non-linear function, i.e.

[0061] H k+1 =f(H k A)

[0062] Among them, H k is the feature information vector of the k-th layer; k is the layer number of the graph convolution at this time; f is the path of information propagation in the graph convolution; A is the adjacency matrix of the spatiotemporal graph structure at this time.

[0063] Commonly used graph convolution calculation methods mainly involve spectral domain graph convolution. This method requires spectral decomposition, resulting in high computational cost, and requires retraining once the distribution network topology changes, exhibiting poor adaptability. Therefore, this application adopts spatial domain graph convolution, directly aggregating information on node neighbors. Its general form is as follows:

[0064] H k+1 =σ(AH l W+b)

[0065] Where A is the adjacency matrix, W is the weight matrix, σ is the activation function, and b considers additional bias with a certain margin.

[0066] It should be noted that, due to the inherent structure of the distribution network, in actual operation, faults can spread to other areas along the connections between nodes and edges. Furthermore, the degree of impact varies among different nodes, and the timing measurements of fault-related effects also differ. Traditional adjacency matrices A, which strictly use 0 or 1 values ​​to indicate whether nodes are connected, cannot reflect connection strength. This may lead to an overly averaged information transmission during graph convolution aggregation, failing to reflect the actual topological differences and physical conditions of the distribution network. Constructing weighted adjacency matrices directly using line impedance, distance, etc., presents the following problems: some line parameters in the distribution network are difficult to obtain accurately; a single physical quantity cannot fully characterize the coupling characteristics between nodes; and network topology and operating states under different scenarios can lead to parameter failure or generalization difficulties.

[0067] In an optional embodiment, this application introduces a learnable mask matrix to adjust the weights of the adjacency matrix. By guiding the model to automatically learn the effectiveness and importance of each edge during training, hidden physical parameter constraints and influence patterns between nodes are mined from a large number of samples, achieving dynamic optimization of adjacency relationships. The final adjacency matrix A used for information propagation in graph convolution is expressed as follows:

[0068] A = AST ⊙W m

[0069] Among them, A ST Based on the spatiotemporal fusion adjacency matrix constructed earlier, connections between nodes in the distribution network in spatial and temporal series are established. m The adjacency matrix is ​​obtained through continuous learning, where each element represents the weight of the corresponding edge. The adjacency matrix is ​​dynamically adjusted based on the learned weights. This structure enables graph convolution to simultaneously consider adjacency relationships and connection strength during information aggregation, accurately modeling the real information transmission process between nodes.

[0070] In this embodiment of the application, step S200, which uses the spatiotemporal adjacency matrix as input data for the distribution network fault location model and performs preprocessing on the input data, includes: inputting the spatiotemporal adjacency matrix as input data into the fully connected layer of the distribution network fault location model for dimensionality upscaling, and inputting it into the batch normalization layer for batch normalization.

[0071] In this embodiment of the application, step S200 further includes: adjusting the adjacency strength of distribution network nodes based on the normalized input data using a learnable mask matrix to obtain the connection weights between different nodes, and simultaneously aggregating the temporal and spatial information of the distribution network.

[0072] In this embodiment of the application, step S200 further includes: introducing a skip connection and attention mechanism in the graph convolution extraction of the spatiotemporal signal, and using global max pooling + average pooling to obtain key fault features.

[0073] It should be noted that, compared with traditional methods that rely solely on spatial or temporal information for graph convolution operations, the spatiotemporal fusion graph convolution strategy can integrate both time-series and spatial topology data in the same feature extraction process, comprehensively reflecting the structural relationships between nodes in the distribution network and their dynamic characteristics over time. Through fusion processing, the propagation process of electrical quantities in the network after a fault can be better simulated, avoiding the loss of key information, thereby effectively improving the accuracy of fault location and the model's generalization ability.

[0074] Although the deployment of measuring devices in the current distribution network is progressing steadily, achieving full node coverage remains difficult in the short term due to economic constraints and practical implementation conditions. Therefore, the graph data acquired by the network will continue to exhibit significant sparsity for a considerable period. In this situation, if the model relies solely on single-step graph convolution to aggregate neighbor information, the acquired features are limited to the first-order neighborhood, making it difficult to capture fault impact features across a wide area and affecting the extraction of global features. Therefore, it is necessary to design a graph neural network structure with greater information perception capabilities to improve the overall performance of the model under sparse measuring point conditions.

[0075] In an optional embodiment, this application introduces a skip-layer connection mechanism to replace the traditional layer-by-layer stacking graph convolution method. This mechanism allows direct connections to be established between non-adjacent graph convolutional layers, enabling the preservation and utilization of low-order features from shallow layers in the generation of high-level features, thereby achieving information fusion across different scales. This structure can effectively alleviate the gradient vanishing and feature oversmoothing problems commonly found in deep graph neural networks. For nodes with more connections, low-order features already possess strong expressive power, while for sparsely connected nodes, skip-layer connections help introduce a wider range of contextual information, enhancing the overall robustness of feature representation.

[0076] Building upon the skip-layer structure, an attention mechanism is further introduced to enhance the selectivity and targeting of feature fusion. The attention module can automatically evaluate the contribution of each level of features to the current node in both the feature and node dimensions, thereby achieving adaptive receptive field adjustment for different nodes and overcoming the limitation of uniform aggregation depth. By introducing a trainable scoring function, the model can dynamically allocate weights among multi-scale features, focusing on information regions that contribute more to fault detection.

[0077] In this embodiment of the application, step S300, based on the preprocessed data, locates the faulty line in the distribution network using a distribution network fault location model. This includes: using a multi-scale graph convolutional structure to cover the spatial range of the sparse graph, retaining feature representations of different levels through skip connections, and using an additive attention mechanism to weightedly fuse the convolutional outputs of different orders of graphs to generate a node embedding representation for identifying the fault area, represented as follows:

[0078]

[0079] Among them, F (i) It is the graph convolution output of the i-th layer, α i These are the importance weights of the output of each layer learned by the attention mechanism, satisfying ∑ i α i =1.

[0080] In this embodiment of the application, step S300 further includes: segmenting the waveform sequence into graphs using a sliding window method, extracting features from several time periods in parallel, and assuming the total number of sampling points is T, the sliding window size is s, and the step size is a, the total number of spatiotemporal graph convolutional module groups k generated is calculated as follows:

[0081]

[0082] K spatiotemporal feature modules are constructed in parallel, and their output features after attention fusion are spliced ​​together to express the spatiotemporal fusion features of the distribution network, enabling spatial perception and localization of fault impacts under sparse graph data.

[0083] It should be noted that this application generates a corresponding spatiotemporal adjacency matrix based on the power grid topology and uses it as input data. Weights are adjusted using a mask matrix to better match the actual distribution network topology. The optimized adjacency matrix is ​​then input into a graph convolution function. A graph convolution module combining skip connections and attention mechanisms extracts node feature vectors that integrate temporal and spatial information. Global max pooling and global average pooling form the final feature vector. Pooling operations eliminate redundant information and accelerate model processing. Using two different pooling methods creates differentiated features, resulting in a final feature vector containing more information and improving model robustness. Finally, the fully connected layer uses a softmax activation function to process the output, converting it into a classification result, i.e., fault location labels.

[0084] Example 2: The above example is an illustrative scheme of a fault location method based on synchronous waveform data processing. It should be noted that the technical solution of this fault location system based on synchronous waveform data processing belongs to the same concept as the above-described fault location method based on synchronous waveform data processing. Details not described in detail in this example of the fault location system based on synchronous waveform data processing can be found in the description of the above-described fault location method based on synchronous waveform data processing.

[0085] This embodiment of a fault location system based on synchronous waveform data processing includes:

[0086] The spatiotemporal adjacency matrix construction module is used to obtain the synchronous waveform data of each node in the distribution network, and construct the spatiotemporal adjacency matrix of the distribution network based on the synchronous waveform data and the topology of the distribution network.

[0087] The data preprocessing module is used to preset the distribution network fault location model, take the spatiotemporal adjacency matrix as the input data of the distribution network fault location model, and preprocess the input data.

[0088] The fault location module is used to locate faulty lines in the distribution network based on preprocessed data and a distribution network fault location model.

[0089] This embodiment also provides an electronic device applicable to fault location methods based on synchronous waveform data processing, including:

[0090] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the fault location method based on synchronous waveform data processing as proposed in the above embodiments.

[0091] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the fault location method based on synchronous waveform data processing as proposed in the above embodiments.

[0092] The storage medium proposed in this embodiment and the fault location method based on synchronous waveform data processing proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0093] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0094] Example 3, referring to Figure 4 This is one embodiment of the present invention. In this embodiment, the IEEE 33-node standard system is used to conduct a distribution network fault location experiment to verify the effectiveness of the present invention.

[0095] This embodiment addresses the issue of insufficient fault location accuracy in traditional methods due to the complexity of the distribution network structure in the context of high-proportion distributed power generation. In the simulation model, photovoltaic power is connected to specific nodes, as detailed below. Figure 4 As shown.

[0096] In this embodiment, the system voltage level is set to 10kV, the frequency to 50Hz, and the total active power of the load is 3.715MW. The distributed power source adopts an inverter-type photovoltaic power source model, and the total active power injected into the photovoltaic power source is 2.64MW. Under normal operating conditions, the photovoltaic penetration rate is approximately 71%. The specific grid connection capacity of the photovoltaic power source is shown in Table 1.

[0097] Table 1. Location and Capacity of Photovoltaic Power Supply

[0098]

[0099] The simulation in this embodiment is implemented in PSCAD. The simulation case takes into account most of the fault conditions, including different fault types, different transition resistances, different fault lines, and the output mode of photovoltaic power sources.

[0100] To verify the fault location capability of this invention in scenarios with a high proportion of distributed power sources, this embodiment conducted a detailed simulation model based on the distribution network parameters in Table 2:

[0101] Table 2 Sample Parameters

[0102]

[0103] In the simulation, the distribution network was set to operate at 100% standard load. Systematic fault injection was performed on each line in the network (11 fault types, 4 transition resistance magnitudes, and 4 photovoltaic power output conditions), generating a total of 226 fault samples. Each sample was sampled within 0.02 seconds of the fault occurrence, with a sampling frequency set to 2kHz to capture the high-frequency dynamic characteristics of the initial fault stage. Since each sample covers 50 time steps, 33 nodes, and 6-dimensional measurements per node (such as three-phase voltage and current), each sample constitutes a three-dimensional spatiotemporal feature tensor of size 50×33×6. Simultaneously, 226 fault samples appeared on each of the 32 lines, for a total of 7232 fault samples.

[0104] The model accuracy comparison is shown in Table 3:

[0105] Table 3 Comparison of accuracy rates for each model

[0106]

[0107]

[0108] As shown in Table 3, the improved spatiotemporal graph fusion graph convolution method of this invention, in the context of a distribution network with a high proportion of distributed power sources, exhibits significantly higher fault location accuracy than the simple data processing model CNN, the GCN model which only outputs the spatial structure between system nodes, and the STGCN model which considers temporal and spatial features separately. Simulation results show that although the waveform data contains characteristic signals that can characterize the fault after a distribution network fault, with the large-scale integration of distributed power sources, the distribution network system structure becomes increasingly complex, and signal interference is severe. The waveform signal data extracted in time and space cannot be accurately used as characteristic signals for fault location, further demonstrating the effectiveness of this invention.

[0109] To verify that the skip connections and attention mechanisms in this method also expand the receptive field of the model during signal extraction, resulting in smoother extraction and adaptability to fewer measurement devices, the skip connections and attention mechanisms are removed from the model of this invention, and the traditional ReLU activation function is directly used. The model of this invention is compared with the ablation model. In the simulation, the measurement signals of some nodes (2, 25, 33) are set to 0 to simulate the situation with fewer measurement devices. The experimental results are shown in Table 4:

[0110] Table 4. Comparison of accuracy between models with and without skip connections and attention mechanisms.

[0111]

[0112] The results show that, in practice, even with fewer measurement devices to ensure economy and address the challenges of deployment, the model of this invention still achieves high accuracy. However, once the skip-layer connection and attention mechanism are removed, the number of measurement devices has a more significant impact on the model's positioning accuracy, further demonstrating the effectiveness of the skip-layer connection and attention mechanism.

[0113] This invention addresses the issue of decreased fault location accuracy due to high-proportion distributed power supply access. It maps collected synchronous measurement traveling wave data to a spatiotemporal graph and combines this with distribution network topology information, significantly improving the robustness of signal extraction. Through a spatiotemporal graph fusion graph convolution method, it utilizes traveling wave signal data that integrates network physical structure and temporal information to achieve precise fault location. To ensure the method's location capability under sparse measurement conditions, it employs layer-skipping connections combined with an attention mechanism to increase the model's capture range for fault characteristic signals, greatly enhancing the sensitivity of the method.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A fault location method based on synchronous waveform data processing, characterized in that, include: Acquire the synchronization waveform data of each node in the distribution network, and construct the spatiotemporal adjacency matrix of the distribution network based on the synchronization waveform data and the topology of the distribution network. A pre-defined distribution network fault location model is established, and the spatiotemporal adjacency matrix is ​​used as the input data for the distribution network fault location model. The input data is then pre-processed. Based on the preprocessed data, the faulty lines in the distribution network are located using the distribution network fault location model.

2. The fault location method based on synchronous waveform data processing as described in claim 1, characterized in that, The construction of the spatiotemporal adjacency matrix of the distribution network includes: The distribution network is modeled as an undirected graph structure, where nodes represent substations, load nodes, and distributed power source access points; edges represent electrical connections. Synchronous waveform measurement devices are deployed at key nodes to collect three-phase voltage and current waveform data, forming node features. The measurement value of each node at any time is represented as a vector, and the traveling wave data collected by the node are stacked. All node features are combined into a three-dimensional matrix as the data input for the graph structure. Based on the connection relationship of distribution network nodes and the continuity of the same node in the time dimension, a spatial adjacency matrix and a temporal adjacency matrix are constructed. The spatial adjacency matrix and the temporal adjacency matrix are then fused to generate a spatiotemporal adjacency matrix, thereby synchronously modeling the spatiotemporal information of the distribution network.

3. The fault location method based on synchronous waveform data processing as described in claim 2, characterized in that, Using the spatiotemporal adjacency matrix as input data for the distribution network fault location model, and preprocessing the input data, includes: inputting the spatiotemporal adjacency matrix as input data into the fully connected layer of the distribution network fault location model for dimensionality upscaling, and then inputting it into the batch normalization layer for batch normalization.

4. The fault location method based on synchronous waveform data processing as described in claim 3, characterized in that, Also includes: Based on the normalized input data, the adjacency strength of the distribution network nodes is adjusted using a learnable mask matrix to obtain the connection weights between different nodes, and the temporal and spatial information of the distribution network is aggregated synchronously.

5. The fault location method based on synchronous waveform data processing as described in claim 4, characterized in that, Also includes: In the extraction of spatiotemporal signals by graph convolution, skip-layer connections and attention mechanisms are introduced, and key fault features are obtained by using global max pooling + average pooling.

6. The fault location method based on synchronous waveform data processing as described in claim 5, characterized in that, Based on the preprocessed data, a distribution network fault location model is used to locate faulty lines in the distribution network. This includes: employing a multi-scale graph convolutional structure to cover the spatial range of the sparse graph, retaining feature representations of different levels through skip connections, and using an additive attention mechanism to weightedly fuse the outputs of graph convolutions of different orders to generate node embedding representations for identifying fault areas, represented as follows: Among them, F (i) It is the graph convolution output of the i-th layer, α i These are the importance weights of the output of each layer learned by the attention mechanism, satisfying ∑ i α i =1.

7. The fault location method based on synchronous waveform data processing as described in claim 6, characterized in that, It also includes: segmenting the waveform sequence into graphs using a sliding window approach, extracting features from several time periods in parallel. Let the total number of sampling points be T, the sliding window size be s, and the stride be a. The total number of spatiotemporal graph convolutional module groups k generated is calculated as follows: K spatiotemporal feature modules are constructed in parallel, and their output features after attention fusion are spliced ​​together to express the spatiotemporal fusion features of the distribution network, enabling spatial perception and localization of fault impacts under sparse graph data.

8. A fault location system based on synchronous waveform data processing, applied to the method described in any one of claims 1-7, characterized in that, include: The spatiotemporal adjacency matrix construction module is used to obtain the synchronous waveform data of each node in the distribution network, and construct the spatiotemporal adjacency matrix of the distribution network based on the synchronous waveform data and the topology of the distribution network. The data preprocessing module is used to preset the distribution network fault location model, use the spatiotemporal adjacency matrix as the input data of the distribution network fault location model, and preprocess the input data. The fault location module is used to locate faulty lines in the distribution network based on the preprocessed data and the distribution network fault location model.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the fault location method based on synchronous waveform data processing as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the fault location method based on synchronous waveform data processing as described in any one of claims 1 to 7.

Images