RPA-based marketing meter box archive topology automatic correction method and system

Abstract

The application discloses a marketing metering box file topology automatic rectification method and system based on RPA, relates to the technical field of power analysis, and solves the technical problem that the accuracy of rationality analysis and rectification of the metering box file topology is not considered in terms of node fault type probability, space-time correlation, and multi-dimensional data such as characteristic data; the DTW distance between nodes is calculated to generate a correlation matrix, and the dynamic correlation hidden in the topology is revealed, which is helpful to identify the space-time correlation of the nodes. According to the topology structure after rectification of the nodes, the original digital topology structure is rectified, multi-dimensional risk quantification, hybrid algorithm optimization, and hierarchical rectification strategies are used, the accuracy, intelligence and self-adaptation of the marketing metering box file topology rectification are realized, the node comprehensive risk, the topology importance and the algorithm efficiency are deeply integrated, and the rectification efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of power analysis, specifically an automated correction method and system for marketing metering box archive topology diagrams based on RPA. Background Technology

[0002] Robotic Process Automation (RPA) automates repetitive and routine tasks by simulating human operations on a computer. By using RPA robots to simulate human actions, combined with image recognition (OCR / CV) and rule engines, it automatically compares the marketing metering kit file topology diagram with a standard template, identifies offset nodes, incorrect connections, or missing data, and triggers a correction process, achieving detection-analysis-correction. Marketing metering kit files contain critical data such as equipment information, connection relationships, and operating status, requiring a more automated system.

[0003] Existing automated correction solutions for marketing metering box archive topology maps only use historical traffic data to correct the topology maps. They do not consider the accuracy of the rationality analysis and correction of the metering box archive topology maps based on multi-dimensional data such as node failure type probability, spatiotemporal correlation, and feature data. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; to this end, this invention proposes an automated correction method and system for marketing meter box archive topology diagrams based on RPA, which is used to solve the technical problem of the accuracy of the rationality analysis and correction of meter box archive topology diagrams without considering the probability of node failure types, spatiotemporal correlation, and feature data and other multi-dimensional data.

[0005] To address the aforementioned problems, a first aspect of the present invention provides an automated correction method for the marketing metering box archive topology diagram based on RPA, comprising the following steps:

[0006] The RPA robot automatically logs into the marketing management system, GIS system, and design drawing library, extracts the topology map and related data of the metering box archives, extracts the text information in the topology map through OCR text recognition, and identifies the node positions, connection lines and equipment icon types through computer vision to construct a digital topology structure.

[0007] Acquire historical traffic data for each node to identify metering box nodes, user nodes, and transformer nodes, and set the hierarchy of each node based on historical traffic data and node historical data;

[0008] Calculate the DTW dynamic time warping distance between nodes, identify spatiotemporal correlations, and extract features from node data for nodes with different spatiotemporal correlations and levels.

[0009] Based on the extracted feature data, a fault classification model is trained for nodes with different spatiotemporal correlations, and the probability of node fault type is output.

[0010] Based on the probability analysis results of fault types of each node, node hierarchical data, and node degree centrality, and by introducing hierarchical coefficients, the comprehensive risk of nodes is calculated. Through the improved GAPSO algorithm, fitness functions and iteration termination conditions are set, and iterative calculations are performed to obtain the corrected topology structure of each level of nodes. The original digital topology structure is then corrected based on the corrected topology structure of the nodes.

[0011] Optionally, in one example of the above aspects, obtaining historical traffic data for each node to identify metering box nodes, user nodes, and transformer nodes, and obtaining metering box data, business system data, and external environment data for the nodes, includes the following steps:

[0012] Historical flow data of metering box nodes, user nodes and transformer nodes are obtained, the corresponding node types of the flow data are labeled, and the data is input into the LSTM model to train the model to identify the node types.

[0013] Historical traffic data for each node is obtained and input into the trained LSTM model to identify the node type.

[0014] Optionally, in one example of the above aspects, setting each node level based on historical traffic data and node historical data includes the following steps:

[0015] Based on node type data, user nodes are divided into access layer nodes;

[0016] Based on historical traffic data and node historical data, nodes that meet the following conditions are selected: daily average traffic is greater than the 80th percentile of all nodes; traffic standard deviation is less than the 20th percentile of all nodes; and the number of directly connected nodes is greater than 10.

[0017] The selected nodes will be used as core layer nodes;

[0018] The remaining unclassified nodes are classified as aggregation layer nodes.

[0019] Optionally, in one example of the above aspects, calculating the DTW dynamic time warping distance between nodes and identifying spatiotemporal correlations includes the following steps:

[0020] Obtain the historical traffic data for each node, convert the historical traffic data of the node into equal-length time series X and Y, with lengths |X| and |Y| respectively, and ensure that the timestamps of all time series are aligned.

[0021] Calculate the Euclidean distance D(i,j) between any two nodes to form an m×n distance matrix, where m and n are the sequence lengths;

[0022] Create a cumulative distance matrix D of (m+1)×(n+1), D(0,0)=0, and set the first row and first column to infinity;

[0023] Form a node set from all nodes, and calculate the Euclidean distance between any two nodes using the recursive formula: D(i,j) = d(i,j) + min(D(i-1,j), D(i,j-1),D(i-1,j-1)), where d(i,j) is the flow difference between node i and node j within the time series range, D(i-1,j) is the Euclidean distance between node i-1 and node j, D(i,j-1) is the Euclidean distance between node i and node j-1, and D(i-1,j-1) is the Euclidean distance between node i-1 and node j-1.

[0024] Using DTW distance as edge weights, a spatial relationship matrix between nodes is established;

[0025] DTW distances at different time granularities (daily, weekly, and monthly) are calculated to obtain multi-scale DTW distance features.

[0026] Multi-scale DTW distance matrices are fused using weighted summation or attention mechanisms to form a spatiotemporally correlated DTW distance matrix;

[0027] A hierarchical tree is constructed using the spatiotemporal correlation DTW distance matrix, and clustering results are obtained through pruning.

[0028] We directly use the DTW distance as a similarity metric to divide the nodes into K clusters;

[0029] Treating nodes as graph vertices and using the reciprocal of the DTW distance as edge weights, the Louvain algorithm is used for node clustering.

[0030] Nodes clustered into a unified cluster are considered as spatiotemporally relevant nodes.

[0031] Optionally, in one example of the above aspects, feature extraction is performed on node data for nodes with different spatiotemporal correlations and levels, including the following steps:

[0032] Based on the historical traffic data of each node in the spatiotemporal correlation nodes, a time-shifted sliding window W is introduced to calculate the feature data of the data within the window W, including: mean, variance, slope and frequency domain features.

[0033] Based on the node hierarchy, the transmission delay time of the upstream nodes connected to the node and the load of the downstream nodes connected to the node are statistically analyzed.

[0034] The ratio of the transmission delay time of the upstream node to the average transmission delay time in the historical data of the entire meter box archive topology is calculated to form the propagation delay coefficient matrix of the upstream node.

[0035] The ratio of the load of the downstream node connected to the computing node to the average node load in the historical data of the entire metering box archive topology diagram constitutes the downstream node load influence coefficient matrix.

[0036] The obtained coefficient matrix and feature data are used as the feature data of the nodes.

[0037] Optionally, in one example of the above aspects, a fault classification model for nodes with different spatiotemporal correlations is trained based on the extracted feature data, and the probability of node fault type is output, including the following steps:

[0038] The feature data of the nodes are fused according to different time granularities of daily, weekly and monthly. The weights are dynamically allocated using an attention mechanism to obtain the fused feature data and the feature matrix of the node: (Nn,Twindows,G), where Nn is the node number, Twindows is the feature sequence and G corresponds to the fused feature data.

[0039] Extract the topological structure of the connection relationships of nodes with different spatiotemporal correlations, number the nodes with different spatiotemporal correlations, and obtain the adjacency matrix of the connection relationships of each node based on the node numbering information of the adjacent nodes of the nodes with different spatiotemporal correlations.

[0040] Constructing a GCN-LSTM fault classification model:

[0041] Establish a GCN layer, with the input being the node feature matrix and the adjacency matrix;

[0042] By aggregating node feature matrix and adjacency matrix information through the GCN layer, the spatiotemporal feature matrix of the node is output.

[0043] Connect the output of the GCN layer to the input of the LSTM layer, and input the spatiotemporal feature matrix output by the GCN into the LSTM layer;

[0044] The LSTM layer outputs time-coded features. The features output by the LSTM layer are then input into a fully connected layer + Softmax layer to output the probability of the fault type.

[0045] Optionally, in one example of the above aspects, based on the probability analysis results of each node's failure type, node hierarchical data, and node degree centrality, and by introducing a hierarchical coefficient, the comprehensive risk of a node is calculated, including the following steps:

[0046] Obtain the probability analysis results of each node's failure type, node-level data, and node degree centrality to quantify node risk;

[0047] Calculate node risk value Where, Ri is the overall failure probability;

[0048] Li is the hierarchical coefficient, with the core layer hierarchical coefficient = 3, the aggregation layer hierarchical coefficient = 2, and the access layer hierarchical coefficient = 1.

[0049] Di′ represents the normalized degree centrality = number of node connections / maximum possible number of connections, and β is the weight that controls the influence of centrality.

[0050] Calculate the link risk value Sij from node i to connected node j:

[0051] Where Delayij is the delay time from node i to connected node j, Max_Delay is the maximum transmission delay time in the historical data of the entire meter box archive topology, Lossij is the energy loss of power transmission from node i to connected node j, Max_Loss is the maximum energy loss of power transmission between nodes in the historical data of the entire meter box archive topology, and wloss and wdelay are the corresponding weights, wloss=0.3 and wdelay=0.7 respectively;

[0052] The average of the link risk values ​​of the obtained nodes and their connected nodes is then weighted and averaged with the node risk values ​​to obtain the overall node risk value.

[0053] Optionally, in one example of the above aspects, the improved GAPSO algorithm is used to set the fitness function and the iteration termination condition. After iterative calculation, the topology structure of each level node after correction is obtained, including the following steps:

[0054] The improved GAPSO algorithm is used to encode particles, where each particle represents a topological structure with the dimension being the number of node connections.

[0055] Set the fitness function:

[0056] Where Fitness is the fitness value corresponding to the topology; Z_Risk is the mean of the comprehensive risk values ​​of each node under the corresponding topology; and Connectivity is the reciprocal of the mean number of connected components of each node under the corresponding topology.

[0057] Configure dynamic neighbor rules: core layer nodes are forced to retain 3 redundant paths;

[0058] The degree of aggregation layer nodes is limited to 8; if the degree is exceeded, high-risk links will be randomly deleted.

[0059] Set termination conditions: 100 iterations or the fitness value changes by less than 2% in 10 consecutive iterations;

[0060] The particle with the highest fitness value is selected as the optimal topology, which is used as the corrected topology.

[0061] Optionally, in one example of the above aspects, correcting the original digital topology based on the node-corrected topology includes the following steps:

[0062] A1: The comprehensive risk value of each node in the topology after statistical correction, and the comprehensive risk value of each node in the original digital topology.

[0063] A2: Compare the comprehensive risk values ​​of each corresponding node, calculate the difference between the comprehensive risk values ​​of the nodes before and after the correction, and if it is greater than the preset threshold, modify the topology connection of the corresponding node according to the topology structure of the node after the correction.

[0064] A3: For nodes that are missing connections after the topology is modified, make up the missing connections according to the topology of the corrected nodes, and return to step A1 until the difference between the comprehensive risk values ​​of each node before and after the correction is less than the preset threshold.

[0065] According to another aspect of this disclosure, an RPA-based automated correction system for marketing meter box archive topology diagrams is provided. This system uses the RPA-based automated correction method for marketing meter box archive topology diagrams as described above to achieve automated correction of the marketing meter box archive topology diagram.

[0066] Compared with the prior art, the beneficial effects of the present invention are:

[0067] This invention generates a correlation matrix by calculating the DTW distance between nodes, revealing hidden dynamic relationships in the topology graph and facilitating the identification of spatiotemporal correlations between nodes. For nodes with different spatiotemporal correlations, fault classification models for different node types are trained. Input features include traffic characteristics and hierarchy coefficients, and the model outputs the probability of node fault types, improving the accuracy of fault diagnosis.

[0068] This invention corrects the original digital topology based on the corrected topology of nodes. Through multi-dimensional risk quantification, hybrid algorithm optimization, and hierarchical correction strategies, it achieves precise, intelligent, and adaptive correction of the marketing metering box archive topology map. Its core advantage lies in the deep integration of node comprehensive risk, topological importance, and algorithm efficiency. This not only solves the problems of one-sided risk assessment and low correction efficiency in traditional methods, but also drives business optimization through data closed-loop, providing a replicable technical paradigm for the digital transformation of power marketing. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0071] Figure 2 This is a schematic diagram of the method for correcting the original digital topology structure according to the present invention. Detailed Implementation

[0072] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figures 1-2 The first aspect of this invention provides an automated correction method for the marketing metering box archive topology map based on RPA, comprising the following steps:

[0074] The RPA robot automatically logs into the marketing management system, GIS system, and design drawing library, extracts the topology map and related data of the metering box archives, extracts the text information in the topology map through OCR text recognition, and identifies the node positions, connection lines and equipment icon types through computer vision to construct a digital topology structure.

[0075] Acquire historical traffic data for each node to identify metering box nodes, user nodes, and transformer nodes, and set the hierarchy of each node based on historical traffic data and node historical data;

[0076] Calculate the DTW dynamic time warping distance between nodes, identify spatiotemporal correlations, and extract features from node data for nodes with different spatiotemporal correlations and levels.

[0077] Based on the extracted feature data, a fault classification model is trained for nodes with different spatiotemporal correlations, and the probability of node fault type is output.

[0078] Based on the probability analysis results of fault types of each node, node hierarchical data, and node degree centrality, and by introducing hierarchical coefficients, the comprehensive risk of nodes is calculated. Through the improved GAPSO algorithm, which integrates the crossover and mutation mechanism of genetic algorithm (GA) and the global search capability of particle swarm optimization (PSO), fitness functions and iteration termination conditions are set, and iterative calculations are performed to obtain the corrected topology structure of each level of nodes. The original digital topology structure is then corrected based on the corrected topology structure of the nodes.

[0079] Specifically, in this embodiment, the RPA robot automatically logs into the marketing management system, GIS system, and design drawing library to extract the metering box archive topology map and related data. The RPA robot can work 24 / 7 without interruption, completing cross-system data extraction, topology map parsing, and correction tasks without human intervention, completely eliminating the time constraints of manual operation. Metering box archive verification, which traditionally takes several hours manually, can be completed by RPA in minutes, improving efficiency by tens of times. The robot can simultaneously log into multiple platforms such as the marketing management system, GIS system, and design drawing library, extracting data and constructing topology structures in parallel, avoiding the inefficient operation of manually switching between systems one by one.

[0080] By automating repetitive manual operations, businesses can reduce their reliance on dedicated personnel and reallocate human resources to high-value tasks. Manual operations are prone to data errors or topological deviations due to fatigue and negligence, while RPA executes according to preset rules, resulting in an error rate close to zero and avoiding the costs of rework or decision-making errors caused by inaccurate data.

[0081] RPA generates detailed logs for each step of its operation, including login time, extracted data, and correction results, ensuring full traceability. The robot can have a built-in compliance rule library to verify data compliance in real time during the correction process, ensuring that the records meet industry standards and regulatory requirements.

[0082] OCR text recognition accurately extracts text information (such as meter box numbers and equipment parameters) from the topology diagram, solving the error-prone problem of manual data entry. It constructs a digital topology structure through intelligent recognition of node locations, connecting line routes, and equipment icon types.

[0083] By acquiring historical traffic data of each node, metering box nodes, user nodes, and transformer nodes are identified, and the hierarchical level of each node is set according to the historical traffic data and node historical data; the DTW dynamic time warping distance between nodes is calculated to identify spatiotemporal correlations, and different spatiotemporal correlations and hierarchical nodes are targeted.

[0084] Traditional topology correction relies solely on static information such as node location and connectivity. This method, however, transforms nodes from "spatial symbols" into "spatiotemporal entities" by incorporating historical traffic data, revealing their operational patterns. For example, transformer node traffic typically exhibits periodic fluctuations, while user node traffic may be affected by lifestyle habits, showing a nighttime trough. These characteristics can help further differentiate node types.

[0085] By calculating the DTW distance between nodes, a correlation matrix is ​​generated, revealing the hidden dynamic relationships in the topology graph, which helps to identify the spatiotemporal correlation of nodes.

[0086] For nodes with different spatiotemporal correlations, a fault classification model for different node types is trained. The input features include traffic features, hierarchical coefficients, etc., and the model outputs the probability of node fault type, thereby improving the accuracy of fault diagnosis.

[0087] By integrating historical traffic data, DTW spatiotemporal correlation analysis, and hierarchical fault models, a leap from "static topology correction" to "dynamic fault prediction" has been achieved. Its core advantage lies in using data to drive node cognitive upgrades and using algorithms to reveal hidden correlations, providing an intelligent and forward-looking solution for the management of electricity marketing metering boxes.

[0088] Based on the probability analysis results of fault types of each node, the comprehensive risk of the node is calculated. Using the improved GAPSO algorithm, a fitness function and an iteration termination condition are set. After iterative calculation, the topology structure of each level of node is obtained after correction. The original digital topology structure is then corrected based on the topology structure of the corrected nodes.

[0089] Reflecting the health status of nodes, the output is a classification model trained on historical traffic data. High-level node failures have a wider impact and require higher weighting. Considering node degree centrality, failures of nodes with many connected edges (such as hub metering boxes) are prone to triggering chain reactions, and their risks need to be considered cumulatively. Using an improved GAPSO algorithm, a fitness function and iteration termination condition are set, and iterative calculations yield the corrected topology structure for each level of nodes. Based on the corrected topology structure, the original digital topology structure is corrected. Through multi-dimensional risk quantification, hybrid algorithm optimization, and hierarchical correction strategies, the accuracy, intelligence, and adaptability of the marketing metering box archive topology map correction are achieved. Its core advantage lies in the deep integration of node comprehensive risk, topological importance, and algorithm efficiency. This not only solves the problems of one-sided risk assessment and low correction efficiency in traditional methods but also drives business optimization through data closed-loop, providing a replicable technical paradigm for the digital transformation of power marketing.

[0090] In one optional embodiment, historical traffic data of each node is acquired to identify metering box nodes, user nodes, and transformer nodes, and metering box data, business system data, and external environment data of the nodes are acquired, including the following steps:

[0091] Historical flow data of metering box nodes, user nodes and transformer nodes are obtained, the corresponding node types of the flow data are labeled, and the data is input into the LSTM model to train the model to identify the node types.

[0092] Historical traffic data for each node is obtained and input into the trained LSTM model to identify the node type.

[0093] In one optional embodiment, each node level is set based on historical traffic data and node historical data, including the following steps:

[0094] Based on node type data, user nodes are divided into access layer nodes;

[0095] Based on historical traffic data and node historical data, nodes that meet the following conditions are selected: daily average traffic is greater than the 80th percentile of all nodes; traffic standard deviation is less than the 20th percentile of all nodes; and the number of directly connected nodes is greater than 10.

[0096] The selected nodes will be used as core layer nodes;

[0097] The remaining unclassified nodes are classified as aggregation layer nodes.

[0098] In one optional embodiment, calculating the DTW dynamic time warping distance between nodes and identifying spatiotemporal correlations includes the following steps:

[0099] Obtain the historical traffic data for each node, convert the historical traffic data of the node into equal-length time series X and Y, with lengths |X| and |Y| respectively, and ensure that the timestamps of all time series are aligned.

[0100] Calculate the Euclidean distance D(i,j) between any two nodes to form an m×n distance matrix, where m and n are the sequence lengths;

[0101] Create a cumulative distance matrix D of (m+1)×(n+1), D(0,0)=0, and set the first row and first column to infinity;

[0102] Form a node set from all nodes, and calculate the Euclidean distance between any two nodes using the recursive formula: D(i,j) = d(i,j) + min(D(i-1,j), D(i,j-1),D(i-1,j-1)), where d(i,j) is the flow difference between node i and node j within the time series range, D(i-1,j) is the Euclidean distance between node i-1 and node j, D(i,j-1) is the Euclidean distance between node i and node j-1, and D(i-1,j-1) is the Euclidean distance between node i-1 and node j-1.

[0103] Using DTW distance as edge weights, a spatial relationship matrix between nodes is established;

[0104] DTW distances at different time granularities (daily, weekly, and monthly) are calculated to obtain multi-scale DTW distance features.

[0105] Multi-scale DTW distance matrices are fused using weighted summation or attention mechanisms to form a spatiotemporally correlated DTW distance matrix;

[0106] A hierarchical tree is constructed using the spatiotemporal correlation DTW distance matrix, and clustering results are obtained through pruning.

[0107] We directly use the DTW distance as a similarity metric to divide the nodes into K clusters;

[0108] Treating nodes as graph vertices and the reciprocal of the DTW distance as edge weights, we use the Louvain or Label Propagation algorithm to cluster nodes.

[0109] Nodes clustered into a unified cluster are considered as spatiotemporally relevant nodes.

[0110] In one optional embodiment, feature extraction is performed on node data for nodes with different spatiotemporal correlations and levels, including the following steps:

[0111] Based on the historical traffic data of each node in the spatiotemporal correlation nodes, a time-shifted sliding window W is introduced to calculate the feature data of the data within the window W, including: mean, variance, slope and frequency domain features.

[0112] Based on the node hierarchy, the transmission delay time of the upstream nodes connected to the node and the load of the downstream nodes connected to the node are statistically analyzed.

[0113] The ratio of the transmission delay time of the upstream node to the average transmission delay time in the historical data of the entire meter box archive topology is calculated to form the propagation delay coefficient matrix of the upstream node.

[0114] The ratio of the load of the downstream node connected to the computing node to the average node load in the historical data of the entire metering box archive topology diagram constitutes the downstream node load influence coefficient matrix.

[0115] The obtained coefficient matrix and feature data are used as the feature data of the nodes.

[0116] In one optional embodiment, based on the extracted feature data, a fault classification model is trained for nodes with different spatiotemporal correlations, and the probability of node fault type is output, including the following steps:

[0117] The feature data of the nodes are fused according to different time granularities of daily, weekly and monthly. The weights are dynamically allocated using an attention mechanism to obtain the fused feature data and the feature matrix of the node: (Nn,Twindows,G), where Nn is the node number, Twindows is the feature sequence and G corresponds to the fused feature data.

[0118] Extract the topological structure of the connection relationships of nodes with different spatiotemporal correlations, number the nodes with different spatiotemporal correlations, and obtain the adjacency matrix of the connection relationships of each node based on the node numbering information of the adjacent nodes of the nodes with different spatiotemporal correlations.

[0119] Constructing a GCN-LSTM fault classification model:

[0120] Establish a GCN layer, with the input being the node feature matrix and the adjacency matrix;

[0121] By aggregating node feature matrix and adjacency matrix information through the GCN layer, the spatiotemporal feature matrix of the node is output.

[0122] Connect the output of the GCN layer to the input of the LSTM layer, and input the spatiotemporal feature matrix output by the GCN into the LSTM layer;

[0123] The LSTM layer outputs time-coded features. The features output by the LSTM layer are then input into a fully connected layer + Softmax layer to output the probability of the fault type.

[0124] In one optional embodiment, based on the probability analysis results of each node's failure type, node hierarchy data, and node degree centrality, and by introducing a hierarchy coefficient, the comprehensive risk of a node is calculated, including the following steps:

[0125] Obtain the probability analysis results of each node's failure type, node-level data, and node degree centrality to quantify node risk;

[0126] Calculate node risk value Where Ri is the overall failure probability, such as hardware failure, software vulnerability, configuration error, etc., which is obtained through historical data statistics or expert scoring, such as Ri=0.8 for core routers;

[0127] Li is the hierarchical coefficient, with the core layer hierarchical coefficient = 3, the aggregation layer hierarchical coefficient = 2, and the access layer hierarchical coefficient = 1.

[0128] Di′ represents the normalized degree centrality = number of node connections / maximum possible number of connections, and β is the weight that controls the influence of centrality. In this embodiment, β = 0.2.

[0129] Calculate the link risk value Sij from node i to connected node j:

[0130] Where Delayij is the delay time from node i to connected node j, Max_Delay is the maximum transmission delay time in the historical data of the entire meter box archive topology, Lossij is the energy loss of power transmission from node i to connected node j, Max_Loss is the maximum energy loss of power transmission between nodes in the historical data of the entire meter box archive topology, and wloss and wdelay are the corresponding weights, wloss=0.3 and wdelay=0.7 respectively;

[0131] The average of the link risk values ​​of the obtained nodes and their connected nodes is then weighted and averaged with the node risk values ​​to obtain the overall node risk value.

[0132] In one optional embodiment, an improved GAPSO algorithm is used, which integrates the crossover and mutation mechanism of genetic algorithm (GA) with the global search capability of particle swarm optimization (PSO). Fitness functions and iteration termination conditions are set, and iterative calculations are performed to obtain the corrected topology of nodes at each level. This includes the following steps:

[0133] The improved GAPSO algorithm is used to encode particles, each particle representing a topology, with the dimension being the number of node connections. For example, if the core layer has 4 nodes, and each node connects to 3 neighboring nodes, then the particle dimension is 4×3=12.

[0134] Set the fitness function:

[0135] Where Fitness is the fitness value corresponding to the topology; Z_Risk is the average comprehensive risk value of each node under the corresponding topology; Connectivity is the reciprocal of the average number of connected components of each node under the corresponding topology, which is 1 when fully connected;

[0136] Configure dynamic neighbor rules: core layer nodes are forced to retain 3 redundant paths;

[0137] The degree of aggregation layer nodes is limited to 8; if the degree is exceeded, high-risk links will be randomly deleted.

[0138] Set termination conditions: 100 iterations or the fitness value changes by less than 2% in 10 consecutive iterations;

[0139] The particle with the highest fitness value is selected as the optimal topology, which is used as the corrected topology.

[0140] In this embodiment, the correction step is performed:

[0141] Initialize the particle swarm by randomly generating 50 particles, each representing an initial topology.

[0142] The initial number of connections for the core layer nodes is set to 3, and for the aggregation layer it is set to 5.

[0143] Calculate fitness:

[0144] For each particle, calculate Z_Risk and Connectivity;

[0145] Set the fitness function:

[0146] Where Fitness is the fitness value corresponding to the topology; Z_Risk is the mean of the comprehensive risk values ​​of each node under the corresponding topology; and Connectivity is the reciprocal of the mean number of connected components of each node under the corresponding topology.

[0147] Configure dynamic neighbor rules: core layer nodes are forced to retain 3 redundant paths;

[0148] The degree of aggregation layer nodes is limited to 8; if the degree is exceeded, high-risk links will be randomly deleted.

[0149] Speed ​​updates:

[0150] Introducing inertia weights , where t is the current iteration number, T is the maximum iteration number, and T=100.

[0151] Example: In the 20th iteration, ω=0.8, maintaining a large search range.

[0152] Set termination conditions: 100 iterations or the fitness value changes by less than 2% in 10 consecutive iterations;

[0153] Output the optimal topology:

[0154] Select the particle with the highest fitness as the topology after correction, ensuring that there are 3 redundant paths between the core layer nodes and the degree of the convergence layer nodes is ≤8.

[0155] In one optional embodiment, the original digital topology is corrected based on the corrected topology of the nodes, including the following steps:

[0156] A1: The comprehensive risk value of each node in the topology after statistical correction, and the comprehensive risk value of each node in the original digital topology.

[0157] A2: Compare the comprehensive risk values ​​of each corresponding node, calculate the difference between the comprehensive risk values ​​of the nodes before and after the correction, and if it is greater than the preset threshold, modify the topology connection of the corresponding node according to the topology structure of the node after the correction.

[0158] A3: For nodes that are missing connections after the topology is modified, make up the missing connections according to the topology of the corrected nodes, and return to step A1 until the difference between the comprehensive risk values ​​of each node before and after the correction is less than the preset threshold.

[0159] In this embodiment, the comprehensive risk values ​​of each corresponding node are compared, and a preset threshold is calculated for the difference between the comprehensive risk values ​​of the nodes before and after the correction. The comprehensive risk values ​​of the nodes of various topological connections are calculated based on the historical data of each node in the entire metering box archive topology diagram, and the average value is calculated. Half of the average value is used as the preset threshold.

[0160] In another embodiment of the present invention, an RPA-based automated correction system for the marketing meter box archive topology map is used to achieve automated correction of the marketing meter box archive topology map using the RPA-based automated correction method described above.

[0161] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An RPA-based automated correction method for marketing meter box archive topology diagrams, characterized in that, Includes the following steps: The RPA robot automatically logs into the marketing management system, GIS system, and design drawing library, extracts the topology map and related data of the metering box archives, extracts the text information in the topology map through OCR text recognition, and identifies the node positions, connection lines and equipment icon types through computer vision to construct a digital topology structure. Acquire historical traffic data for each node to identify metering box nodes, user nodes, and transformer nodes, and set the hierarchy of each node based on historical traffic data and node historical data; Calculate the DTW dynamic time warping distance between nodes, identify spatiotemporal correlations, and extract features from node data for nodes with different spatiotemporal correlations and levels. Based on the extracted feature data, a fault classification model is trained for nodes with different spatiotemporal correlations, and the probability of node fault type is output. Based on the probability analysis results of fault types of each node, node hierarchical data, and node degree centrality, and by introducing hierarchical coefficients, the comprehensive risk of nodes is calculated. Through the improved GAPSO algorithm, fitness functions and iteration termination conditions are set, and iterative calculations are performed to obtain the corrected topology structure of each level of nodes. The original digital topology structure is then corrected based on the corrected topology structure of the nodes.

2. The RPA-based automated correction method for marketing metering box archive topology diagrams according to claim 1, characterized in that, To acquire historical traffic data for each node, identify metering box nodes, user nodes, and transformer nodes, and obtain metering box data, business system data, and external environment data for each node, including the following steps: Historical flow data of metering box nodes, user nodes and transformer nodes are obtained, the corresponding node types of the flow data are labeled, and the data is input into the LSTM model to train the model to identify the node types. Historical traffic data for each node is obtained and input into the trained LSTM model to identify the node type.

3. The RPA-based automated correction method for the marketing metering box archive topology diagram according to claim 1, characterized in that, Configure each node level based on historical traffic data and node historical data, including the following steps: Based on node type data, user nodes are divided into access layer nodes; Based on historical traffic data and node historical data, nodes that meet the following conditions are selected: daily average traffic is greater than the 80th percentile of all nodes; traffic standard deviation is less than the 20th percentile of all nodes; and the number of directly connected nodes is greater than 10. The selected nodes will be used as core layer nodes; The remaining unclassified nodes are classified as aggregation layer nodes.

4. The method for automated correction of marketing meter box archive topology based on RPA according to claim 1, characterized in that, Calculate the DTW (Dynamic Time Warped Distance) between nodes and identify spatiotemporal correlations, including the following steps: Obtain the historical traffic data for each node, convert the historical traffic data of the node into equal-length time series X and Y, with lengths |X| and |Y| respectively, and ensure that the timestamps of all time series are aligned. Calculate the Euclidean distance D(i,j) between any two nodes to form an m×n distance matrix, where m and n are the sequence lengths; Create a cumulative distance matrix D of (m+1)×(n+1), D(0,0)=0, and set the first row and first column to infinity; Form a node set from all nodes, and calculate the Euclidean distance between any two nodes using the recursive formula: D(i,j) = d(i,j) + min(D(i-1,j), D(i,j-1),D(i-1,j-1)), where d(i,j) is the flow difference between node i and node j within the time series range, D(i-1,j) is the Euclidean distance between node i-1 and node j, D(i,j-1) is the Euclidean distance between node i and node j-1, and D(i-1,j-1) is the Euclidean distance between node i-1 and node j-1. Using DTW distance as edge weights, a spatial relationship matrix between nodes is established; DTW distances at different time granularities (daily, weekly, and monthly) are calculated to obtain multi-scale DTW distance features. Multi-scale DTW distance matrices are fused using weighted summation or attention mechanisms to form a spatiotemporally correlated DTW distance matrix; A hierarchical tree is constructed using the spatiotemporal correlation DTW distance matrix, and clustering results are obtained through pruning. We directly use the DTW distance as a similarity metric to divide the nodes into K clusters; Treating nodes as graph vertices and using the reciprocal of the DTW distance as edge weights, the Louvain algorithm is used for node clustering. Nodes clustered into a unified cluster are considered as spatiotemporally relevant nodes.

5. The RPA-based automated correction method for the marketing metering box archive topology diagram according to claim 1, characterized in that, For nodes with different spatiotemporal correlations and levels, feature extraction is performed on the node data, including the following steps: Based on the historical traffic data of each node in the spatiotemporal correlation nodes, a time-shifted sliding window W is introduced to calculate the feature data of the data within the window W, including: mean, variance, slope and frequency domain features. Based on the node hierarchy, the transmission delay time of the upstream nodes connected to the node and the load of the downstream nodes connected to the node are statistically analyzed. The ratio of the transmission delay time of the upstream node to the average transmission delay time in the historical data of the entire meter box archive topology is calculated to form the propagation delay coefficient matrix of the upstream node. The ratio of the load of the downstream node connected to the computing node to the average node load in the historical data of the entire metering box archive topology diagram constitutes the downstream node load influence coefficient matrix. The obtained coefficient matrix and feature data are used as the feature data of the nodes.

6. The RPA-based automated correction method for marketing metering box archive topology diagrams according to claim 1, characterized in that, Based on the extracted feature data, a fault classification model is trained for nodes with different spatiotemporal correlations, and the probability of node fault type is output, including the following steps: The feature data of the nodes are fused according to different time granularities of daily, weekly and monthly. The weights are dynamically allocated using an attention mechanism to obtain the fused feature data and the feature matrix of the node: (Nn,Twindows,G), where Nn is the node number, Twindows is the feature sequence and G corresponds to the fused feature data. Extract the topological structure of the connection relationships of nodes with different spatiotemporal correlations, number the nodes with different spatiotemporal correlations, and obtain the adjacency matrix of the connection relationships of each node based on the node numbering information of the adjacent nodes of the nodes with different spatiotemporal correlations. Constructing a GCN-LSTM fault classification model: Establish a GCN layer, with the input being the node feature matrix and the adjacency matrix; By aggregating node feature matrix and adjacency matrix information through the GCN layer, the spatiotemporal feature matrix of the node is output. Connect the output of the GCN layer to the input of the LSTM layer, and input the spatiotemporal feature matrix output by the GCN into the LSTM layer; The LSTM layer outputs time-coded features, which are then input into a fully connected layer + Softmax layer to output the probability of the fault type.

7. The method for automated correction of marketing metering box file topology diagram based on RPA according to claim 1, characterized in that, Based on the probability analysis results of each node's failure type, node hierarchy data, and node degree centrality, and by introducing hierarchy coefficients, the comprehensive risk of a node is calculated, including the following steps: Obtain the probability analysis results of each node's failure type, node-level data, and node degree centrality to quantify node risk; Calculate node risk value Where, Ri is the overall failure probability; Li is the hierarchical coefficient, with the core layer hierarchical coefficient = 3, the aggregation layer hierarchical coefficient = 2, and the access layer hierarchical coefficient = 1. Di′ represents the normalized degree centrality = number of node connections / maximum possible number of connections, and β is the weight that controls the influence of centrality. Calculate the link risk value Sij from node i to connected node j: Where Delayij is the delay time from node i to connected node j, Max_Delay is the maximum transmission delay time in the historical data of the entire meter box archive topology, Lossij is the energy loss of power transmission from node i to connected node j, Max_Loss is the maximum energy loss of power transmission between nodes in the historical data of the entire meter box archive topology, and wloss and wdelay are the corresponding weights, wloss=0.3 and wdelay=0.7 respectively; The average of the link risk values ​​of the obtained nodes and their connected nodes is then weighted and averaged with the node risk values ​​to obtain the overall node risk value.

8. The method for automated correction of marketing meter box archive topology based on RPA according to claim 1, characterized in that, By using the improved GAPSO algorithm, setting the fitness function and iteration termination condition, and performing iterative calculations, the corrected topology structure of each level of nodes is obtained, including the following steps: The improved GAPSO algorithm is used to encode particles, where each particle represents a topological structure with the dimension being the number of node connections. Set the fitness function: Where Fitness is the fitness value corresponding to the topology; Z_Risk is the mean of the comprehensive risk values ​​of each node under the corresponding topology; and Connectivity is the reciprocal of the mean number of connected components of each node under the corresponding topology. Configure dynamic neighbor rules: core layer nodes are forced to retain 3 redundant paths; The degree of aggregation layer nodes is limited to 8; if the degree is exceeded, high-risk links will be randomly deleted. Set termination conditions: 100 iterations or the fitness value changes by less than 2% in 10 consecutive iterations; The particle with the highest fitness value is selected as the optimal topology, which is used as the corrected topology.

9. The RPA-based automated correction method for the marketing metering box archive topology diagram according to claim 1, characterized in that, The original digital topology is corrected based on the corrected topology of the nodes, including the following steps: A1: The comprehensive risk value of each node in the topology after statistical correction, and the comprehensive risk value of each node in the original digital topology. A2: Compare the comprehensive risk values ​​of each corresponding node, calculate the difference between the comprehensive risk values ​​of the nodes before and after the correction, and if it is greater than the preset threshold, modify the topology connection of the corresponding node according to the topology structure of the node after the correction. A3: For nodes that are missing connections after the topology is modified, make up the missing connections according to the topology of the corrected nodes, and return to step A1 until the difference between the comprehensive risk values ​​of each node before and after the correction is less than the preset threshold.

10. An RPA-based automated correction system for marketing metering box file topology diagrams, characterized in that: The system employs the RPA-based automated correction method for the marketing meter box archive topology map as described in any one of claims 1-9 to achieve automated correction of the marketing meter box archive topology map.

Images