Power system wiring detection method based on block parallel processing

Abstract

The application discloses a power system wiring detection method based on block parallel processing and relates to the current detection field, which comprises the following steps: collecting real-time operation data through sensors arranged at key nodes, constructing a basic data set after Kalman filter denoising and normalization processing; establishing a sub-block identification system based on topological structure blocks, and extracting electrical and topological feature vectors; adopting a master-slave parallel computing architecture, assigning tasks by priority by the master node, and performing sub-block level anomaly preliminary screening by using an improved random forest model; realizing cross-sub-block collaborative verification by checking boundary tie switch state, branch current and bus voltage; finally, constructing a global wiring state atlas for difference analysis, accurately positioning abnormal equipment and types, and verifying the results through high-precision re-measurement. The application has the advantages that the efficiency is improved through scientific block and distributed parallel computing, the accuracy is guaranteed by combining multi-dimensional detection, boundary checking and re-measurement mechanism, wiring abnormalities are efficiently located, and reliable support is provided.

Description

Technical Field

[0001] This invention relates to the field of power detection, and in particular to a power system wiring detection method based on block parallel processing. Background Technology

[0002] As power systems continue to expand in scale and become increasingly complex, traditional serial connection detection methods face severe challenges in terms of real-time performance and computational efficiency. On the one hand, modern power grids contain massive numbers of nodes and branches, making manual verification or single-machine serial processing insufficient to meet the timeliness requirements of high-dimensional topology analysis. On the other hand, the high proportion of new energy integration and the dynamic reconfiguration requirements of smart grids have significantly increased the frequency of system topology changes, necessitating more efficient automated detection methods.

[0003] Current power system wiring detection methods on the market suffer from several shortcomings. Firstly, most lack systematic noise reduction and data completion processes and standardized procedures, making them susceptible to random noise and missing data, resulting in insufficient accuracy of basic data and creating potential errors for subsequent detection. Secondly, in terms of computational efficiency, traditional methods often employ centralized processing architectures without scientifically partitioning data based on topological characteristics. When dealing with large-scale power systems, the concentrated data processing pressure easily leads to delays and stuttering, and the absence of priority task allocation mechanisms results in low resource utilization efficiency. Thirdly, in terms of detection accuracy, they are often limited to detecting single electrical features, ignoring topological feature correlations and lacking cross-regional boundary collaborative verification. Relying solely on the output of a single model makes them prone to misjudgments due to signal interference. Finally, in the anomaly localization and verification stages, most methods fail to construct a global state map comparison system, achieving only device-level accuracy rather than specific port-level accuracy. Furthermore, the lack of high-precision retesting and secondary verification mechanisms leads to high rates of missed and false detections, making it difficult to meet the precise detection requirements of complex power grids. Summary of the Invention

[0004] To improve existing methods, a power system wiring detection method based on block parallel processing is provided. This method improves efficiency through precise data processing, scientific block partitioning and distributed parallel computing, and ensures accuracy by combining multi-dimensional detection, boundary verification and retesting mechanisms. It can efficiently locate power system wiring anomalies and provide reliable support for stable system operation.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A power system wiring detection method based on block parallel processing includes:

[0007] Real-time operational data is collected by sensing devices deployed at key nodes of the power system, and a basic dataset is generated by Kalman filtering for noise reduction and normalization.

[0008] A power system topology model is constructed based on the basic dataset. The topology is then divided into blocks based on voltage level hierarchy and equipment functional cluster characteristics. A sub-block identification system is established, and a unique topology code and data index are assigned.

[0009] Local data of each sub-block is obtained from the basic dataset and the data volume is statistically analyzed. Feature extraction is performed in the electrical feature dimension and the topology feature dimension respectively to obtain the task feature vector of each sub-block.

[0010] A distributed parallel computing architecture is built, consisting of one master control node and at least four computing nodes. The master control node constructs a priority model based on the importance weight of sub-blocks, data volume, and feature extraction complexity, calculates the priority coefficient of sub-block tasks, and uses a round-robin method to allocate tasks.

[0011] By comparing the feature vectors of each sub-block with the normal feature templates obtained during model training using a pre-trained anomaly detection model based on an improved random forest algorithm, preliminary detection results for each sub-block are obtained.

[0012] The main control node summarizes the preliminary detection results of each sub-block, extracts the boundary data of abnormal sub-blocks and their adjacent sub-blocks, and performs cross-sub-block boundary collaborative verification, including verifying the status of tie switches, cross-sub-block branch currents, and boundary bus voltages. It then merges the boundary collaborative verification and preliminary detection results to confirm the validity of the abnormal sub-blocks.

[0013] The main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-blocks, constructs a global wiring status map, compares it with the normal topology map, and locates the abnormal device and identifies the abnormal type by analyzing the position of the difference nodes and edges.

[0014] Based on the anomaly localization and type identification results, the anomaly location is re-measured with high precision. The re-measured data is then input into the anomaly detection model for secondary detection to verify the detection results.

[0015] Preferably, the step of collecting real-time operational data through sensing devices deployed at key nodes of the power system, and generating a basic dataset through Kalman filtering for noise reduction and normalization, specifically includes:

[0016] By deploying sensing devices at key nodes of the power system, real-time operating data of the power system is collected. The real-time operating data includes bus voltage, branch current, switch status signals, transformer operating parameters and line topology correlation data.

[0017] The collected real-time operating data is preprocessed by using an adaptive Kalman filter algorithm to eliminate random noise interference in the data, using Lagrange interpolation to fill in missing values ​​generated during the data acquisition process, and using the Z-Score standardization method to standardize electrical parameters to obtain the basic dataset.

[0018] Preferably, the step of constructing a power system topology model based on the basic dataset, and dividing the topology into blocks based on voltage level hierarchies and equipment functional cluster characteristics, establishing a sub-block identification system, and assigning unique topology codes and data indexes specifically includes:

[0019] Based on the acquired basic dataset, the types and distribution of all key equipment, including busbars, transformers, and switchgear, are summarized, and the actual connection relationships between each piece of equipment are obtained.

[0020] Using devices as nodes and the connecting lines between devices as edges, an adjacency matrix is ​​used to represent the connection relationships between devices, and a power system topology model is constructed.

[0021] Based on the voltage level hierarchical characteristics of the power system and combined with the functional cluster characteristics of equipment, equipment clusters that are at the same voltage level and can directly transmit power are divided into a sub-block.

[0022] Each sub-block after division contains at least one main power supply line, two busbar nodes and three key power equipment, and sets boundary constraints, including the state threshold of tie switches, the allowable range of voltage fluctuations of boundary buses and the current transmission limit value of cross-sub-block branches.

[0023] Each sub-block is assigned unique identification information, including a unique topology code and data index.

[0024] Preferably, the step of obtaining local data of each sub-block from the basic dataset and statistically analyzing the data, and performing feature extraction in both the electrical feature dimension and the topological feature dimension to obtain the task feature vector of each sub-block specifically includes:

[0025] Local data corresponding to each sub-block is extracted from the basic dataset, and multi-dimensional feature extraction is performed, including electrical feature dimension and topological feature dimension;

[0026] The electrical feature dimension extracts the transient features of branch current through wavelet transform algorithm, calculates the steady-state features of bus voltage through sliding window method, and obtains the active power, reactive power and power factor features of line by combining voltage and current data through power calculation formula.

[0027] The topological feature dimension constructs a local topological adjacency matrix based on the device connection relationship within the sub-block, generates a switch state vector based on the real-time status data of the device, and converts the switch's open / closed state into binary feature values.

[0028] Based on the extracted multi-dimensional features, normalization processing is performed to obtain the standardized feature vector of each sub-block.

[0029] Preferably, the distributed parallel computing architecture includes one master control node and at least four computing nodes. The master control node constructs a priority model based on the importance weight of sub-blocks, data volume, and feature extraction complexity, calculates the priority coefficients of sub-block tasks, and uses a round-robin method for task allocation, specifically including:

[0030] A distributed parallel computing architecture is constructed, which includes one master control node and at least four computing nodes. The master control node and the computing nodes interact with each other via Ethernet.

[0031] Construct a sub-block task priority model, the input parameters of which include the importance weight of the sub-block, the data size of the sub-block, and the feature extraction complexity of the sub-block;

[0032] The priority coefficient of each sub-block processing task is calculated based on the priority model, and the sub-block processing tasks are allocated to each computing node through a round-robin scheduling algorithm.

[0033] Preferably, the step of obtaining preliminary detection results for each sub-block by comparing the feature vectors of each sub-block with the normal feature templates obtained during model training using a pre-trained anomaly detection model based on an improved random forest algorithm specifically includes:

[0034] Each computing node processes tasks based on the assigned sub-blocks and obtains its standardized feature vector based on the sub-block identification information;

[0035] Historical power system operation data, including wiring anomaly data and normal operation data, are collected to form the original dataset for model training. An anomaly detection model is built based on the random forest algorithm. The number of decision trees is set according to the size of the training dataset and the feature dimension, and the model is trained.

[0036] Based on the trained anomaly detection model, the standardized feature vector of the current sub-block is compared with the normal feature template in the model;

[0037] If the matching degree between the feature vector and the normal feature template is greater than or equal to the threshold, a normal identifier is output; if the matching degree is less than the threshold, an abnormal identifier is output, and the deviation degree of the abnormal feature is calculated to obtain the preliminary detection result of the sub-block.

[0038] Preferably, the main control node summarizes the preliminary detection results of each sub-block, extracts the boundary data of the abnormal sub-block and its adjacent sub-blocks, and performs cross-sub-block boundary collaborative verification, including verifying the status of tie switches, cross-sub-block branch currents, and boundary bus voltages. It then fuses the boundary collaborative verification with the preliminary detection results to confirm the validity of the abnormal sub-block. Specifically, this includes:

[0039] The main control node receives the preliminary detection results of sub-blocks uploaded by all computing nodes, filters out sub-blocks marked with anomaly indicators, obtains the information of adjacent sub-blocks of each abnormal sub-block, and extracts the boundary data between the abnormal sub-block and its adjacent sub-blocks.

[0040] For the communication switch shared by the abnormal sub-block and the adjacent sub-block, the real-time status signal of the switch in the two sub-blocks is obtained and compared to perform a communication switch status consistency check.

[0041] Select the cross-sub-block branch connecting the abnormal sub-block and the adjacent sub-block, collect the current data of the branch on the abnormal sub-block side and the current data on the adjacent sub-block side, compare the numerical difference of the current data on both sides, and perform branch current continuity verification.

[0042] Determine the boundary busbar between the abnormal sub-block and the adjacent sub-block, collect real-time voltage data of the boundary busbar on both sides of the sub-block, obtain the changing trend of the voltage data on both sides, and perform collaborative verification of the boundary busbar voltage.

[0043] The results of boundary collaborative verification are matched with the preliminary detection results of sub-blocks to determine the validity of the abnormal identifiers of abnormal sub-blocks.

[0044] Preferably, the main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-blocks to construct a global wiring status map. This map is compared with the normal topology map. By analyzing the positions of differing nodes and edges, the abnormal device is located, and the specific abnormal type is identified, including:

[0045] The main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-blocks, and constructs a global wiring status map based on a three-layer structure, including the device layer, connection layer, and feature layer.

[0046] Retrieve the normal wiring topology map of the power system, compare the constructed global wiring status map with the normal wiring topology map element by element, and calculate the overall similarity.

[0047] If the similarity is less than 90%, a global wiring anomaly is identified. By analyzing the positions of the differing nodes and edges, the specific device and its connection port where the anomaly occurred are determined, and the anomaly type is identified.

[0048] Preferably, the step of performing a high-precision retest of the abnormal location based on the anomaly localization and type identification results, and inputting the retest data into the anomaly detection model for secondary detection, specifically includes:

[0049] Based on the anomaly localization and type identification results, retest data of the anomaly locations are continuously collected, and retest feature vectors are generated.

[0050] The retested feature vector is input into the anomaly detection model, and the presence of anomalies is determined by comparing its matching degree with that of normal feature templates.

[0051] The retest feature vector is compared with the feature vector of the first anomaly detection, and the consistency between the two is calculated. The second detection result is output to verify the first anomaly detection result.

[0052] Compared with the prior art, the advantages of the present invention are:

[0053] At the data processing level, Kalman filtering for noise reduction, Lagrange interpolation for completion, and Z-score standardization ensure the quality of the basic dataset, laying a precise data foundation for subsequent detection. In terms of block partitioning, sub-blocks are divided according to voltage level and equipment functional cluster characteristics, with boundary constraints set. This achieves both refined topology management and creates conditions for parallel computing. Regarding computational efficiency, a distributed parallel architecture combined with polling task allocation and a priority model, relying on multiple computing nodes to synchronously process sub-block tasks, significantly shortens detection time and adapts to the needs of large-scale power systems. In terms of detection accuracy, multi-dimensional feature extraction combined with an improved random forest anomaly detection model accurately captures electrical and topology anomaly signals; cross-sub-block boundary collaborative verification further filters out false positives, ensuring the validity of anomaly sub-blocks. In the anomaly localization stage, comparison of the global wiring status map with the normal map accurately locates anomaly devices and ports and identifies their types; high-precision retesting and secondary detection mechanisms provide dual verification results, significantly reducing missed and false detection rates, providing efficient and reliable technical support for the stable operation of the power system. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the method proposed in this invention;

[0055] Figure 2 This is a schematic diagram illustrating the generation of the basic dataset proposed in this invention;

[0056] Figure 3 This is a schematic diagram of the power system topology model proposed in this invention;

[0057] Figure 4 This is a schematic diagram of the sub-block task feature vector acquisition method proposed in this invention;

[0058] Figure 5 This is a schematic diagram of the distributed parallel computing architecture proposed in this invention;

[0059] Figure 6 This is a schematic diagram illustrating the preliminary detection results of the sub-block proposed in this invention;

[0060] Figure 7 This is a schematic diagram illustrating the verification of the validity of the abnormal sub-block proposed in this invention;

[0061] Figure 8 This is a schematic diagram of the anomaly location device and anomaly type identification proposed in this invention;

[0062] Figure 9 This is a schematic diagram of the secondary detection proposed in this invention. Detailed Implementation

[0063] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0064] See Figure 1 As shown, the power system wiring detection method based on block parallel processing includes:

[0065] Step 1: Collect real-time operational data by deploying sensing devices at key nodes of the power system, and generate a basic dataset by denoising and normalizing the data using Kalman filtering.

[0066] Step 2: Construct a power system topology model based on the basic dataset, and divide the topology into blocks based on voltage level hierarchy and equipment functional cluster characteristics, establish a sub-block identification system, and assign unique topology codes and data indexes.

[0067] Step 3: Obtain local data of each sub-block from the basic dataset and count the data volume. Perform feature extraction in the electrical feature dimension and topology feature dimension respectively to obtain the task feature vector of each sub-block.

[0068] Step 4: Build a distributed parallel computing architecture, including one master control node and at least four computing nodes. The master control node constructs a priority model based on the importance weight of sub-blocks, data volume, and feature extraction complexity, calculates the priority coefficient of sub-block tasks, and uses a round-robin method to allocate tasks.

[0069] Step 5: Using a pre-trained anomaly detection model based on an improved random forest algorithm, compare the feature vectors of each sub-block with the normal feature templates obtained during model training to obtain preliminary detection results for the sub-blocks;

[0070] Step 6: The main control node summarizes the preliminary detection results of each sub-block, extracts the boundary data of abnormal sub-blocks and their adjacent sub-blocks, and performs cross-sub-block boundary collaborative verification, including verifying the status of tie switches, cross-sub-block branch currents, and boundary bus voltages. The boundary collaborative verification and preliminary detection results are then merged to confirm the validity of the abnormal sub-blocks.

[0071] Step 7: The main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-block, constructs a global wiring status map, compares it with the normal topology map, and locates the abnormal device and identifies the abnormal type by analyzing the position of the difference nodes and edges.

[0072] Step 8: Based on the anomaly location and type identification results, perform high-precision retesting of the anomaly location, input the retest data into the anomaly detection model for secondary detection, and verify the detection results.

[0073] See Figure 2 As shown, real-time operational data is collected by sensing devices deployed at key nodes of the power system, and then processed through Kalman filtering for noise reduction and normalization to generate a basic dataset, which specifically includes:

[0074] By deploying sensing devices at key nodes of the power system, real-time operating data of the power system is collected. The real-time operating data includes bus voltage, branch current, switch status signals, transformer operating parameters and line topology correlation data.

[0075] The collected real-time operating data is preprocessed by using an adaptive Kalman filter algorithm to eliminate random noise interference in the data, using Lagrange interpolation to fill in missing values ​​generated during the data acquisition process, and using the Z-Score standardization method to standardize electrical parameters to obtain the basic dataset.

[0076] See Figure 3 As shown, a power system topology model is constructed based on the basic dataset. The topology is then divided into blocks based on voltage level hierarchies and equipment functional cluster characteristics. A sub-block identification system is established, and unique topology codes and data indexes are assigned. Specifically, this includes:

[0077] Based on the acquired basic dataset, the types and distribution of all key equipment, including busbars, transformers, and switchgear, are summarized, and the actual connection relationships between each piece of equipment are obtained.

[0078] Using devices as nodes and the connecting lines between devices as edges, an adjacency matrix is ​​used to represent the connection relationships between devices, and a power system topology model is constructed.

[0079] Based on the voltage level hierarchical characteristics of the power system and combined with the functional cluster characteristics of equipment, equipment clusters that are at the same voltage level and can directly transmit power are divided into a sub-block.

[0080] Each sub-block after division contains at least one main power supply line, two busbar nodes and three key power equipment, and sets boundary constraints, including the state threshold of tie switches, the allowable range of voltage fluctuations of boundary buses and the current transmission limit value of cross-sub-block branches.

[0081] Each sub-block is assigned unique identification information, including a unique topology code and data index.

[0082] Specifically, the state threshold of the inter-sub-block interconnection switch specifies the criteria for determining the normal opening and closing state of the interconnection switch, such as the normal duration range of "continuously closed" and "continuously open"; the allowable range of voltage fluctuation of the boundary busbar specifies the maximum and minimum values ​​that the boundary busbar voltage can fluctuate during normal operation; and the current transmission limit value of the cross-sub-block branch determines the maximum current value that the cross-sub-block line can carry under the premise of safe operation, so as to avoid the risk of overcurrent.

[0083] By recording information in text, the identification information of sub-blocks is matched one-to-one with the equipment list and boundary constraints within the sub-blocks, ensuring that sub-blocks can be quickly located and all relevant data of the sub-blocks can be retrieved through the identification.

[0084] See Figure 4 As shown, local data for each sub-block is obtained from the basic dataset, and the data volume is statistically analyzed. Feature extraction is performed in both the electrical and topological feature dimensions to obtain the task feature vector for each sub-block. Specifically, this includes:

[0085] Local data corresponding to each sub-block is extracted from the basic dataset, and multi-dimensional feature extraction is performed, including electrical feature dimension and topological feature dimension;

[0086] The electrical feature dimension extracts the transient features of branch current through wavelet transform algorithm, calculates the steady-state features of bus voltage through sliding window method, and obtains the active power, reactive power and power factor features of line by combining voltage and current data through power calculation formula.

[0087] The topological feature dimension constructs a local topological adjacency matrix based on the device connection relationship within the sub-block, generates a switch state vector based on the real-time status data of the device, and converts the switch's open / closed state into binary feature values.

[0088] Based on the extracted multi-dimensional features, normalization processing is performed to obtain the standardized feature vector of each sub-block.

[0089] Specifically, the transient characteristics of branch current are processed using wavelet transform algorithm to capture the characteristics of current during sudden changes, specifically extracting the peak value of transient current (the maximum value of a sudden increase in current), attenuation coefficient (the rate at which current decreases from the peak value), and oscillation frequency (the number of cycles of current fluctuation).

[0090] The steady-state characteristics of the bus voltage are analyzed using the sliding window method to calculate the effective voltage value (equivalent stable voltage value), voltage deviation rate (the percentage difference between the actual voltage and the rated voltage), and three-phase voltage imbalance (the degree of difference between the three-phase voltages).

[0091] The power characteristics of the line are derived by using the power calculation logic common to power systems and combining voltage and current data to obtain the active power (the power actually doing work), reactive power (the power maintaining the electric or magnetic field) and power factor (the ratio of active power to total power), thus completing the extraction of electrical characteristic dimensions.

[0092] The local topology features are constructed based on the actual connection of buses, lines and equipment within the sub-block. The local topology adjacency matrix of the sub-block is then constructed, and topology parameters are extracted from the matrix, including node degree (the number of other devices connected to a single device), edge density (the ratio of the actual number of lines in the sub-block to the maximum possible number of lines), and loop number (the number of closed loops formed within the sub-block).

[0093] The switch status feature is obtained by collecting the real-time status signals (open or closed) of all switching devices in the sub-block, converting the "closed" status into a unified binary identifier and the "open" status into another binary identifier, forming a switch status vector containing all switch statuses, thus completing the extraction of topological feature dimensions.

[0094] See Figure 5 As shown, a distributed parallel computing architecture is constructed, including one master control node and at least four computing nodes. The master control node constructs a priority model based on the importance weight of sub-blocks, data volume, and feature extraction complexity, calculates the priority coefficients of sub-block tasks, and uses a round-robin method for task allocation, specifically including:

[0095] A distributed parallel computing architecture is constructed, which includes one master control node and at least four computing nodes. The master control node and the computing nodes interact with each other via Ethernet.

[0096] Construct a sub-block task priority model, the input parameters of which include the importance weight of the sub-block, the data size of the sub-block, and the feature extraction complexity of the sub-block;

[0097] The priority coefficient of each sub-block processing task is calculated based on the priority model, and the sub-block processing tasks are allocated to each computing node through a round-robin scheduling algorithm.

[0098] Specifically, the importance weight of sub-blocks is determined based on the power supply attributes of the area where the sub-block is located. The weight of sub-blocks in the core load power supply area is set to 0.8-1.0, and the weight of sub-blocks in non-core areas is set to 0.3-0.6. The size of sub-block data is determined by the total amount of data in the local dataset of each sub-block. The larger the data volume, the longer the task processing time may be, and this needs to be taken into consideration in the priority. The complexity of sub-block feature extraction is determined based on the number of device types and feature dimensions within the sub-block, such as whether it includes complex transient feature extraction requirements. The higher the complexity, the more appropriate the priority needs to be adjusted.

[0099] Based on the established priority model, a comprehensive calculation is performed on each sub-block processing task. The importance weight of the sub-block, the data volume, and the feature extraction complexity are assigned reasonable influence weights, and the priority coefficient of each task is obtained through comprehensive evaluation. The higher the coefficient value, the higher the urgency and importance of the sub-block processing task, and the more computing resources should be allocated to it.

[0100] By employing a round-robin algorithm, tasks are sorted from high to low priority and then distributed to each computing node in a cyclical manner. After distribution, the resource usage status of each computing node is monitored to ensure that the difference in CPU utilization between nodes is controlled within ±15% and that the memory utilization of all nodes does not exceed 80%.

[0101] See Figure 6 As shown, by using a pre-trained anomaly detection model based on an improved random forest algorithm, and comparing the feature vectors of each sub-block with the normal feature templates obtained during model training, the preliminary detection results for each sub-block are obtained, specifically including:

[0102] Each computing node processes tasks based on the assigned sub-blocks and obtains its standardized feature vector based on the sub-block identification information;

[0103] Historical power system operation data, including wiring anomaly data and normal operation data, are collected to form the original dataset for model training. An anomaly detection model is built based on the random forest algorithm. The number of decision trees is set according to the size of the training dataset and the feature dimension, and the model is trained.

[0104] Based on the trained anomaly detection model, the standardized feature vector of the current sub-block is compared with the normal feature template in the model;

[0105] If the matching degree between the feature vector and the normal feature template is greater than or equal to the threshold, a normal identifier is output; if the matching degree is less than the threshold, an abnormal identifier is output, and the deviation degree of the abnormal feature is calculated to obtain the preliminary detection result of the sub-block.

[0106] Specifically, the wiring anomaly data covers operational data for typical abnormal scenarios such as switch malfunction, incorrect line connection, and incorrect transformer wiring group; normal operation data includes similar data for power system periods with no anomalies and stable operation.

[0107] The model is built based on the random forest algorithm. First, the number of decision trees, the core building block of the model, is determined. Based on the size of the training dataset and the feature dimension, the number of decision trees is set to 100-200. The model parameters are optimized by cross-validation. Specifically, the training dataset is divided into multiple subsets. The model is trained on some subsets and validated on the rest. The decision criteria for splitting nodes are adjusted repeatedly until the detection accuracy of the model on the validation set reaches the preset requirements.

[0108] The processed training dataset is input into the improved random forest model to start the model training process. The model learns the feature patterns of normal data, extracts the feature distribution patterns of various sub-blocks under normal conditions, and forms normal feature templates. At the same time, the model will memorize the feature deviation patterns under different abnormal scenarios, establish a preliminary correspondence between abnormal features and abnormal types, and provide a basis for judgment for subsequent detection.

[0109] The core of matching the standardized feature vector with the normal feature template is to quantify the difference between the current feature and the normal template. Let the standardized feature vector of the current sub-block be... Normal feature template is The formula is:

[0110]

[0111] in, For feature matching degree, This is the standardized feature vector of the current sub-block. The normal feature template obtained from model training. The dimension of the feature vector. This is the k-th element of the current feature vector. The k-th element of the normal feature template. It is a local minimum.

[0112] See Figure 7 As shown, the main control node summarizes the preliminary detection results of each sub-block, extracts the boundary data of the abnormal sub-block and its adjacent sub-blocks, and performs cross-sub-block boundary collaborative verification, including verifying the status of tie switches, cross-sub-block branch currents, and boundary bus voltages. It then merges the boundary collaborative verification results with the preliminary detection results to confirm the validity of the abnormal sub-block. Specifically, this includes:

[0113] The main control node receives the preliminary detection results of sub-blocks uploaded by all computing nodes, filters out sub-blocks marked with anomaly indicators, obtains the information of adjacent sub-blocks of each abnormal sub-block, and extracts the boundary data between the abnormal sub-block and its adjacent sub-blocks.

[0114] For the communication switch shared by the abnormal sub-block and the adjacent sub-block, the real-time status signal of the switch in the two sub-blocks is obtained and compared to perform a communication switch status consistency check.

[0115] Select the cross-sub-block branch connecting the abnormal sub-block and the adjacent sub-block, collect the current data of the branch on the abnormal sub-block side and the current data on the adjacent sub-block side, compare the numerical difference of the current data on both sides, and perform branch current continuity verification.

[0116] Determine the boundary busbar between the abnormal sub-block and the adjacent sub-block, collect real-time voltage data of the boundary busbar on both sides of the sub-block, obtain the changing trend of the voltage data on both sides, and perform collaborative verification of the boundary busbar voltage.

[0117] The results of boundary collaborative verification are matched with the preliminary detection results of sub-blocks to determine the validity of the abnormal identifiers of abnormal sub-blocks.

[0118] Specifically, in the consistency verification of the status of the tie switch, if the two statuses are consistent, both being closed or both being open, the tie switch is determined to be in normal status; if the two statuses are inconsistent, the duration of the status deviation is recorded. If the deviation duration exceeds 500 milliseconds, it is determined to be an abnormal boundary status, and the position of the tie switch is marked.

[0119] In the continuity verification of branch current across sub-blocks, if the difference value does not exceed 5% of the rated value of the branch current, the branch current transmission is considered continuous and the status is normal; if the difference value exceeds 5% of the rated value of the current, it indicates that there is an abnormal change in the branch current at the boundary, which is judged as an abnormal boundary current, and the identification information of the branch is recorded.

[0120] During the joint verification of boundary bus voltage, if the voltage change trends on both sides are consistent and the voltage difference does not exceed 3% of the rated voltage of the bus, the boundary bus voltage is judged to be normal; if the voltage change trends on both sides are opposite and the voltage difference exceeds 3% of the rated voltage, the boundary voltage is judged to be abnormal, and the number of the boundary bus is recorded.

[0121] For boundary devices of abnormal sub-blocks and adjacent sub-blocks, extract the status data and electrical quantity data of these devices within 10 consecutive sampling periods, and organize them into a time series in chronological order. Calculate the similarity of the time series of the same boundary devices on the two sub-block sides using a dynamic time warping algorithm: if the similarity is higher than 85%, the timing is considered consistent and the status is normal; if the similarity is lower than 85%, the timing consistency is considered abnormal. Combine this result with previous results of boundary status anomalies, boundary current anomalies, and boundary voltage anomalies. If any one of these anomaly types exists, it is determined that there is a cooperative anomaly at that boundary.

[0122] See Figure 8 As shown, the main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-blocks to construct a global wiring status map. This map is compared with the normal topology map. By analyzing the positions of differing nodes and edges, the abnormal device is located, and the specific abnormal type is identified, including:

[0123] The main control node retrieves the collected data, feature vectors, and boundary verification data of the abnormal sub-blocks, and constructs a global wiring status map based on a three-layer structure, including the device layer, connection layer, and feature layer.

[0124] Retrieve the normal wiring topology map of the power system, compare the constructed global wiring status map with the normal wiring topology map element by element, and calculate the overall similarity.

[0125] If the similarity is less than 90%, a global wiring anomaly is identified. By analyzing the positions of the differing nodes and edges, the specific device and its connection port where the anomaly occurred are determined, and the anomaly type is identified.

[0126] Specifically, the diagram includes the following layers: Equipment layer: All equipment in the power system is marked one by one in the diagram, including equipment model, equipment parameters, and real-time operating status; Connection layer: For the connection relationships between equipment, the corresponding equipment is connected by edges in the diagram, and the type, physical length, and impedance parameters of each connection line are marked to clarify the actual connection method between equipment; Feature layer: Key electrical feature values ​​are marked next to the corresponding equipment and lines, such as the real-time effective voltage value of the bus, the real-time active power of the line, and the power transmission value of the transformer, forming a global diagram integrating equipment, connection, and feature.

[0127] A detailed analysis of the differences between the global wiring status diagram and the normal wiring topology diagram is conducted:

[0128] First, locate the discrepancy nodes and check which device nodes in the global map, such as a switch, a line, or a transformer, have inconsistent labeling information with the normal map. These nodes are suspected abnormal devices.

[0129] Next, locate the difference edge and check whether the connection edge between devices in the global map matches the normal map. For example, check whether there are redundant connections, missing connections or incorrect connections, and determine the specific line of the abnormal connection.

[0130] Finally, the details of the anomaly location are clarified, and the specific location of the anomaly is determined by combining the difference nodes and difference edges, such as "the No. 2 connection port of a 110kV switch" or "the two ends of a 10kV line are incorrectly connected to busbars No. 2 and No. 5", so as to achieve precise location of the anomaly.

[0131] Retrieve the standardized feature vectors of valid anomaly sub-blocks and check which features deviate significantly; match the current anomaly features with the historical anomaly case library stored in the system to verify whether the feature deviation pattern is consistent with a certain type of historical anomaly; if the current feature completely matches the feature pattern of a certain type of anomaly and highly matches the corresponding case in the historical case library, the anomaly type can be determined.

[0132] See Figure 9 As shown, based on the anomaly localization and type identification results, the anomaly location is re-measured with high precision. The re-measured data is then input into the anomaly detection model for secondary detection. The verification of the detection results specifically includes:

[0133] Based on the anomaly localization and type identification results, retest data of the anomaly locations are continuously collected, and retest feature vectors are generated.

[0134] The retested feature vector is input into the anomaly detection model, and the presence of anomalies is determined by comparing its matching degree with that of normal feature templates.

[0135] The retest feature vector is compared with the feature vector of the first anomaly detection, and the consistency between the two is calculated. The second detection result is output to verify the first anomaly detection result.

[0136] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0137] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for power system connection detection based on block-parallel processing, characterized in that, The application relates to a power system anomaly detection method and system. Real-time operation data is collected by a sensing device arranged at a key node of a power system, and basic data sets are generated by Kalman filter denoising and normalization processing; A power system topology structure model is constructed based on the basic data set data, and the model is blocked based on voltage level layering of the topology structure and device function cluster characteristics, a sub-block identification system is established, and unique topology codes and data indexes are allocated; Local data of each sub-block is obtained from the basic data set, and the data volume is counted, and feature extraction is performed in the electrical characteristic dimension and the topology characteristic dimension to obtain a feature vector of each sub-block task; A distributed parallel computing architecture is built, including one main control node and at least four computing nodes, the main control node constructs a priority model based on the importance weight, data volume and feature extraction complexity of the sub-block, calculates a sub-block task priority coefficient, and adopts a polling method to perform task allocation; An anomaly detection model based on an improved random forest algorithm is constructed, and the model is compared with a normal feature template obtained by training to obtain a preliminary detection result of the sub-block; The main control node collects the preliminary detection results of the sub-blocks, extracts abnormal sub-blocks and their adjacent sub-block boundary data, performs cross-sub-block boundary collaborative verification, including verifying the state of a tie-in switch, the branch current across the sub-block and the voltage of a boundary bus, fuses the boundary collaborative verification and the preliminary detection result, and confirms the validity of the abnormal sub-blocks; The main control node calls the acquisition data, the feature vector and the boundary verification data of the abnormal sub-block, constructs a global wiring state atlas, compares the atlas with a normal topology atlas, analyzes the positions of the difference nodes and edges, locates the abnormal equipment, and identifies the abnormal type; Based on the abnormal positioning and type identification result, high-precision re-measurement is performed on the abnormal position, and the re-measurement data is input into the anomaly detection model for secondary detection to verify the detection result.

2. The method for power system connection detection based on block-parallel processing according to claim 1, characterized in that, The real-time operation data is collected by a sensing device arranged at a key node of a power system, and basic data sets are generated by Kalman filter denoising and normalization processing; Real-time operation data of the power system is collected by a sensing device arranged at each key node of the power system, and the real-time operation data includes bus voltages, branch currents, switch device state signals, transformer operation parameters and line topology correlation data; The collected real-time operation data is preprocessed, random noise interference in the data is eliminated by an adaptive Kalman filter algorithm, missing values generated in the data acquisition process are completed by a Lagrange interpolation method, electrical parameters are standardized based on a Z-Score standardization method, and basic data sets are obtained.

3. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, Based on the obtained basic data sets, the types and distributions of all key devices, including busbars, transformers and switch devices, are summarized, and the actual connection relationship between the devices is obtained; The connection relationship between the devices is represented in the form of an adjacency matrix, and a power system topology structure model is constructed. ​ Based on the voltage level hierarchical characteristics of the power system, combined with the equipment function cluster characteristics, the equipment clusters at the same voltage level and capable of direct power transmission are divided into a sub-block; For each divided sub-block, each sub-block contains at least one main supply line, two bus nodes and three key power transformation equipment, and boundary constraint conditions are set, including the state threshold of the tie-in switch, the voltage fluctuation allowed range of the boundary bus and the current transmission limit value of the cross-sub-block branch; Each sub-block is assigned a unique identification information, including a dedicated topology code and a data index.

4. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, The local data of each sub-block is obtained from the basic data set, and the data volume is counted, and feature extraction is performed in the electrical characteristic dimension and the topology characteristic dimension to obtain the task feature vector of each sub-block, which specifically includes: The local data corresponding to each sub-block is extracted from the basic data set, and multi-dimensional feature extraction is performed, including the electrical characteristic dimension and the topology characteristic dimension; The electrical characteristic dimension extracts the transient characteristics of the branch current through the wavelet transform algorithm, calculates the steady-state characteristics of the bus voltage through the sliding window method, and calculates the active power, reactive power and power factor characteristics of the line through the power calculation formula combined with the voltage and current data; The topology characteristic dimension constructs a local topology adjacency matrix based on the connection relationship of the equipment in the sub-block, generates a switch state vector based on the real-time state data of the equipment, and converts the on-off state of the switch into a binary feature value; Based on the extracted multi-dimensional features, normalization processing is performed to obtain the standardized feature vector of each sub-block.

5. The method for power system connection detection based on block-parallel processing according to claim 4, characterized in that, The distributed parallel computing architecture is built, including 1 main control node and at least 4 computing nodes, the main control node constructs a priority model based on the importance weight, data volume and feature extraction complexity of the sub-block, calculates the priority coefficient of the sub-block task, and the task allocation is performed by using the polling method, which specifically includes: The distributed parallel computing architecture is built, including 1 main control node and at least 4 computing nodes, the main control node and the computing node interact data through Ethernet; A sub-block task priority model is constructed, and the input parameters of the priority model include the importance weight of the sub-block, the data volume of the sub-block, and the feature extraction complexity of the sub-block; Based on the priority model, the priority coefficient of each sub-block processing task is calculated, and the sub-block processing task is allocated to each computing node by using the polling scheduling algorithm.

6. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, The abnormal detection model constructed based on the improved random forest algorithm is trained in advance, the task feature vector of each sub-block is compared with the normal feature template obtained by model training, and the preliminary detection result of the sub-block is obtained, which specifically includes: Each computing node obtains the standardized feature vector based on the sub-block identification information based on the allocated sub-block processing task; The historical operation data of the power system is collected, including abnormal connection data and normal operation data, to form a model training original data set, and an abnormal detection model is constructed based on the random forest algorithm, the number of decision trees is set based on the size and feature dimension of the training data set, and the model is trained; Based on the trained abnormal detection model, the standardized feature vector of the current sub-block is compared with the normal feature template in the model. If the matching degree of the eigenvector with the normal feature template is greater than or equal to the threshold value, a normal identification is output, and if the matching degree is less than the threshold value, an abnormal identification is output, and an abnormal feature deviation degree is calculated to obtain a sub-block preliminary detection result.

7. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, The main control node collects the sub-block preliminary detection results, extracts abnormal sub-block and adjacent sub-block boundary data, and performs cross-sub-block boundary collaborative verification, including verifying the state of the tie-in switch, the cross-sub-block branch current, and the boundary bus voltage, fusing the boundary collaborative verification and the preliminary detection result, and confirming the effectiveness of the abnormal sub-block, specifically including: The main control node receives the sub-block preliminary detection results uploaded by all computing nodes, filters out the sub-blocks labeled with abnormal identification, obtains the adjacent sub-block information of each abnormal sub-block, and extracts the boundary data of the abnormal sub-block and the adjacent sub-block; For the tie-in switch shared by the abnormal sub-block and the adjacent sub-block, the real-time state signals of the switch in the two sub-blocks are obtained and compared to perform consistency verification of the state of the tie-in switch; The cross-sub-block branch connecting the abnormal sub-block and the adjacent sub-block is selected, the current data on the abnormal sub-block side and the current data on the adjacent sub-block side are collected, the numerical difference between the two sides is compared, and the continuity of the branch current is verified; The boundary bus of the abnormal sub-block and the adjacent sub-block is determined, the real-time voltage data on both sides of the boundary bus is collected, the change trend of the voltage data on both sides is obtained, and the collaborative verification of the voltage of the boundary bus is performed; The various results of the boundary collaborative verification are matched with the sub-block preliminary detection results to determine the effectiveness of the abnormal identification of the abnormal sub-block.

8. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, The main control node retrieves the collected data, eigenvector, and boundary verification data of the abnormal sub-block, constructs a global wiring state graph, and compares it with the normal topology graph to locate the abnormal equipment and identify the abnormal type, specifically including: The main control node retrieves the collected data, eigenvector, and boundary verification data of the abnormal sub-block, and constructs a global wiring state graph based on a three-layer structure, including a device layer, a connection layer, and a feature layer. The normal wiring topology graph of the power system is retrieved, the constructed global wiring state graph is compared with the normal wiring topology graph element by element, and the overall similarity is calculated. If the similarity is less than 90%, it is determined that there is a global wiring abnormality, the positions of the difference nodes and edges are analyzed to determine the specific equipment and the specific connection port of the equipment where the abnormality occurs, and the abnormal type is identified.

9. The method for power system connection detection based on block-parallel processing according to claim 1, wherein, Based on the abnormal positioning and type identification results, the abnormal position is re-measured with high precision, the re-measured data is input into the abnormal detection model for secondary detection, and the detection result is verified, specifically including: Based on the abnormal positioning and type identification results, the re-measured data of the abnormal position is continuously collected, and a re-measured eigenvector is generated; The re-measured eigenvector is input into the abnormal detection model, and the matching degree with the normal feature template is compared to determine whether there is an abnormality; The re-measured eigenvector and the eigenvector of the first abnormal detection are compared, the consistency of the two is calculated, the second detection result is output, and the first abnormal detection result is verified.

Images