Electric power topology rapid identification method based on depth map matching

Abstract

The invention relates to the technical field of electric power topology identification, and discloses an electric power topology rapid identification method based on depth map matching, and the method comprises the following steps: S1, carrying out the multi-source heterogeneous data fusion and adaptive noise reduction, and employing an adaptive threshold noise reduction algorithm; s2, modeling a dynamic space-time diagram; s3, attention-driven graph matching is carried out, and a two-channel graph neural network is used; and S4, updating the incremental topology library. For the condition that the real-time data of the low-voltage power distribution network generates significant noise due to equipment aging and acquisition errors, a self-adaptive threshold noise reduction algorithm is adopted, high-frequency noise components are accurately filtered out according to a threshold value, missing data are intelligently repaired in combination with a space-time Kriging interpolation method, the purity and integrity of the data are ensured, and a data basis is provided for subsequent topology recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of power topology recognition, and in particular to a fast power topology recognition method based on deep graph matching. Background Art

[0002] The fast power topology recognition method based on deep graph matching is mainly used to quickly and accurately identify the topological structure in a power system. When a fault occurs in the power system, quickly and accurately identifying the topological structure helps to quickly locate the fault location. By comparing with the normal topological structure, abnormal points can be found in time, so as to guide maintenance personnel to quickly reach the fault site for repair, reduce the power outage time and scope, and improve the power supply reliability.

[0003] After retrieval, the patent with the Chinese patent number CN120067700A discloses a method for verifying the data distribution of topological similarity in a power system, including the following steps: collecting the operation data of the power system; calculating the similarity between topological states; screening out multiple historical topological states closest to the current topological state; performing a fast approximate similarity query to quickly match similar topological states; verifying the power system operation data distribution under different topological states, analyzing the impact of topological changes on the system operation data, identifying abnormal distributions or inconsistent operation states, and introducing a streaming data processing framework. The present invention uses a combination of different similarity measurement methods.

[0004] However, in the above technical solution, there are significant noises in the real-time data of the low-voltage distribution network, such as equipment aging and acquisition errors, and the multi-source data formats are heterogeneous. The number of network nodes and connection relationships grows exponentially, and the computational complexity of traditional algorithms is high, making it difficult to meet the real-time response requirements in fault scenarios. Although the above patent introduces an approximate similarity query, its parallel processing ability for large-scale topologies is limited, resulting in network deployment delays. In addition, the power network load fluctuates and the topology switches frequently, and the existing models rely on historical data training, with insufficient generalization ability for sudden structural changes and need to be repeatedly trained to adapt to new scenarios. Based on this, the present invention designs a fast power topology recognition method based on deep graph matching to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a fast power topology recognition method based on deep graph matching, which solves the real-time response problem in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: A fast power topology recognition method based on deep graph matching, including the following steps: Step S1, Multi-source Heterogeneous Data Fusion and Adaptive Denoising. Synchronously collect high-precision real-time data streams of voltage amplitude U, current phase angle θ, active power P, and reactive power Q of each node in the power system, with a sampling frequency not lower than 1 kHz; Apply the adaptive threshold denoising algorithm. According to the threshold λ = 0.6745·median(|D high-frequency|), accurately filter out the high-frequency noise components generated by equipment aging to ensure data purity; Step S2, Dynamic Spatiotemporal Graph Modeling. Using the electrical connection relationship between power equipment as edges, initialize and construct a binary adjacency matrix A0. The matrix element A0(i,j) is 1 indicating that node i and node j are electrically connected, and 0 indicating non-connection; Build a 3-layer spatiotemporal graph convolutional network, reasonably configure the number of convolutional kernels K = 3, and the dimension of the hidden layer dh = 128. Use this network to deeply fuse the dynamic data of the power system within the sliding time window Tw = 15 min to achieve real-time dynamic update of the adjacency matrix; Step S3, Attention-driven Graph Matching. Use a two-channel graph neural network to perform deep feature extraction on the currently constructed topological graph Gc and the historical topological graph library {Gh} respectively; Step S4, Incremental Topological Library Update. Inject the newly identified topological structure Gc into the historical topological library indexed by the ball tree structure in an orderly manner to ensure that the complexity of the nearest neighbor search is maintained at O(logM), achieving efficient data organization and retrieval.

[0007] Preferably, in step S1, the spatio-temporal Kriging interpolation method is adopted, the spatial correlation scale L = 1.2 km is set, and the missing data is intelligently repaired by combining spatio-temporal correlation to ensure data integrity and temporal continuity; The preprocessed multi-dimensional data is uniformly encoded into a 128-dimensional feature vector Fv, and a node feature matrix X of size N×128 is constructed, where N represents the total number of nodes in the power system.

[0008] Preferably, in step S2, to enhance the training effect of the spatio-temporal graph convolutional network, a transfer learning strategy is introduced. Use the model parameters pre-trained on a large-scale general power topology dataset to initialize the current network, which is used to reduce the number of samples required for training and accelerate the convergence speed.

[0009] Preferably, in step S3, to improve the calculation efficiency, a block parallel calculation strategy is adopted. The complex topological graph is evenly divided into multiple relatively independent subgraphs. With the zero-copy RDMA communication technology, and the communication delay is strictly controlled within ≤2 ms, asynchronous parallel processing is realized in the GPU cluster environment to accelerate the graph matching process; By scientifically calculating the node similarity matrix S, abnormal power nodes are accurately marked according to the preset threshold max(Si) less than 0.85, and the geographical fault coordinates (xi,yi) of the abnormal points are further output.

[0010] Preferably, in step S4, whenever the number of newly added topologies reaches Nnew = 100, the online knowledge distillation mechanism is automatically triggered to prompt the student model to quickly learn new features while retaining key knowledge and continuously optimize the model parameters.

[0011] Preferably, in the noise reduction process of step S1, the improved sliding window CUSUM algorithm is further introduced to monitor current mutation events in real time, and the event extraction sensitivity coefficient β = 0.35 is accurately set; the power data is detected and filtered for anomalies by combining the isolation forest algorithm, and the anomaly score threshold τ = 0.65 is strictly defined.

[0012] Preferably, in step S3, a comprehensive similarity metric function is also constructed by combining two metrics, cosine similarity and Euclidean distance, to match the current topology graph with the graphs in the historical topology graph library; the dropout technique is used to randomly discard some neurons in the hidden layer of the graph neural network with a certain probability, which can prevent overfitting and enable the model to perform graph matching more stably in the face of different scenarios and data changes.

[0013] Preferably, in step S4, the historical topology library is periodically compressed and streamlined, and the usage frequency of each topology structure in the recent T days is counted. When the usage frequency is lower than ω = 0.01, it is removed from the topology library and stored in the cold backup area to optimize the storage space and maintain the efficiency and timeliness of the topology library.

[0014] Preferably, in step S4, the topology library also needs to be backed up regularly, and the backup period is set to once a week. The backup data is stored in a distributed storage system in a different location.

[0015] Preferably, during the entire power topology recognition process, a real-time monitoring and logging system is established to monitor and record each link such as data acquisition, data processing, model training, graph matching, and topology update in real time; this system can timely detect abnormal situations during the system operation process and record the abnormal information in the log file.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are: 1. For the situation where the real-time data of the low-voltage distribution network is significantly noisy due to equipment aging and acquisition errors, the present invention adopts an adaptive threshold noise reduction algorithm to accurately filter out high-frequency noise components according to the threshold, and combines the spatio-temporal Kriging interpolation method to intelligently repair missing data, ensuring the data purity and integrity, and providing a data basis for subsequent topology recognition.

[0017] 2. The present invention constructs a spatio-temporal graph convolutional network, deeply fuses the dynamic data of the power system within a sliding time window, and updates the adjacency matrix in real time and dynamically, enabling real-time capture of the dynamic changes in the topology of the power system; uses a dual-channel graph neural network to perform deep feature extraction on the current topology graph and the historical topology graph library respectively, and adopts a block parallel computing strategy to accelerate the graph matching process, meeting the real-time response requirements in fault scenarios and effectively avoiding the problem of network deployment delay.

[0018] 3. In the present invention, whenever the number of newly added topologies reaches a certain value, an online knowledge distillation mechanism is automatically triggered, prompting the student model to quickly learn new features while retaining key knowledge, continuously optimizing the model parameters, solving the problems of the existing model relying on historical data training, insufficient generalization ability for sudden structural changes, and the need for repeated training, enabling the historical topology graph library to be dynamically updated and improved, and improving the adaptability and accurate description ability of the model to the topology changes of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the overall working process and monitoring architecture diagram of the system of the present invention; Figure 2 is the technical decomposition diagram of data noise reduction and repair of the present invention; Figure 3 is the working flow diagram of the present invention; Figure 4 is the dual-channel working flow diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1; Please refer to Figures 1-4 , in the embodiment of the present invention, a method for quickly identifying power topology based on deep graph matching includes the following steps: Step S1, multi-source heterogeneous data fusion and adaptive noise reduction, synchronously collect high-precision real-time data streams of the voltage amplitude U, current phase angle θ, active power P, and reactive power Q of each node in the power system, and the sampling frequency is not less than 1 kHz; Use the adaptive threshold noise reduction algorithm, and according to the threshold λ = 0.6745·median(|D high-frequency|), accurately filter out the high-frequency noise components generated due to equipment aging to ensure data purity; Step S2: Dynamic spatiotemporal graph modeling. Using the electrical connection relationships between power equipment as edges, a binary adjacency matrix A0 is initialized and constructed. A matrix element A0(i,j) of 1 indicates that node i is electrically connected to node j, and 0 indicates that it is not connected. A three-layer spatiotemporal graph convolutional network is constructed, with a reasonable configuration of the number of convolution kernels K=3 and the hidden layer dimension dh=128. The network is used to deeply fuse the power system dynamic data within the sliding time window Tw=15 minutes to achieve real-time dynamic update of the adjacency matrix. Step S3: attention-driven graph matching, using a dual-channel graph neural network to perform deep feature extraction on the currently constructed topology graph Gc and the historical topology graph library {Gh}; Step S4, incremental topology library update, injects the newly identified topology structure Gc into the historical topology library based on the ball tree structure index in an orderly manner, ensuring that the neighbor search complexity is maintained at O(logM) to achieve efficient data organization and retrieval.

[0022] In step S1, the spatiotemporal kriging interpolation method is used, the spatial correlation scale L is set to 1.2 km, and the missing data is intelligently repaired in combination with the spatiotemporal correlation to ensure data integrity and time series continuity; the preprocessed multi-dimensional data is uniformly encoded into a 128-dimensional feature vector Fv, and a node feature matrix X of a scale of N×128 is constructed, where N represents the total number of nodes in the power system.

[0023] In step S2, in order to enhance the training effect of the spatiotemporal graph convolutional network, a transfer learning strategy is introduced to initialize the current network using the model parameters pre-trained on a large-scale general power topology dataset, in order to reduce the number of samples required for training and accelerate the convergence speed.

[0024] In step S3, to improve computing efficiency, a block parallel computing strategy is adopted to evenly divide the complex topology graph into multiple relatively independent subgraphs. With the help of zero-copy RDMA communication technology, and the communication delay is strictly controlled to ≤2ms, asynchronous parallel processing is realized in a GPU cluster environment to accelerate the graph matching process; through scientific calculation of the node similarity matrix S, the abnormal power nodes are accurately marked according to the preset threshold max(Si) less than 0.85, and the geographic fault coordinates (xi,yi) of the abnormal points are further output.

[0025] In step S4, whenever the number of new topologies reaches Nnew=100, the online knowledge distillation mechanism is automatically triggered to encourage the student model to quickly learn new features while retaining key knowledge and continuously optimize model parameters.

[0026] The working principle of the embodiments of the present invention is as follows: In step S1, focusing on the working principle of multi-source heterogeneous data fusion and adaptive noise reduction, when the power system operates, four types of key data, namely voltage amplitude U, current phase angle θ, active power P, and reactive power Q, will be generated at each node and synchronously collected in the form of a high-precision real-time data stream. The sampling frequency is strictly set to be not less than 1 kHz, aiming to finely capture the rapidly changing dynamic characteristics of the power system. Considering that factors such as equipment aging are likely to introduce high-frequency noise, the present invention uses an adaptive threshold noise reduction algorithm. Its core principle is based on the threshold λ = 0.6745·median(|Dhigh-frequency|), which can accurately identify and filter out high-frequency noise components. Here, Dhigh-frequency represents the signal after being filtered by a specific high-frequency band, and median represents the median operation, realizing the accurate quantification and effective filtering of noise, thereby ensuring the data purity; at the same time, aiming at the possible missing situations in the data collection process, the spatio-temporal Kriging interpolation method is introduced, and the spatial correlation scale is set to L = 1.2 km. Combining spatio-temporal correlation, the missing data is intelligently repaired to ensure data integrity and temporal continuity, and the true operating state of the power system is restored to the greatest extent.

[0027] Entering step S2, the core lies in the working principle of dynamic spatio-temporal graph modeling. First, a binary adjacency matrix A0 is constructed based on the electrical connection relationship between power equipment. The matrix element A0(i,j) is 1 or 0, intuitively indicating whether there is an electrical connection between node i and node j. This matrix initially constructs the topological framework of the power system. On this basis, a 3-layer spatio-temporal graph convolutional network is built, where the number of convolutional kernels K is determined to be 3, and the hidden layer dimension is set to dh = 128. Such a configuration aims to achieve in-depth fusion and efficient processing of the dynamic data of the power system. By setting the sliding time window Tw = 15 min, the dynamic data within a period of time is included in the analysis scope. Using the powerful feature extraction ability of the spatio-temporal graph convolutional network, the dynamic changes of the topological structure of the power system are captured in real time, and the adjacency matrix is dynamically updated accordingly, so that the topological model always fits the actual operating conditions.

[0028] In step S3, the key principle lies in attention-driven graph matching. A dual-channel graph neural network architecture is employed, where each channel separately conducts deep feature extraction for the currently constructed topological graph Gc and the historical topological graph library {Gh}. The dual-channel design enables the network to parallelly and focus on mining topological graph features from two different sources, fully extracting deep feature information of nodes, edges, and the overall topological structure, providing accurate feature representations for subsequent matching operations. Meanwhile, to improve computational efficiency, a block parallel computing strategy is adopted, evenly dividing the complex power topological graph into multiple relatively independent subgraphs according to specific rules. In a GPU cluster environment, with the help of zero-copy RDMA (Remote Direct Memory Access) communication technology, efficient data transmission and exchange between subgraphs are achieved, and the communication delay is strictly controlled within ≤2ms to ensure the efficiency of asynchronous parallel processing and significantly accelerate the graph matching process. Additionally, by scientifically calculating the node similarity matrix S, abnormal power nodes are accurately marked according to the judgment rule that the preset threshold max(Si) is less than 0.85. This can not only quickly locate abnormal parts in the power system but also further output the geographical fault coordinates (xi, yi) of the abnormal points, providing key location information for fault repair and system maintenance.

[0029] Moving on to step S4, its working principle revolves around incremental topological library update. Whenever the newly identified topological structure Gc reaches a certain quantity, that is, when the number of newly added topologies reaches Nnew = 100, an online knowledge distillation mechanism is automatically triggered. The core principle of this mechanism is to enable the student model to quickly learn new features while retaining the key knowledge of the existing model. By constructing an effective inheritance and fusion path between old and new knowledge and continuously optimizing model parameters, the historical topological library can be dynamically updated and continuously improved, always maintaining high adaptability and accurate description ability for topological changes in the power system. Meanwhile, based on the data organization method of the ball tree structure index, the complexity of nearest neighbor search is ensured to be maintained at O(logM), achieving efficient data organization and retrieval.

[0030] Embodiment 2; Please refer to Figures 1-4 In the noise reduction processing flow of step S1 in the embodiments of the present invention, an improved sliding window CUSUM algorithm is further introduced to monitor current mutation events in real time, and the event extraction sensitivity coefficient β = 0.35 is accurately set; the isolation forest algorithm is combined to perform anomaly detection and filtering on power data, and the anomaly score threshold τ = 0.65 is strictly defined.

[0031] In step S3, a comprehensive similarity metric function is also constructed by combining two metric methods, cosine similarity and Euclidean distance, to match the current topology graph with the graphs in the historical topology graph library; the dropout technique is used to randomly discard some neurons in the hidden layer of the graph neural network with a certain probability, which can not only prevent overfitting but also enable the model to perform graph matching more stably in the face of different scenarios and data changes.

[0032] In step S4, the historical topology library is periodically compressed and streamlined. The usage frequency of each topology structure in the past T days is counted. When the usage frequency is lower than ω = 0.01, it is removed from the topology library and stored in the cold backup area to optimize the storage space and maintain the efficiency and timeliness of the topology library.

[0033] In step S4, the topology library also needs to be backed up regularly, and the backup period is set to once a week. The backup data is stored in a distributed storage system in a different location.

[0034] During the entire power topology recognition process, a real-time monitoring and logging system is established to monitor and record each link such as data acquisition, data processing, model training, graph matching, and topology update in real time; this system can promptly detect abnormal situations during the system operation process and record the abnormal information in the log file.

[0035] The working principle of the embodiment of the present invention is as follows: in the noise reduction processing flow of step S1, an improved sliding window CUSUM algorithm is introduced. By setting the event extraction sensitivity coefficient β = 0.35, this algorithm can monitor the current mutation events in the power system in real time. Its core principle is based on the statistical process control theory to perform cumulative sum test within the sliding window for the current data. When the current data mutates, the algorithm can quickly detect that the cumulative sum exceeds the set threshold, thereby accurately identifying the current mutation events.

[0036] The Isolation Forest algorithm is used to detect and filter abnormal power data. By combining the Isolation Forest algorithm to detect and filter abnormal power data, the abnormal score threshold τ = 0.65 is strictly defined. The Isolation Forest algorithm is based on the principle of random forest and measures the abnormality degree of data points by constructing multiple isolation trees. For each data point in the power data, the algorithm calculates its abnormal score. If the score exceeds the threshold τ = 0.65, it is determined as abnormal data and filtered.

[0037] Construct a comprehensive similarity metric function by combining two metrics, cosine similarity and Euclidean distance, to match the current topology graph with the graphs in the historical topology graph library. Specifically, cosine similarity is used to measure the directional similarity of two topology graphs in the feature vector space. By calculating the ratio of their dot product to the norm length, it reflects the angular relationship between the topology graphs. Euclidean distance, on the other hand, measures the straight-line distance between two topology graphs in the feature vector space, reflecting their numerical differences.

[0038] Compression and refinement of the historical topology library. Regularly compress and refine the historical topology library, and count the usage frequency of each topology structure in the past T days. When the usage frequency is lower than ω = 0.01, remove it from the topology library and store it in the cold backup area. This process is based on the analysis of the usage frequency of topology structures. Topology structures with low usage frequency have less reference value in the current operation of the power system. Removing them can optimize the storage space, and at the same time store these topology structures in the cold backup area for future needs.

[0039] During the entire power topology recognition process, establish a real-time monitoring and logging system. This system monitors and records each link of data acquisition, data processing, model training, graph matching, topology update, etc. in real time. By setting monitoring points at each key link, collect data such as system operation status information, data flow information, model training progress, and performance indicators, and record them in the log file. When an abnormal situation occurs during the system operation, the real-time monitoring and logging system can detect it in time and issue an alarm, and at the same time record the abnormal information in detail in the log file.

[0040] Embodiment 3; Please refer to Figures 1-4 , this embodiment provides a fast power topology recognition method based on deep graph matching. The specific steps are as follows: Apply the adaptive threshold noise reduction algorithm. According to the threshold λ = 0.6745×median(|D high-frequency|), where D high-frequency represents the signal after filtering in a specific high-frequency band, and median represents the median operation, accurately filter out the high-frequency noise components generated by equipment aging to ensure data purity. Introduce the spatio-temporal Kriging interpolation method, set the spatial correlation scale L = 1.2 km, and intelligently repair the missing data in combination with spatio-temporal correlation to ensure data integrity and temporal continuity. Uniformly encode the preprocessed multi-dimensional data into 128-dimensional feature vectors Fv, and construct a node feature matrix X with a size of N×128, where N represents the total number of nodes in the power system.

[0041] Meanwhile, in the noise reduction process, an improved sliding window CUSUM algorithm is further introduced to monitor current mutation events in real time, and the sensitivity coefficient β = 0.35 for event extraction is accurately set; the Isolation Forest algorithm is combined to perform anomaly detection and filtering on power data, and the anomaly score threshold τ = 0.65 is strictly defined.

[0042] Taking the electrical connection relationship between power equipment as edges, a binary adjacency matrix A0 is initialized to construct the topological framework of the power system. The matrix element A0(i,j) is 1 indicating that node i and node j are electrically connected, and 0 indicating non-connection.

[0043] Adopting a block parallel computing strategy, the complex power topology map is evenly divided into 8 relatively independent subgraphs. With the zero-copy RDMA communication technology, and the communication delay is strictly controlled within ≤2ms, asynchronous parallel processing is realized in the GPU cluster environment to accelerate the graph matching process. By scientifically calculating the node similarity matrix S, abnormal power nodes are accurately marked according to the preset threshold max(Si) less than 0.85, and the geographical fault coordinates (xi,yi) of the abnormal points are further output.

[0044] Whenever the newly recognized topological structure Gc is orderly injected into the historical topology library indexed by the ball tree structure, when the number of newly added topologies Nnew = 100, the online knowledge distillation mechanism is automatically triggered to enable the student model to quickly learn new features while retaining key knowledge and continuously optimize the model parameters.

[0045] The historical topology library is regularly compressed and streamlined. The usage frequency of each topological structure in the past T = 30 days is counted. When the usage frequency is lower than ω = 0.01, it is removed from the topology library and stored in the cold backup area to optimize the storage space and maintain the efficiency and timeliness of the topology library; at the same time, the topology library is regularly backed up, and the backup period is set to once a week, and the backup data is stored in a distributed storage system in a different location.

[0046] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fast identification method for power topology based on depth map matching, characterized in that It includes the following steps: Step S1, multi-source heterogeneous data fusion and adaptive noise reduction. Synchronously collect high-precision real-time data streams of voltage amplitude U, current phase angle θ, active power P, and reactive power Q of each node in the power system, with a sampling frequency not lower than 1 kHz; Apply the adaptive threshold noise reduction algorithm. According to the threshold λ = 0.6745·median(|D high-frequency|), accurately filter out the high-frequency noise components generated by equipment aging to ensure data purity; Step S2, dynamic spatio-temporal graph modeling. Using the electrical connection relationship between power equipment as edges, initialize and construct a binary adjacency matrix A0. The matrix element A0(i,j) is 1 indicating that node i and node j are electrically connected, and 0 indicating non-connection; Build a 3-layer spatio-temporal graph convolutional network, reasonably configure the number of convolutional kernels K = 3, and the hidden layer dimension dh = 128. Use this network to deeply fuse the dynamic data of the power system within the sliding time window Tw = 15 min to achieve real-time dynamic update of the adjacency matrix; Step S3, attention-driven graph matching. Use a two-channel graph neural network to perform deep feature extraction on the currently constructed topology graph Gc and the historical topology graph library {Gh} respectively; Step S4, incremental topology library update. Inject the newly identified topology structure Gc into the historical topology library indexed by the ball tree structure in an orderly manner to ensure that the complexity of nearest neighbor search is maintained at O(logM), and achieve efficient data organization and retrieval.

2. The fast identification method of power topology based on depth map matching according to claim 1, wherein In step S1, use the spatio-temporal Kriging interpolation method, set the spatial correlation scale L = 1.2 km, and intelligently repair missing data in combination with spatio-temporal correlation to ensure data integrity and temporal continuity; Uniformly encode the preprocessed multi-dimensional data into a 128-dimensional feature vector Fv, and construct a node feature matrix X of size N×128, where N represents the total number of nodes in the power system.

3. A fast recognition method for power topology based on depth map matching according to claim 1, characterized in that In step S2, to enhance the training effect of the spatio-temporal graph convolutional network, introduce a transfer learning strategy, and use the model parameters pre-trained on a large-scale general power topology dataset to initialize the current network, which is used to reduce the number of samples required for training and accelerate the convergence speed.

4. A fast identification method for power topology based on depth map matching according to claim 1, characterized in that: In step S3, to improve the computing efficiency, adopt a block parallel computing strategy, evenly divide the complex topology graph into multiple relatively independent subgraphs, and use the zero-copy RDMA communication technology, and strictly control the communication delay to ≤2 ms. Achieve asynchronous parallel processing in the GPU cluster environment to accelerate the graph matching process; Scientifically calculate the node similarity matrix S, and accurately mark abnormal power nodes according to the preset threshold max(Si) < 0.85, and further output the geographical fault coordinates (xi,yi) of the abnormal points.

5. A fast identification method for power topology based on depth map matching according to claim 1, characterized in that: In step S4, whenever the number of newly added topologies reaches Nnew = 100, automatically trigger the online knowledge distillation mechanism, which is used to prompt the student model to quickly learn new features while retaining key knowledge, and continuously optimize the model parameters.

6. The fast identification method of power topology based on depth map matching according to claim 1, characterized in that: In the noise reduction processing flow of step S1, an improved sliding window CUSUM algorithm is further introduced to monitor current mutation events in real time, and the event extraction sensitivity coefficient β = 0.35 is accurately set; the Isolation Forest algorithm is combined to perform anomaly detection and filtering on power data, and the anomaly score threshold τ = 0.65 is strictly defined.

7. A fast recognition method for power topology based on depth map matching according to claim 1, characterized in that: In step S3, a comprehensive similarity metric function is also constructed by combining two metric methods, cosine similarity and Euclidean distance, to match the current topology graph with the graphs in the historical topology graph library; through the dropout technique, some neurons are randomly discarded with a certain probability in the hidden layer of the graph neural network, which can prevent overfitting and enable the model to perform graph matching more stably in the face of different scenarios and data changes.

8. A fast recognition method for power topology based on depth map matching according to claim 1, characterized in that: In step S4, the historical topology library is regularly compressed and streamlined. The usage frequency of each topology structure in the recent T days is counted. When the usage frequency is lower than ω = 0.01, it is removed from the topology library and stored in the cold backup area to optimize the storage space and maintain the efficiency and timeliness of the topology library.

9. A fast recognition method for power topology based on depth map matching according to claim 1, characterized in that: In step S4, the topology library also needs to be backed up regularly. The backup period is set to once a week, and the backup data is stored in a distributed storage system in a remote location.

10. A method for rapid identification of power topology based on depth map matching according to claim 1, characterized in that: During the entire power topology recognition process, a real-time monitoring and logging system is established to monitor and record each link, including data acquisition, data processing, model training, graph matching, topology update, etc.; this system can timely detect abnormal situations during the system operation process and record the abnormal information in the log file.

Images