Power distribution network fault identification method and system based on adaptive visual view and multi-source fusion
By using an adaptive visual diagram and multi-source fusion method, and employing a correlation variance contribution algorithm and graph neural network, the problem of accurately identifying high-impedance faults in distribution networks was solved, and high-precision fault diagnosis was achieved in complex noise environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID HUNAN ELECTRIC POWER CO LTD ELECTRIC POWER SCI RES INST
- Filing Date
- 2026-04-07
- Publication Date
- 2026-07-10
Smart Images
Figure CN122365129A_ABST
Abstract
Description
Technical Field
[0001] This invention mainly relates to the field of power distribution network technology, specifically to a power distribution network fault identification method and system based on adaptive visibility and multi-source fusion. Background Technology
[0002] As the final link in the power system, the distribution network operates in an extremely complex environment and is prone to frequent faults. High-impedance faults (HIFs), in particular, are typically caused by broken conductors coming into contact with high-impedance media such as trees, gravel surfaces, or asphalt roads. Unlike metallic short circuits, HIFs generate extremely weak fault currents, often smaller than the load current, and their transient characteristics are easily masked by normal load fluctuations or background noise. If such faults are not cleared promptly and accurately, they can lead to continuous arcing to ground, easily igniting surrounding dry branches and leaves, causing forest fires, and may also pose a fatal electric shock threat to pedestrians due to live conductors falling to the ground.
[0003] With the large-scale integration of distributed power sources, such as photovoltaics and wind power, into the distribution network, these sources are gradually exhibiting significant active, nonlinear, and highly time-varying characteristics. The widespread application of power electronic inverters has injected a large amount of nonlinear harmonics and random high-frequency noise into the power grid, completely altering the background noise spectrum characteristics of the system. Traditional protection methods based on a single electrical quantity (such as using only the zero-sequence current amplitude) or simple fixed thresholds often fail to extract effective fault fingerprints when faced with such strong noise and non-stationary interference, resulting in frequent failures to operate or false trips.
[0004] Existing power distribution network fault diagnosis technologies mainly face the following three insurmountable shortcomings: 1. Limitations of diagnostic methods based on a single signal source: Existing technologies mostly rely on single-ended electrical quantities for analysis. However, when high-impedance faults occur, the fault characteristics are extremely weak and easily affected by environmental noise and load switching interference. A single signal often contains limited information and cannot fully depict the complete picture of the fault, resulting in a significant decrease in identification accuracy in low signal-to-noise ratio environments.
[0005] 2. Shortcomings of simple multi-source information splicing and fusion methods: Although some technologies attempt to combine multi-source data such as voltage and current, they usually only perform simple splicing at the data layer or feature layer. This method ignores the correlation, complementarity, and redundancy between different sensor signals, which may lead to the accumulation of invalid information (noise) or even obscure key fault features, failing to achieve true "complementary advantages".
[0006] 3. Limitations of traditional temporal deep learning methods: Methods such as CNN and LSTM typically treat time series data as Euclidean space data, focusing on extracting local or temporal dependent features, but neglecting the structural dependencies and geometric features between data samples. For transient fault signals in distribution networks with non-stationary and nonlinear characteristics, it is difficult to capture the implicit topological change patterns in the signal waveforms, limiting the accuracy and generalization ability of fault diagnosis. Summary of the Invention
[0007] To address the technical problems existing in the prior art, this invention provides a distribution network fault identification method and system with high identification accuracy based on adaptive visual diagrams and multi-source fusion.
[0008] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows: A method for distribution network fault identification based on adaptive visibility and multi-source fusion includes the following steps: S1: Acquire multi-source electrical analog data of the power distribution network and preprocess it to obtain preprocessed multi-source signals; S2: Dynamically weighted and fused the preprocessed multi-source signals using the correlation variance contribution algorithm to obtain the fused signal; S3: The fused signal is mapped into a graph structure using an adaptive convergent visual graph algorithm; S4: Input the graph structure into a pre-built graph neural network model to obtain the global graph feature vector; S5: Input the global graph feature vector into the fully connected layer, and output the fault probability distribution through the Softmax classifier.
[0009] Preferably, the specific process of step S2 is as follows: First, calculate the cross-correlation function: assuming the collected data... The signals from the same type of sensor are ; Calculate any two signals and Cross-correlation function between :
[0010] in, This represents the expectation operation. For time delay; Secondly, calculate the total correlation energy: calculate the first The total correlation energy of the sensor signal with all other signals :
[0011] Finally, the fused signal is generated: the dynamic fusion coefficient is calculated based on the total correlation energy of each signal. The original signals are then weighted and fused to obtain the final fused signal. for: .
[0012] Preferably, the dynamic fusion coefficient The calculation formula is:
[0013] in This represents the total correlation energy of the i-th signal at time t. Let represent the sum of the energies of all signals at time t.
[0014] Preferably, the specific process of step S3 is as follows: First, subsampling aggregation: The fused signal is divided into sliding windows, and a maximum value aggregation operation is performed within each window to extract local peak feature sequences. ;
[0015] in Step size; For fused signals; n For window indexing; Secondly, adaptive convolution feature extraction: introducing a one-dimensional convolution operator to the feature sequence Perform convolution processing; Next, weighted adjacency matrix generation: The extracted feature sequences are arranged in parallel diagonal directions, and a weighted adjacency matrix is constructed based on the visibility principle or the strength of feature association, thereby generating graph structure data containing nodes and edges.
[0016] Preferably, a ReLU activation function is introduced after convolution to enhance nonlinear expressive power and filter noise.
[0017] Preferably, in step S4, the pre-built graph neural network model includes a differentiable pooling module for aggregating and updating node features through a graph convolutional network, and also learns an assignment matrix to cluster nodes into coarser-grained nodes, thereby achieving hierarchical downsampling of the graph.
[0018] Preferably, in step S5, the current power grid operating status is determined based on the maximum probability. Typical operating conditions that can be identified include: normal operation, single-phase high-resistance grounding fault, metallic short-circuit fault, two-phase short-circuit fault, and capacitor switching disturbance.
[0019] Preferably, in step S1, the electrical analog data includes the zero-sequence current and three-phase voltage of multiple feeders.
[0020] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, the computer program performing the steps of the method described above when run by a processor.
[0021] The present invention further discloses a power distribution network fault identification system based on adaptive visibility and multi-source fusion, including a memory and a processor connected to each other. The memory stores a computer program, which executes the steps of the method described above when run by the processor.
[0022] Compared with the prior art, the advantages of the present invention are as follows: This invention innovatively transforms fault identification from the traditional "single time-series analysis" paradigm to a deep learning paradigm of "multi-source fusion + graph structure mapping". Specifically, firstly, the Correlation Variance Contribution (CVC) algorithm dynamically allocates weights by utilizing the cross-correlation and variance contribution rate among multi-source signals, effectively leveraging the complementarity of multi-source electrical quantities and significantly enhancing the signal-to-noise ratio of fault features. Secondly, the Adaptive Converging Visible Graph (AcvGraph) technique is introduced to transform one-dimensional power waveform data into graph structure data, using convolution and maximum value aggregation to capture local transient changes and global topological relationships in the signal. Finally, the DiffPool graph neural network is combined to achieve end-to-end feature extraction and classification.
[0023] This invention constructs a feature space that objectively characterizes the essential differences in physical correlation and topological structure between normal operating conditions and fault states, enabling accurate identification of known faults and effective screening and elimination of unknown benign disturbances. Experiments demonstrate that this method exhibits higher accuracy and robustness compared to traditional CNN or LSTM methods when dealing with high-resistance faults and complex noise environments, providing a more reliable and advanced technical solution for the safe and stable operation of power distribution networks. Attached Figure Description
[0024] Figure 1 This is a flowchart of a power distribution network fault identification method based on adaptive visual diagrams and multi-source fusion, according to an embodiment of the present invention. Detailed Implementation
[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0026] like Figure 1 As shown in the figure, the power distribution network fault identification method based on adaptive visibility and multi-source fusion provided by this embodiment of the invention includes the following steps: S1: Multi-source electrical signal acquisition and preprocessing Using synchronous waveform recording devices installed at key nodes of the distribution network, multi-channel electrical analog data (including zero-sequence current and three-phase voltage of multiple feeders) are collected. To eliminate the influence of differences in the dimensions of different sensors and DC bias on the model, the acquired raw signal sequence was Z-Score normalized. To address the high-frequency sampling characteristics of fault signals in distribution networks, a sliding time window technique is employed to extract continuous time-series signals, generating a subsample set containing transient fault information.
[0027] This step ensures that the model input is a valid fragment containing the complete transient process.
[0028] S2: Data-level fusion based on CVC To address the issues of poor noise immunity and incomplete information from single sensors, a Correlation Variance Contribution (CVC) algorithm is introduced to dynamically weight and fuse multi-source signals.
[0029] First, calculate the cross-correlation function: assuming the collected data... The signals from the same type of sensor are First, calculate any two signals. and Cross-correlation function between :
[0030] in, This represents the expectation operation. For time delay.
[0031] Secondly, calculate the total correlation energy: calculate the first The total correlation energy of the sensor signal with all other signals :
[0032] This energy value reflects the first The degree of correlation between road signals and other signals; the higher the correlation, the more common fault characteristics it contains.
[0033] Finally, the fused signal is generated: the dynamic fusion coefficient is calculated based on the total correlation energy of each signal. The original signals are then weighted and fused. A signal in Dynamic fusion coefficient at time Defined as the relevant energy percentage, specifically:
[0034] in This represents the total correlation energy of the i-th signal at time t. This represents the sum of the energies of all signals at time t; The final fusion enhancement signal for:
[0035] This step fully utilizes the complementarity of multi-source signals, resulting in a fused signal with a higher signal-to-noise ratio and more significant transient characteristics at the moment of fault occurrence.
[0036] S3: Adaptive Converging Visual Graph (AcvGraph) Construction To extract the topological features of time-series signals, an Adaptive Converging Visible Graph (AcvGraph) algorithm is used to map the fused one-dimensional time-series data into a graph structure to extract its topological features. Specifically, this includes the following steps: First, subsampling and aggregation: The fused signal is divided into sliding windows, and a maximum value aggregation operation is performed within each window to extract local peak feature sequences. This step effectively preserves the pulse impact information (i.e., peak characteristics) at the moment of the fault occurrence, while reducing data dimensionality.
[0037] in Step size, n For window indexing.
[0038] Secondly, adaptive convolution feature extraction is performed: a one-dimensional convolution operator is introduced to the feature sequence. This is processed by adjusting the kernel length. This allows for the flexible capture of local information at different scales. To enhance nonlinear expressiveness and filter noise, a ReLU activation function is introduced after convolution:
[0039] Next, a weighted adjacency matrix is generated: the extracted feature sequences are arranged along a diagonal parallel direction, and a weighted adjacency matrix is constructed based on the principle of visibility or feature association strength, thereby generating graph structure data containing nodes and edges. The edge weights in the graph reflect the association strength of signal features at different time points.
[0040] S4: Graph Feature Extraction Based on DiffPool Construct a graph neural network (GNN) model (CVC-Net) that includes a differentiable pooling (DiffPool) module.
[0041] The graph data generated in step S3 is input into the model. The DiffPool module not only aggregates and updates node features through a graph convolutional network (GCN), but also learns an assignment matrix to cluster nodes into coarser-grained nodes, achieving hierarchical downsampling of the graph. This process reduces dimensionality while preserving key topological information, enabling the model to extract high-order graph features that are beneficial for fault classification.
[0042] S5: Fault Type Identification The global graph feature vector, obtained after multi-layer DiffPool processing, is input into the fully connected layer, and a Softmax classifier outputs the fault probability distribution. The system determines the current power grid operating state based on the maximum probability value. Typical operating conditions that can be identified include: normal operation, single-phase high-resistance grounding fault, metallic short-circuit fault, two-phase short-circuit fault, and capacitor switching disturbance. This classifier combines the fusion features of multi-source information with the topological features of time-series signals, thereby achieving high-precision identification of various faults, especially the highly concealed high-resistance faults.
[0043] This invention innovatively transforms fault identification from the traditional "single time-series analysis" paradigm to a deep learning paradigm of "multi-source fusion + graph structure mapping". Specifically, firstly, the Correlation Variance Contribution (CVC) algorithm dynamically allocates weights by utilizing the cross-correlation and variance contribution rate among multi-source signals, effectively leveraging the complementarity of multi-source electrical quantities and significantly enhancing the signal-to-noise ratio of fault features. Secondly, the Adaptive Converging Visible Graph (AcvGraph) technique is introduced to transform one-dimensional power waveform data into graph structure data, using convolution and maximum value aggregation to capture local transient changes and global topological relationships in the signal. Finally, the DiffPool graph neural network is combined to achieve end-to-end feature extraction and classification.
[0044] This invention constructs a feature space that objectively characterizes the essential differences in physical correlation and topological structure between normal operating conditions and fault states, enabling accurate identification of known faults and effective screening and elimination of unknown benign disturbances. Experiments demonstrate that this method exhibits higher accuracy and robustness compared to traditional CNN or LSTM methods when dealing with high-resistance faults and complex noise environments, providing a more reliable and advanced technical solution for the safe and stable operation of power distribution networks.
[0045] This invention also discloses a computer-readable storage medium storing a computer program thereon, which, when run by a processor, executes the steps of the method described above. This invention further discloses a distribution network fault identification system based on adaptive visibility and multi-source fusion, comprising an interconnected memory and a processor, wherein the memory stores a computer program, which, when run by a processor, executes the steps of the method described above. The medium and system of this invention, corresponding to the methods described above, also possess the advantages described above.
[0046] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be implemented by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium includes: any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0047] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.
Claims
1. A method for distribution network fault identification based on adaptive visibility and multi-source fusion, characterized in that, Includes the following steps: S1: Acquire multi-source electrical analog data of the power distribution network and preprocess it to obtain preprocessed multi-source signals; S2: Dynamically weighted and fused the preprocessed multi-source signals using the correlation variance contribution algorithm to obtain the fused signal; S3: The fused signal is mapped into a graph structure using an adaptive convergent visual graph algorithm; S4: Input the graph structure into a pre-built graph neural network model to obtain the global graph feature vector; S5: Input the global graph feature vector into the fully connected layer, and output the fault probability distribution through the Softmax classifier.
2. The distribution network fault identification method based on adaptive visible view and multi-source fusion according to claim 1, characterized in that, The specific process of step S2 is as follows: First, calculate the cross-correlation function: assuming the collected data... The signals from the same type of sensor are ; Calculate any two signals and Cross-correlation function between : in, This represents the expectation operation. For time delay; Secondly, calculate the total correlation energy: calculate the first The total correlation energy of the sensor signal with all other signals : Finally, the fused signal is generated: its dynamic fusion coefficient is calculated based on the total correlation energy of each signal. The original signals are then weighted and fused to obtain the final fused signal. for: 。 3. The distribution network fault identification method based on adaptive visual image and multi-source fusion according to claim 2, characterized in that, Dynamic fusion coefficient The calculation formula is: in This represents the total correlation energy of the i-th signal at time t. Let represent the sum of the energies of all signals at time t.
4. The distribution network fault identification method based on adaptive visibility and multi-source fusion according to claim 1, 2, or 3, characterized in that, The specific process of step S3 is as follows: First, subsampling aggregation: The fused signal is divided into sliding windows, and a maximum value aggregation operation is performed within each window to extract local peak feature sequences. ; in Step size; For fused signals; n For window indexing; Secondly, adaptive convolution feature extraction: introducing a one-dimensional convolution operator to the feature sequence Perform convolution processing; Next, weighted adjacency matrix generation: The extracted feature sequences are arranged in parallel diagonal directions, and a weighted adjacency matrix is constructed based on the visibility principle or the strength of feature association, thereby generating graph structure data containing nodes and edges.
5. The distribution network fault identification method based on adaptive visual image and multi-source fusion according to claim 4, characterized in that, A ReLU activation function is introduced after convolution to enhance nonlinear expressiveness and filter noise.
6. The distribution network fault identification method based on adaptive visibility and multi-source fusion according to claim 1, 2, or 3, characterized in that, In step S4, the pre-built graph neural network model includes a differentiable pooling module for aggregating and updating node features through a graph convolutional network. It also learns an assignment matrix to cluster nodes into coarser-grained nodes, thereby achieving hierarchical downsampling of the graph.
7. The distribution network fault identification method based on adaptive visibility and multi-source fusion according to claim 1, 2, or 3, characterized in that, In step S5, the current power grid operating status is determined based on the maximum probability. Typical operating conditions that can be identified include: normal operation, single-phase high-resistance grounding fault, metallic short-circuit fault, two-phase short-circuit fault, and capacitor switching disturbance.
8. The distribution network fault identification method based on adaptive visibility and multi-source fusion according to claim 1, 2, or 3, characterized in that, In step S1, the electrical analog data includes the zero-sequence current and three-phase voltage of multiple feeders.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-8.
10. A power distribution network fault identification system based on adaptive visibility and multi-source fusion, comprising an interconnected memory and a processor, wherein the memory stores a computer program, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-8.