A Method and System for Pre-screening Causal Correlation of Candidate Sources of Voltage Exceedance

By constructing a causal correlation pre-screening method, which comprehensively considers the static structure and dynamic operating status of the distribution network, the source of responsibility for voltage exceeding the limit is identified. This solves the problems of misjudgment of causal relationships and computational redundancy in existing technologies, and achieves efficient and accurate source of responsibility screening and fault analysis.

CN123073818APending Publication Date: 2026-11-10STATE GRID JIANGSU ELECTRIC POWER CO LTD CHANGZHOU BRANCH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611552002.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-09
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for tracing voltage over-limits neglect dynamic power flow and physical causal relationships, leading to misjudgments of irrelevant nodes and redundant computing power, making it difficult to accurately screen the responsible source under complex operating conditions.

Method used

By constructing a causal correlation degree pre-screening method, the static structure and dynamic operation status of the distribution network are comprehensively considered to identify the causal validity of the transmission path, and a screening threshold is dynamically generated based on the causal correlation degree index to eliminate irrelevant nodes.

Benefits of technology

This significantly reduces the computational complexity of the subsequent detailed source tracing stage, improves the real-time response capability and accuracy of distribution network fault analysis, and ensures the physical representativeness of the screening results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123073818A_ABST
    Figure CN123073818A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for pre-screening causal correlation of candidate sources of voltage exceedance. The method includes: acquiring distribution network operation and topology data, identifying voltage exceedance nodes and an initial set of candidate sources; determining the transmission path between the exceedance node and the candidate sources based on the topology; deriving the expected power flow direction of the transmission path based on the operating status and comparing it with the actual power flow characteristics to determine the causal validity of the transmission path; for valid transmission paths, constructing a causal correlation index based on the physical topology and actual power flow characteristics; dynamically generating a screening threshold based on the distribution characteristics of the index, and thereby screening and outputting a subset of target candidate sources of responsibility. This invention integrates static structural connectivity and dynamic power causality, eliminating irrelevant nodes with power flow obstructions, and reducing the complexity of subsequent fine-grained source tracing calculations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of power distribution network operation monitoring and source tracing analysis technology. The invention relates to a method and system for pre-screening the causal correlation of candidate responsibility sources for voltage over-limit. Background Technology

[0002] With the large-scale integration of distributed power sources and diverse loads, the power flow distribution of distribution networks exhibits high uncertainty, and node voltage exceedances are becoming increasingly frequent. In the voltage exceedance source analysis phase, quickly and accurately locating the responsible source that actually contributes to the exceedance event is crucial for implementing subsequent voltage and reactive power control and network reconfiguration. A pre-screening mechanism can effectively eliminate a massive number of irrelevant nodes, significantly compressing the computational space, and is of significant engineering importance for ensuring the real-time response capability of the source analysis algorithm.

[0003] Currently, for the pre-screening of candidate sources for voltage exceedance tracing, the industry typically relies on electrical distance or topological connectivity for rough classification. Conventional methods often center on the exceedance node, directly including all similar devices within a certain topological hop count or electrical distance range into the candidate set. However, this type of screening mechanism based on static network structure can only determine the existence of physical paths, ignoring the dynamic blocking effect of real-time power flow direction evolution on the voltage conduction process; at the same time, when assessing the degree of multi-source impact, existing methods mostly use a single distance index for linear quantification, making it difficult to accurately distinguish the strength differences in power transmission in the network under different operating conditions.

[0004] In summary, when faced with complex operating conditions, existing methods are prone to misjudging a large number of irrelevant nodes as candidate sources of responsibility by ignoring dynamic power flow and physical causal relationships, resulting in serious computational redundancy and misjudgment risks in subsequent analysis stages. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for pre-screening the causal correlation of candidate sources of voltage over-limit, which can comprehensively consider the static structure and dynamic operating status of the distribution network, achieve efficient pre-screening of candidate sources of responsibility, and improve the accuracy and robustness of the pre-processing stage of voltage over-limit tracing in the distribution network.

[0006] The technical solution to achieve the purpose of this invention is as follows: A method for pre-screening causal correlation of candidate sources of voltage exceeding limits includes the following steps: Acquire the operating status data and topology data of the distribution network, identify voltage over-limit nodes, and construct an initial candidate source set; Based on the topology data, determine the transmission path between the over-limit node and the initial candidate source set; The expected power flow direction of the transmission path is obtained based on the operating status data, and compared with the actual power flow characteristics to determine the causal validity of the transmission path. For transmission paths with valid causal validity, a causal correlation index is constructed based on physical topology characteristics and actual power flow characteristics. Based on the distribution characteristics of the causal correlation index, a screening threshold is dynamically generated, and a subset of target candidate responsibility sources is selected from the initial candidate source set according to the generated screening threshold.

[0007] In a preferred embodiment, determining the transmission path between the over-limit node and the initial candidate source set includes: Construct node connection relationships and branch impedance magnitudes based on the topology data; Traverse all simple paths between the out-of-bounds node and the nodes in the initial candidate source set; During the traversal, if the search depth exceeds the preset maximum topology jump limit, the search of the current branch is stopped; The path that meets the hop count limit is taken as the transmission path, and the impedance modulus of each branch on the path is accumulated to obtain the cumulative impedance distance.

[0008] In a preferred technical solution, obtaining the desired power flow direction of the transmission path and comparing it with the actual power flow characteristics includes: The physical logic for responsibility transfer is determined based on the type of over-limit, and the physical logic for responsibility transfer is a causal driving relationship between node injection or absorption power and voltage fluctuations determined based on circuit theory. Based on the responsibility transfer physical logic determined by the out-of-limit type, assign the desired power flow direction code to each branch on the transmission path; The active power sign and amplitude of each branch power flow are extracted from the operating status data as actual power flow characteristics; By comparing the active power symbol of each branch power flow with the expected power flow direction code, the direction consistency determination result at the branch level is obtained.

[0009] In a preferred embodiment, assigning a desired power flow direction code to each branch on the transmission path includes: Determine the lowest common ancestor node of the out-of-bounds node and the initial candidate source nodes in the initial candidate source set in the radial topology; If the out-of-limit node is the lowest common ancestor node, the expected power flow direction is determined to be from the out-of-limit node to the initial candidate source node, and the corresponding path segment is assigned a first confidence weight. If the initial candidate source node is the lowest common ancestor node, the expected power flow direction is determined to be from the initial candidate source node to the over-limit node, and the corresponding path segment is assigned a second confidence weight. Conversely, the transmission path is divided into a first path segment from the initial candidate source to the lowest common ancestor, and a second path segment from the lowest common ancestor to the over-limit node, and the expected power flow direction and corresponding confidence weight of each path segment are determined respectively.

[0010] In a preferred technical solution, determining the causal validity of the transmission path includes: The ratio of the amplitude of the branch power flow to the active power injection intensity of the initial candidate sources in the initial candidate source set is used as the power transmission intensity ratio. If the direction consistency determination result of the branch is consistent, then the smaller value between the power transmission strength ratio and the predetermined value is taken as the consistency score of the corresponding branch. If the absolute value of the branch power flow is less than the preset light load threshold, then the preset light load base score is assigned. If the direction consistency judgment results of the branch are opposite, then zero points are assigned; The causal validity determination result of the transmission path is determined based on the consistency score of each branch on the transmission path.

[0011] In a preferred technical solution, determining the causal validity of the transmission path further includes aggregating the consistency scores of each branch to obtain power flow direction consistency features. The obtained power flow direction consistency features include: Calculate the proportion of the impedance magnitude of each branch to the total impedance of the transmission path to obtain the initial weight of each branch; The initial weights are corrected using the confidence weights corresponding to each path segment to obtain the comprehensive weights; The consistency scores of each branch on the transmission path are aggregated using a weighted geometric mean based on the comprehensive weights to obtain the power flow direction consistency characteristics.

[0012] In a preferred technical solution, constructing a causal correlation index based on physical topology characteristics and actual power flow characteristics includes: Extract the trend consistency component based on the trend direction consistency characteristics; The path complexity penalty component and electrical proximity feature component are extracted based on the physical topology attributes of the transmission path. The power intensity weighted component is extracted based on the power injection intensity of the nodes in the initial candidate source set; By integrating the power flow consistency component, the path complexity penalty component, the electrical proximity feature component, and the power intensity weighted component, a causal correlation index is generated.

[0013] In the preferred technical solution, dynamically generating the screening threshold includes: The mean and standard deviation of all non-zero causal correlation indices are calculated as distribution characteristics; The average severity of the out-of-limit node is assessed based on the node voltage deviation values ​​in the operational status data. An adjustment coefficient is determined based on the average severity of exceeding the limit, and a screening threshold is calculated based on a linear combination of the mean, the standard deviation, and the adjustment coefficient.

[0014] In the preferred technical solution, the subset of candidate sources of responsibility for screening and output includes: Nodes whose causal validity determination results are valid are identified from the initial candidate source set, and a causal valid candidate source benchmark set is constructed; Based on the screening threshold, nodes with a causal correlation index not lower than the screening threshold are selected from the initial candidate source set to obtain a preliminary candidate source subset; Calculate the power coverage of the candidate sources in the preliminary candidate source subset relative to the causal effective candidate source reference set; If the power coverage rate is lower than the preset coverage threshold, then candidate sources are added from the causal effective candidate source benchmark set in descending order of causal correlation index until the coverage threshold is met, thus obtaining a subset of target candidate responsible sources.

[0015] In the preferred technical solution, after filtering and outputting a subset of target candidate responsibility sources, the following steps are also included: Calculate the ratio of the number of nodes in the target candidate source subset to the total number of nodes in the initial candidate source set, and the ratio of the squares of these two terms, respectively, and use them as the node compression ratio and the computation reduction ratio. The node compression ratio, the computational cost reduction ratio, and the target candidate responsibility source subset are combined and output as the pre-screening performance evaluation results.

[0016] This invention also discloses a pre-screening system for the causal correlation of candidate sources of voltage exceedance, used to implement the aforementioned pre-screening method for the causal correlation of candidate sources of voltage exceedance, comprising: The data acquisition and identification module acquires the operating status data and topology data of the distribution network, identifies voltage over-limit nodes, and constructs an initial candidate source set. The transmission path determination module determines the transmission path between the over-limit node and the initial candidate source set based on the topology data. The transmission path causal validity determination module obtains the expected power flow direction of the transmission path based on the operating status data, and compares it with the actual power flow characteristics to determine the causal validity of the transmission path. The causal correlation index construction module constructs a causal correlation index based on physical topology characteristics and actual power flow characteristics for transmission paths with valid causal validity. The filtering module dynamically generates a filtering threshold based on the distribution characteristics of the causal correlation index, and filters and outputs a subset of target candidate responsibility sources from the initial candidate source set according to the generated filtering threshold.

[0017] The present invention also discloses a computer storage medium storing a computer program thereon, wherein when the computer executes the computer program, it implements the pre-screening method for causal correlation of voltage over-limit candidate responsibility sources as described in any of the above claims.

[0018] Compared with the prior art, the significant advantages of this invention are: To address the issues of misjudgment of irrelevant nodes and computational redundancy in existing voltage over-limit tracing of distribution networks due to neglect of dynamic power flow causal logic, this invention constructs a multi-dimensional causal correlation quantification model that comprehensively considers the static structure and dynamic operating status of the distribution network, thereby achieving efficient pre-screening of candidate responsibility sources.

[0019] To address the challenges of misjudging causal direction and identifying strong / weak correlations in complex topologies, this method utilizes least common ancestor analysis to identify the physical path of power transfer and assess confidence levels segment by segment. Simultaneously, it introduces the ratio of branch current amplitude to source injection to continuously correct the consistency score. This approach elevates traditional qualitative direction matching to a semi-quantitative transmission strength assessment, enabling the precise elimination of power flow obstruction nodes and remote nodes with extremely weak physical connections at the source.

[0020] To address the numerical instability and rigidity in setting screening thresholds during the correlation quantification process, this invention constructs an exponentially decaying correlation index with a fully multiplicative structure and establishes an adaptive threshold adjustment mechanism driven by the severity of exceeding the limit. This method ensures that the quantified index is strictly distributed within the normalized interval and allows the screening depth to automatically tighten or loosen according to the urgency of the fault, thus avoiding redundant calculations while ensuring the sensitivity of source tracing.

[0021] By implementing power coverage verification based on causal validity and multi-level elastic compensation logic, the logical conflicts caused by single statistical screening in traditional algorithms are eliminated. This invention significantly reduces the scale of nodes to be analyzed in the subsequent fine-grained tracing stage, reducing the computational complexity from the scale of a global search to the scale of a local core association set. While ensuring that the screening results have high physical representativeness, it greatly improves the real-time response capability of distribution network fault analysis. Attached Figure Description

[0022] Figure 1 This is a flowchart of the pre-screening method for the causal correlation of candidate sources of voltage exceeding limits in this embodiment; Figure 2This is a flowchart illustrating the process of determining the transmission path between the over-limit node and the initial candidate source set in this embodiment. Figure 3 A flowchart illustrating the derivation of the desired power flow direction of the transmission path in this embodiment and its comparison with actual power flow characteristics; Figure 4 A flowchart illustrating the process of assigning desired power flow direction codes to each branch of the transmission path in this embodiment. Detailed Implementation

[0023] The principle of this invention is as follows: The desired power flow direction of the transmission path is derived based on operational status data and compared with actual power flow characteristics to determine the causal validity of the transmission path; for transmission paths with valid causal validity, a causal correlation index is constructed by integrating physical topology characteristics and actual power flow characteristics; a screening threshold is dynamically generated based on the distribution characteristics of the causal correlation index, and a subset of target candidate responsibility sources is selected and output accordingly. This invention comprehensively considers both the static structure and dynamic operating status of the distribution network, achieving efficient pre-screening of candidate responsibility sources.

[0024] Example 1: like Figure 1 As shown, a method for pre-screening the causal correlation of candidate sources of voltage exceeding limits includes the following steps: Step 101: Obtain the operating status data and topology data of the distribution network, and identify voltage over-limit nodes and the initial candidate source set.

[0025] In a further embodiment, the method also includes identifying out-of-limit nodes and their corresponding out-of-limit types.

[0026] Specifically, the operating status data of the distribution network can be obtained in real time through the monitoring and data acquisition system. The operating status data includes the voltage amplitude measurement value of each node, the active power flow measurement value and reactive power flow measurement value of each branch, the active power output value of distributed power sources, and the active power consumption value of each electrical load.

[0027] Topology data can be extracted from the geographic information system or single-line map of the power grid management system. Specifically, it includes the set of node numbers of the distribution network, the branch connection relationship, the resistance parameters and reactance parameters of each branch.

[0028] After acquiring the above data (operational status data and topology data), the system scans and detects the voltage amplitude of each node.

[0029] In one specific implementation, if the voltage amplitude of a node is detected to be higher than a preset upper voltage threshold, for example, the voltage amplitude exceeds 1.07 per unit, then the node is identified as an upper limit node among the over-limit nodes; if the voltage amplitude of a node is detected to be lower than a preset lower voltage threshold, for example, the voltage amplitude is lower than 0.90 per unit, then the node is identified as a lower limit node among the over-limit nodes.

[0030] Based on the type of voltage limit exceedance, the system establishes a corresponding mapping relationship for responsible source types. Specifically, for upper limit exceedance events, all nodes connected to photovoltaic systems are extracted to form an initial candidate source set; for lower limit exceedance events, all nodes connected to loads are extracted to form an initial candidate source set. Simultaneously, the system records the injected power value or consumed power value of each candidate source, forming a candidate source power vector.

[0031] Step 102: Determine the transmission path between the over-limit node and the initial candidate source set based on the topology data.

[0032] In this embodiment, a node adjacency matrix describing network connectivity is constructed based on the branch connection relationships contained in the topology data. Using a depth-first search algorithm, all simple paths between each out-of-bounds node and each candidate source node in the initial candidate source set are traversed; these simple paths constitute the transmission paths. From a physical topology perspective, the transmission paths reflect the potential physical channels through which power is transferred from the source side to the observation point in the power grid. The system records the ordered sequence of intermediate nodes and branch sequences contained in the transmission paths to form a complete path information database, thus laying the physical structural foundation for subsequent causal relationship assessment.

[0033] Step 103: Derive the expected power flow direction of the transmission path based on the operating status data, and compare it with the actual power flow characteristics to determine the causal validity of the transmission path.

[0034] In other words, the actual power flow characteristics of each branch on the transmission path are extracted from the operational status data; the expected power flow direction of the transmission path is derived based on the over-limit type and topology data, and compared with the actual power flow characteristics to determine the causal validity and power flow direction consistency characteristics of the transmission path.

[0035] Specifically, based on the type of voltage overrun and the physical logic of power flow in the power system, the system derives the ideal power flow direction, i.e., the desired power flow direction, that each transmission path should possess when it has causal responsibility. The derivation principle of the desired power flow direction is as follows: if a candidate source is responsible for the overrun event, then power must be able to be transmitted from the candidate source to the overrun node, or the power change trend must be able to cause a directional evolution of the voltage at the overrun node.

[0036] Furthermore, the system extracts the actual active power flow measurements of each branch from the operational status data, determines the power flow direction based on the sign of the power flow, and obtains the actual power flow characteristics. By comparing the actual power flow direction of each branch with the corresponding expected power flow direction branch by branch, the system calculates the power flow direction consistency index of that path.

[0037] If the overall power flow direction consistency of the path meets the preset causal reachability condition, the path is marked as a causal valid path, that is, the transmission path is determined to have causal validity; otherwise, if there is significant power flow obstruction on the path, the path is determined to have no physical causal relationship.

[0038] Step 104: For transmission paths with valid causal validity, construct a causal correlation index by combining the physical topology characteristics and actual power flow characteristics of the transmission path.

[0039] Assuming the transmission path is deemed causally valid, the system further quantifies the impact of this path on voltage exceedance events, generating a causal correlation index. Specifically, physical topology features include the cumulative impedance distance between the exceedance node and the candidate source node, as well as the number of hops along the path's branches. Actual power flow features include the consistency score between the power flow of each branch along the transmission path and the desired direction, as well as the injected power intensity of the candidate source. The system integrates these components by establishing a multi-factor fusion quantification model.

[0040] In one specific embodiment, the causal correlation index is calculated as follows: ρ_ij=C_flow P_comp W_power D_prox; Wherein, ρ_ij is the causal correlation index between the over-limit node i and the candidate source node j, C_flow is the power flow direction consistency component, P_comp is the path complexity penalty component, W_power is the power intensity weighted component of the candidate source, and D_prox is the electrical proximity component characterizing the intensity of the electrical distance influence. All of the above components have been normalized so that the final causal correlation index falls within the value range of 0 to 1. This index allows for the quantification of the potential contribution of candidate sources to a specific voltage over-limit event from the perspective of physical causality.

[0041] Step 105: Dynamically generate a screening threshold based on the distribution characteristics of the causal correlation index, and then select and output a subset of target candidate responsibility sources from the initial candidate source set accordingly.

[0042] Specifically, the system performs statistical analysis on the causal correlation indices corresponding to all paths determined to be causally valid, obtaining the distribution characteristics of the causal correlation indices such as mean, standard deviation, and quantiles under the current power grid operation scenario. Simultaneously, the system calculates the average severity of current voltage exceedance events based on operational status data. Based on the severity of the exceedance, the system adaptively determines the adjustment coefficient, thereby dynamically generating a screening threshold. Correspondingly, the system retains candidate source nodes whose causal correlation indices are not lower than the screening threshold, thus forming a subset of target candidate responsible sources. This subset serves as input for the subsequent refined contribution calculation module, performing refined sensitivity analysis or contribution ranking only on the candidate sources within this subset. Through this two-stage pre-screening process, a large number of nodes that do not have a physical causal relationship with the exceedance node or have an extremely weak relationship are eliminated, significantly improving the real-time performance and accuracy of the source tracing analysis.

[0043] Example 2: Based on the above embodiment 1, this embodiment further details the specific steps of determining the transmission path between the over-limit node and the initial candidate source set.

[0044] In one possible implementation, such as Figure 2 As shown, it includes the following steps: Step 201: Construct node connection relationships and branch impedance magnitudes based on topology data.

[0045] Specifically, constructing node connection relationships refers to the system parsing and acquiring the distribution network topology data, extracting attribute parameters that characterize the interconnection logic of equipment, and generating corresponding adjacency lists or directed graph structures in computing memory. In this data structure, buses or low-voltage sides of distribution transformers in the system are considered nodes, and tie lines, feeder sections, and other line equipment are considered branches. Based on this, constructing branch impedance modulus refers to the system performing modulus conversion based on physical parameters such as wire diameter type and line length, or the acquired branch resistance and reactance values.

[0046] In practical engineering calculations, the complex impedance of a branch can be expressed as Z = R + jX, where R is resistance, X is reactance, and j is the imaginary unit. The system calculates the branch impedance modulus by determining the modulus of this complex number. Z_abs=sqrt(R 2 +X 2 ); Where Z_abs is the branch impedance magnitude, sqrt is the square root function, R is the branch resistance, and X is the branch reactance. This magnitude characterizes the degree of physical obstruction encountered by power during transmission in this branch.

[0047] Optionally, when constructing the adjacency list, the system can use sparse matrix compression storage technology to reduce the dimensionality of the large-scale power distribution network model, thereby reducing the memory overhead during the search process.

[0048] Step 202: Use a depth-first search algorithm to traverse all simple paths between the bounded node and the nodes in the initial candidate source set.

[0049] Specifically, depth-first search (DFS) is a conventional algorithm used for traversing or searching graph structures. The system starts the search from a specific node that has exceeded the limit, or from a specific initial candidate source node, progressing along the unvisited adjacent edges of the current node towards deeper nodes. All simple paths are paths where, during the search process, every node (including intermediate nodes) traversed from the starting node to the target node appears exactly once, without any topological loops or backtracking. For example, in radial distribution networks or weak ring networks with limited interconnecting switches, the system uses a stack structure to record the node push sequence of the current search path.

[0050] Furthermore, to avoid invalid searches caused by data silos or topological breaks, the system can generate a set of connected subgraphs of the distribution network through breadth-first traversal before performing a depth-first search. If the out-of-limit node and the target candidate source node belong to different disconnected subgraphs, the system directly interrupts the path search for that node pair to improve pre-screening efficiency.

[0051] Step 203: During the traversal, if the search depth exceeds the preset maximum topology jump limit, then stop the search of the current branch.

[0052] In detail, the search depth during traversal refers to the total number of branches contained in the currently pushed search path. The preset maximum topology hop count limit is a truncation parameter set by the system to control the computational scale and conform to the physical causal decay law. In power distribution systems, as electrical distance increases, the node voltage response sensitivity caused by power injection changes exhibits a significant nonlinear decay. When the topology hop count exceeds a certain threshold, the impact of power injection changes from distant candidate sources on the voltage of nodes exceeding the limit becomes negligible. Based on this, the system extracts the current search stack depth variable and compares it with the maximum topology hop count limit each time the search algorithm undergoes a state transition. For example, if the maximum topology hop count limit is set to 15, when the algorithm progresses along a branch to the 16th node (i.e., crossing 15 branches), the system determines that the condition has been triggered, immediately performs a pop operation and switches to the backtracking state, no longer continuing to explore deeper-level related nodes of that branch.

[0053] Optionally, the maximum topology hop count limit can be configured as a fixed global constant (such as a positive integer between 10 and 15) or as a dynamic parameter. For example, the system can obtain the longest electrical feeder length or network diameter of the current distribution network and dynamically set the maximum topology hop count limit to a rounded value of 1.2 to 1.5 times the network diameter.

[0054] Step 204: The path that meets the hop count limit is taken as the transmission path, and the impedance modulus of each branch on the path is accumulated to obtain the cumulative impedance distance, which is used to construct the causal correlation index in the future.

[0055] Specifically, when the search process reaches the target node and the number of path hops does not exceed the aforementioned limit, the system saves the currently recorded sequence containing ordered nodes and corresponding branches. This sequence constitutes the physical connection channel, i.e., the transmission path, used to characterize the potential power flow between nodes. Subsequently, for each extracted transmission path, the system extracts all its branches and reads the previously calculated corresponding branch impedance magnitudes.

[0056] The system arithmetically adds the magnitudes of these branch impedances: ; in, Z_path SUM represents the cumulative impedance distance corresponding to the transmission path, and SUM is the summation operator. Z_abs_k Let k be the impedance magnitude of the kth branch on the transmission path, where k ranges from 1 to the total number of hops in the branches of the path.

[0057] In scenarios with multiple parallel transmission channels or ring network structures, the system may extract multiple transmission paths for a specific pair of nodes. In this case, the system can determine the path with the smallest calculated cumulative impedance distance as the preferred transmission path (i.e., the main power supply or main propagation channel), and discard the remaining high-impedance alternative paths. Subsequently, the data of the determined preferred transmission path is loaded into subsequent modules for causal analysis of the desired power flow direction.

[0058] Example 3: Building upon Example 2 above, this example further details the specific steps for deriving the desired power flow direction of the transmission path and determining the confidence weights. Since distribution networks typically exhibit a radial topology, the voltage impact mechanism of candidate loads on over-limit nodes depends on their relative positions in the topology tree. Therefore, this example introduces least common ancestor analysis to improve the accuracy of causal determination.

[0059] In one possible implementation, such as Figure 3 As shown, the expected power flow direction of the transmission path is derived based on the over-limit type and topology data, and compared with the actual power flow characteristics, including the following steps: Step 301: Based on the responsibility transfer physical logic determined by the over-limit type, assign the expected power flow direction code to each branch on the transmission path; extract the active power symbol and amplitude of the power flow of each branch from the operating status data as the actual power flow characteristics; compare the active power symbol of the power flow of each branch with the expected power flow direction code to obtain the direction consistency judgment result at the branch level.

[0060] Specifically, the physical logic of responsibility transfer refers to the causal driving relationship between node-injected or absorbed power and voltage fluctuations, determined based on circuit theory. For events exceeding the upper limit, power should flow from the photovoltaic node to the node exceeding the limit to support the voltage rise. For events exceeding the lower limit, since load power absorption causes a voltage drop along the line, the direction of responsibility transfer manifests as power flowing from the power source side to the load node. The system constructs a branch sequence in memory from candidate source node j through intermediate node sequences to the node exceeding the limit i, denoted as... l_0,l_1 ,..., l_K For each branch, the system generates a corresponding direction identifier according to preset direction encoding rules.

[0061] Furthermore, the system extracts the active power data of each branch at the corresponding moment from the real-time collected operational status data. The active power sign represents the actual current flow direction, while the amplitude represents the intensity of energy exchange. By performing a sign bit logic comparison operation, the system determines whether the actual flow direction matches the derived expected direction. If they match, the branch is determined to have a preliminary causal relationship. Through this comparison mechanism, the system can identify whether there is a reverse current flow obstruction on the path, thereby initially filtering out paths that do not meet the physical causal transmission conditions.

[0062] For events exceeding the upper limit, the injected power from photovoltaic (PV) nodes raises the voltage along the line. If the condition that the lowest common ancestor node *a* and the initial candidate source node *j* are the same node is met, it indicates that PV node *j* is located on the path from the root node to the over-limit node *i*. The injected PV power flows directly downstream to the over-limit node *i*, causing a voltage rise. In this case, the desired power flow direction is determined to be from the candidate source node *j* to the over-limit node *i*, and the system assigns a first confidence weight to the corresponding path segment. If the condition that the lowest common ancestor node *a* and the over-limit node *i* are the same node is met, it indicates that the over-limit node *i* is an upstream node of the candidate source node *j*. The power injection from PV node *j* causes the voltage of the upstream node to rise due to reverse current injection. In this case, the desired power flow direction is determined to be from the over-limit node *i* to the candidate source node *j*, and the system assigns a second confidence weight to the corresponding path segment. If the lowest common ancestor node *a* is different from both node *i* and node *j*, it indicates that they are located on different feeder branches. The system divides the path into two segments, determines the expected power flow direction and corresponding confidence weight for each segment, and processes the same logic as for cross-branch cases that exceed the lower limit.

[0063] Among them, such as Figure 4 As shown, based on the responsibility transfer physical logic determined by the boundary type, each branch on the transmission path is assigned a desired power flow direction code, including: Determine the lowest common ancestor node of the out-of-bounds node and the candidate source nodes (in the initial candidate source set) in the radial topology; If the node that exceeds the limit is the lowest common ancestor node, the expected power flow direction is determined to be from the node that exceeds the limit to the candidate source node, and the corresponding path segment is assigned the first confidence weight. If the initial candidate source node is the lowest common ancestor node, the expected power flow direction is determined to be from the candidate source node to the over-limit node, and the corresponding path segment is assigned a second confidence weight. If the lowest common ancestor node is different from the two types of nodes mentioned above, the transmission path is divided into a first path segment from the candidate source to the lowest common ancestor and a second path segment from the lowest common ancestor to the over-limit node, and the expected power flow direction and corresponding confidence weight of each path segment are determined respectively.

[0064] In detail, the lowest common ancestor (LCA) node is the node furthest from the root node in a radial tree topology rooted at the substation node, serving as the common ancestor of both the boundary-crossing node i and the candidate source node j. The system performs backtracking searches from node i and node j towards the root node, obtaining two path node sets A_i and A_j respectively. The node with the largest depth in the intersection of these two sets is taken as the LCA node a. For boundary-crossing events, the relative topological positions between nodes determine the differentiated causal mechanism.

[0065] If the lowest common ancestor node *a* and the over-limit node *i* are the same node, it indicates that the over-limit node *i* is located on the unique path from the root node to the candidate source node *j*, meaning that the over-limit node *i* is an upstream node of the candidate source node *j*. In this case, the current absorbed by the candidate source node *j* must pass through the branch where the over-limit node *i* is located, resulting in a voltage drop along the path. Based on this, the system determines the desired power flow direction as flowing from the over-limit node *i* to the candidate source node *j*, i.e., from upstream to downstream.

[0066] If the lowest common ancestor node *a* and the candidate source node *j* are the same node, it indicates that the candidate source node *j* is located on the unique path from the root node to the over-limit node *i*, meaning that the candidate source node *j* is an upstream node of the over-limit node *i*. The physical mechanism manifests as a voltage drop at the candidate source node *j* propagating downstream. In this case, the desired power flow direction is determined to be from the candidate source node *j* to the over-limit node *i*.

[0067] If the lowest common ancestor node 'a' is different from both nodes 'i' and 'j', it indicates that they are located on different feeder branches. In this case, power flows from the root node to the lowest common ancestor node 'a' and then splits. The system divides the path into two segments: the first segment represents the influence of candidate sources on the voltage of the common ancestor node, and the second segment represents the process of voltage drop propagation across branches. Through this segmented derivation method, the system achieves accurate modeling of the responsibility transfer path under complex topologies.

[0068] The confidence weights are positively correlated with the strength of the corresponding physical causal transmission mechanism; among them, the first confidence weight, which represents direct current conduction, is greater than the second confidence weight, which represents voltage drop propagation; the confidence weights, which represent indirect propagation across branches, corresponding to the first and second path segments, are the smallest.

[0069] Specifically, confidence weights are used to quantify the degree of certainty of causal relationships under different topological relationships. The first confidence weight characterizes the strong causal relationship generated by current flowing directly through the over-limit node. In this embodiment, the first confidence weight κ_1 is preset to 1.0. The second confidence weight κ_2 characterizes the process of voltage drop propagating downstream along the line, and its influence intensity is adjusted by the load shunting effect along the way; therefore, κ_2 is preset to 0.8.

[0070] For cross-branch propagation scenarios, the causal transmission mechanism is the most indirect. The system assigns a weight of 1.0 to the first path segment and a lower weight, such as 0.5, to the second path segment involving voltage propagation. Through this differentiated weighting mechanism, the system can reflect the attenuation characteristics of the physical mechanism. The system uses the confidence weights of each branch to weight and correct subsequent consistency indices. w_final=w_initial κ_l; Where w_final is the corrected branch composite weight, w_initial is the initial weight calculated based on the proportion of branch impedance to total path impedance, and κ_l is the confidence weight corresponding to the path segment to which the branch belongs.

[0071] This topology-aware weighting method ensures that paths with clear causal mechanisms receive higher correlation scores, while paths with weak causal mechanisms or high uncertainty are reasonably suppressed, thus effectively solving the problem of misjudgment of responsibility sources caused by the inability to distinguish upstream and downstream relationships in traditional methods.

[0072] Example 4: Based on the inventive concept of Embodiment 3 above, this embodiment further details the specific steps for determining the causal validity of the transmission path and the consistency characteristics of the power flow direction. Since traditional causal determination relies only on qualitative comparison of direction and cannot distinguish the differences in causal transmission intensity under different power carrying levels, this embodiment introduces a power transmission intensity ratio to achieve a semi-quantitative evaluation between pure direction judgment and complete power flow tracing.

[0073] In one possible implementation, determining the causal validity and power flow direction consistency characteristics of the transmission path includes the following steps: Step 401: The ratio of the amplitude of the branch power flow to the active power injection intensity of the initial candidate source is used as the power transmission intensity ratio.

[0074] Specifically, the system extracts the absolute value of the active power flow of the target branch from the operational status data and extracts the absolute value of the active power injection of the corresponding candidate source from the candidate source power vector. To quantify the ratio of the actual active power carried by the branch to the injected power from the candidate source, the system calculates the quotient of the two: ; Where r_l is the power transmission intensity ratio of the l-th branch, P_actual_abs is the amplitude of the branch power flow, P_j_abs is the active power injection intensity of the initial candidate source, and ε_p is a preset small positive real number compensation term to prevent division-by-zero anomalies. The preset small positive real number compensation term can be 0.01 MW. The physical mechanism of this ratio is that if the value is close to or greater than 1, it indicates that the power carried by the branch is not less than the injection amount of the candidate source, and it has sufficient physical conditions for power transmission; if the value is much less than 1, it indicates that the power of the candidate source has been significantly attenuated due to the diversion or loss of network nodes before reaching the branch.

[0075] Step 402: If the direction consistency determination result of the branch is consistent, then the smaller value of the power transmission strength ratio and one is taken as the consistency score of the corresponding branch.

[0076] In detail, when the actual power flow direction is the same as the expected power flow direction, that is, when the direction consistency judgment result is consistent, the system does not assign a uniform full score, but truncates the score based on the transmission strength.

[0077] s_l=min(r_l,1.0); Where s_l is the consistency score of the branch, and min is the minimum value function. The physical basis of this truncation mechanism is that if the power carried by the branch is much smaller than the power injected by the candidate source, for example, if the candidate source injects 5 MW but the branch only has 0.5 MW of forward power flow, it means that a large amount of power has been diverted upstream, and only a small amount of power is transmitted through the branch, indicating a weak causal relationship. In this case, the score is taken as a smaller ratio. If the power carried by the branch is greater than the power injected by the candidate source, it means that the branch has gathered power from multiple sources and is fully capable of transmitting the power of the candidate source. In this case, the score is taken as the upper limit of 1.0.

[0078] Step 403: If the absolute value of the branch power flow is less than the preset light load threshold, then assign the preset light load base score.

[0079] In actual power grid operation, some branches may be in a critical state of power flow balance. The system determines whether the power flow amplitude of the branch is lower than a preset light load threshold. The preset light load threshold can be set to 0.01 MW. When this condition is met, it indicates that there is almost no effective power transmission on the branch. The system assigns a preset light load base score to the branch. The preset light load base score can be set to 0.1. The reason for assigning a lower base score is that light load branches have very weak supporting effect on causal transmission. If a higher neutral score is assigned, it will lead to a path containing multiple light load branches receiving an artificially high overall consistency assessment.

[0080] Step 404: If the direction consistency determination result of the branch is opposite, then assign zero points.

[0081] When the actual power flow direction is opposite to the desired power flow direction, a reverse power flow exists at the physical entity level. This reverse power flow constitutes an absolute physical barrier to power transfer, preventing the power from candidate sources from reaching the over-limit node along this path. Therefore, the system directly assigns a consistency score of 0 to this branch.

[0082] Step 405: Determine the causal validity of the transmission path based on the consistency scores of each branch on the transmission path.

[0083] The system collects the consistency scores of all branches along the transmission path. If the path contains a branch with a score of 0, or if the aggregate score of the entire path is lower than a preset validity threshold, the system outputs an invalid result; otherwise, it outputs a valid result. The preset validity threshold can be set to 0.3.

[0084] The process of aggregating the consistency scores of each branch to obtain the power flow direction consistency characteristics includes steps such as 406.

[0085] Step 406: Based on the topology data, calculate the ratio of the impedance magnitude of each branch to the cumulative impedance distance of the total impedance of the transmission path, and obtain the initial weight of each branch.

[0086] To reasonably aggregate the scores of each branch along the path, the system introduces the electrical impedance of the branch as a weighting benchmark. The physical principle is that the branch with higher impedance generates a more significant voltage drop during power transmission, and its influence on the voltage of over-limit nodes is dominant.

[0087] The initial weights of each branch are: w_l=Z_abs_l / Z_path; Where w_l is the initial weight of the l-th branch, Z_abs_l is the impedance magnitude of the l-th branch, and Z_path is the cumulative impedance distance of the transmission path.

[0088] Step 407: Correct the initial weights using the confidence weights corresponding to each path segment on the transmission path to obtain the comprehensive weights; wherein, the confidence weights are predetermined based on the topological positional relationship between the over-limit node and the candidate source node in the transmission path.

[0089] The system extracts the confidence weights derived from the topological positional relationship in Example 3, and uses these confidence weights to correct the initial weights of the purely electrical parameters at the physical mechanism level.

[0090] The overall weight is: w_final_l=w_l κ_l; Where w_final_l is the overall weight of the l-th branch, and κ_l is the confidence weight of the path segment to which this branch belongs. Through this correction, the weight of cross-branch path segments with weaker causal mechanisms in the overall path evaluation is proportionally reduced.

[0091] Step 408: Use the comprehensive weight to perform weighted geometric mean aggregation on the consistency scores of each branch on the transmission path to obtain the power flow direction consistency characteristics.

[0092] To accurately reflect the weakest physical link in the responsibility transmission path, the system uses a weighted geometric mean algorithm to aggregate the continuous consistency scores. Compared to the arithmetic mean, the geometric mean has a significant bottleneck amplification effect.

[0093] γ_path=exp(SUM(w_final_l ln(s_l)) / SUM(w_final_l)); Where γ_path represents the power flow direction consistency characteristic of the final output, exp is the natural exponential function, SUM is the summation operator over all branches on the transmission path, and ln is the natural logarithm function. This calculation mode ensures that when there is any bottleneck branch with extremely weak carrying capacity on the path, the overall consistency characteristic of the path will be significantly reduced.

[0094] In some alternative implementations, if the processor's nonlinear computational resources are limited, the system can use a bottleneck truncation operator for consistency scores instead of weighted geometric mean aggregation. Specifically, the system directly traverses all branches on the path and extracts the global minimum of the consistency score as the power flow direction consistency feature. This alternative reduces the overhead of floating-point operations while retaining the core function of identifying physical transmission bottlenecks.

[0095] In another aspect of this embodiment, determining the causal validity and power flow direction consistency characteristics of the transmission path can also be: Extract the active power injection intensity of the corresponding candidate source node from the initial candidate source set from the operating status data; The ratio of the amplitude of the power flow in each branch to the active power injection intensity in the actual power flow characteristics is taken as the power transmission intensity ratio. The following scoring is performed on each branch of the transmission path in sequence: if the absolute value of the power flow of the branch is less than the preset light load threshold, the preset light load base score is assigned as the consistency score of the corresponding branch; otherwise, if the directional consistency determination result of the branch is consistent, the smaller value of the power transmission intensity ratio and one (predetermined value) is taken as the consistency score of the corresponding branch; if the directional consistency determination result of the branch is opposite, zero score is assigned as the consistency score of the corresponding branch. The consistency scores of each branch on the transmission path are aggregated to obtain the power flow direction consistency characteristics of the transmission path. If the power flow direction consistency characteristics are lower than the preset validity threshold or there is a branch on the transmission path with a consistency score of zero, the causal validity of the transmission path is determined to be invalid; otherwise, it is determined to be valid.

[0096] In another aspect of this embodiment, steps 401 to 405 may also be: For each branch in the transmission path, the system performs consistency scoring according to the following priority order: Determine whether the absolute value of the active power flow of this branch is less than a preset light load threshold. The preset light load threshold can be set to 0.01 MW. If the light load condition is met, then regardless of the power flow direction, a preset light load base score of 0.1 is assigned as the consistency score for this branch.

[0097] If a branch does not meet the light load condition, then directional consistency is further assessed. If the directional consistency assessment result of the branch is consistent, then the power transmission intensity ratio is calculated, and the smaller value of the power transmission intensity ratio and one is taken as the consistency score of the corresponding branch.

[0098] If the direction consistency determination result of a branch is opposite, then the consistency score of that branch is assigned to zero.

[0099] Collect the consistency scores of all branches along the transmission path. If the path contains a branch with a consistency score of zero, the system directly outputs an invalid causal validity determination result without further aggregation calculation. If the consistency scores of all branches are greater than zero, the system performs the weighted geometric mean aggregation processing in steps 406 to 408 to obtain the power flow direction consistency feature. If the power flow direction consistency feature is lower than a preset validity threshold (which can be set to 0.3), an invalid determination result is output; otherwise, a valid determination result is output.

[0100] Example 5: Based on the technical solution of Embodiment 4 above, this embodiment further details the specific steps for constructing a causal correlation index by integrating physical topology characteristics and actual power flow characteristics. This embodiment introduces an exponentially decaying electrical proximity and constructs a causal correlation quantification model with a fully multiplicative structure.

[0101] In one possible implementation, a causal correlation index is constructed by combining the physical topology characteristics of the transmission path with the actual power flow characteristics, including the following steps: Step 501: Extract the trend direction consistency component based on the trend direction consistency feature.

[0102] Specifically, the system directly assigns the calculated power flow direction consistency characteristics to the corresponding evaluation variables as quantitative parameters characterizing the degree to which the power flow direction supports the transfer of responsibility.

[0103] The trend direction consistency component is: C_flow=γ_path; Where C_flow is the power flow direction consistency component, and γ_path is the power flow direction consistency feature obtained after weighted geometric mean aggregation.

[0104] This component preserves information about the bottleneck effect of the path during power transmission, and its value range is strictly limited to between 0 and 1.

[0105] Step 502: Extract the path complexity penalty component and electrical proximity feature component based on the physical topology attributes (physical topology features) of the transmission path.

[0106] In detail, the physical topology attributes include the number of hops in the transmission path and the cumulative impedance distance. The system first calculates the path complexity penalty component based on the number of hops. Considering that power shunting and loss occur at each intermediate node during power transmission in the distribution network, and that this shunting effect exhibits nonlinear characteristics, the system uses an exponential decay model based on the number of hops to calculate this penalty component.

[0107] The path complexity penalty component is: P_comp=exp(-λ (n_hop-1)); Where P_comp is the path complexity penalty component, exp is the natural exponential function, λ is the preset attenuation coefficient, and n_hop is the topology hop count of the transmission path.

[0108] The preset attenuation coefficient can be set to 0.2. Through this calculation, the more hops a path in the topology has, the smaller the penalty component value it receives, indicating that its causal contribution to voltage over-limit events is reduced due to multiple current shunting.

[0109] Step 503: Calculate the normalized ratio of the cumulative impedance distance corresponding to the sum of the impedance magnitudes of each branch on the transmission path to the cumulative impedance distance of all candidate paths, and obtain the normalized electrical distance.

[0110] For the extraction of electrical proximity feature components, the system reads the cumulative impedance distance of each preferred transmission path between the current out-of-limit node and all nodes in the initial candidate source set from the cache record, and extracts the maximum impedance value as a global reference benchmark. Subsequently, the system calculates the ratio of the impedance value of the currently processed transmission path to this global reference benchmark.

[0111] The normalized electrical distance is: D_elec = Z_path / Z_max; Where Delec is the normalized electrical distance, Z_path is the cumulative impedance distance corresponding to the current transmission path, and Z_max is the maximum cumulative impedance distance among all candidate paths for the same out-of-limit node.

[0112] Step 504: The normalized electrical distance is mapped using a preset exponential decay function to obtain an electrical proximity feature component that decreases exponentially as the normalized electrical distance increases.

[0113] In a power distribution network, the physical sensitivity of node voltage to injected power at a distance decreases exponentially with increasing electrical distance. Based on this physical prior law, the system converts the aforementioned linear ratio into an exponential proximity evaluation index.

[0114] D_prox=exp(-α_d D_elec); Where D_prox is the electrical proximity feature component, exp is the natural exponential function, α_d is the preset distance attenuation sensitivity coefficient, and D_elec is the normalized electrical distance.

[0115] The preset distance attenuation sensitivity coefficient controls the strength of the influence of electrical distance on correlation, and its value can be set between 1.5 and 3.0. The preset distance attenuation sensitivity coefficient is usually set to 2.0.

[0116] Step 505: Extract the power intensity weighted component based on the power injection intensity of the nodes in the initial candidate source set.

[0117] Alternatively, the power injection intensity of the corresponding candidate source node in the initial candidate source set can be extracted from the running status data, and the power intensity weighted component can be calculated accordingly.

[0118] The power magnitude of a candidate source directly determines its potential influence on nodes exceeding the limit. The system extracts the absolute value of the active power injection of the initial candidate source node and divides it by the maximum value of the absolute power of all candidate sources in the current set. The calculation logic is as follows: W_power = P_j_abs / P_max; Where W_power is the power intensity weighted component, P_j_abs is the absolute value of the active power injection intensity of the candidate source node, and P_max is the maximum value of the absolute value of the active power injection intensity of all nodes in the initial candidate source set.

[0119] Step 506: Integrate the power flow direction consistency component, path complexity penalty component, electrical proximity feature component, and power intensity weighted component to generate a causal correlation index.

[0120] The system establishes a multi-dimensional feature evaluation architecture, which incorporates the components representing power flow support, topological simplicity, physical accessibility, and power volume into a unified mathematical calculation logic, thereby serving as the quantitative basis for the final responsibility pre-screening.

[0121] Step 507: Normalize the power flow direction consistency component, path complexity penalty component, power intensity weighted component, and electrical proximity characteristic component to a preset value range.

[0122] Before performing the final fusion calculation, the system verifies the value range of each independently calculated evaluation component. The generation algorithms for the four components all incorporate a normalization mechanism, and the system checks whether the values ​​of each component are strictly distributed within a preset value range of 0 to 1. This verification process eliminates numerical interference caused by inconsistencies in dimensions in subsequent calculations.

[0123] Step 508: Calculate the product of the normalized components to obtain the causal correlation index.

[0124] The system uses a full multiplication architecture to process the four components.

[0125] ρ_ij=C_flow P_comp W_power D_prox; Wherein, ρ_ij is the causal correlation index, C_flow is the power flow direction consistency component, P_comp is the path complexity penalty component, W_power is the power intensity weighted component, and D_prox is the electrical proximity feature component.

[0126] The full multiplication framework embodies the conditional joint feature in causal judgments, that is, four supporting conditions must be satisfied simultaneously to generate a strong causal relationship.

[0127] As an alternative to this embodiment, when the processor's computing resources are sufficient to handle the dynamic value range, the system can adopt a division architecture that introduces a median regularization benchmark to generate the correlation index.

[0128] ρ_ij_alt=(C_flow P_comp W_power) / (D_elec+D_base); Wherein, ρ_ij_alt is the candidate causal correlation index, C_flow is the power flow direction consistency component, P_comp is the path complexity penalty component, W_power is the power intensity weighted component, D_elec is the normalized electrical distance, and D_base is the regularization baseline parameter.

[0129] In this alternative, the regularization baseline parameter is defined as the median of the normalized electrical distance values ​​for all causal valid node pairs in the current computational scenario. The introduction of this parameter avoids numerical inflation caused by division-by-zero anomalies at extremely close electrical distances.

[0130] Step 509, in which if the causal validity determination result of the transmission path is invalid, the corresponding causal correlation index is directly set to zero.

[0131] Before performing the index generation calculation, the system extracts the causal validity determination results output in the aforementioned steps. If the result is invalid, the system skips the complex component extraction and fusion multiplication calculations and directly assigns a zero value to the causal correlation index variable. This logic isolates physical paths with power flow obstructions, reducing the system's unnecessary computational power consumption.

[0132] Example 6: Based on Embodiments 1 to 5 above, this embodiment further details the specific steps of dynamically generating the screening threshold. Since using a fixed threshold is difficult to adapt to different operating scenarios and varying degrees of limit exceedances in the distribution network, this embodiment provides an adaptive calculation mechanism based on the coupling of correlation distribution statistical characteristics and limit exceedance severity.

[0133] In one possible implementation, the following steps are included: Step 601: Calculate the mean and standard deviation of all non-zero causal correlation indices as distribution characteristics.

[0134] Specifically, the system extracts the causal correlation index matrix output from the aforementioned calculation steps, removes elements with a value of 0, i.e., removes values ​​corresponding to transmission paths deemed invalid due to causality. For the remaining set of valid indicators, the system performs conventional mathematical statistics operations to calculate its arithmetic mean and standard deviation. The mean is calculated by summing all non-zero indicator values ​​and dividing by the total number of valid indicators. The standard deviation is calculated by taking the square root of the sum of the squares of the differences between each non-zero indicator value and the mean, dividing by the total number of valid indicators. These mean and standard deviation constitute the distribution characteristics representing the overall level and dispersion of the indicators in the current screening scenario.

[0135] Furthermore, to prevent extreme degradation in subsequent adaptive adjustments, the system synchronously calculates the first quartile of the non-zero index set, i.e. the boundary value at the 25th percentile of the positive order, and uses it as a backup benchmark for subsequent constraint processing.

[0136] Step 602: Evaluate the average severity of exceeding the limit for nodes based on the node voltage deviation values ​​in the operating status data.

[0137] Furthermore, based on the voltage amplitude of each over-limit node in the operating status data and the corresponding voltage limit benchmark, the node voltage deviation value is calculated, and the average over-limit severity of the over-limit node is evaluated accordingly.

[0138] The system acquires the set of nodes that triggered the current pre-screening process, extracts the actual voltage amplitude of each node and its corresponding limit benchmark. The system calculates the absolute percentage of each node's voltage amplitude deviating from the limit benchmark, and then arithmetically averages the deviation percentages of all nodes in the set to obtain a comprehensive index that quantitatively characterizes the current fault state level.

[0139] Δ_over=SUM(abs(V_i-V_lim) / V_lim) / N_v; Where Δ_over represents the average severity of the over-limit, SUM is the summation operator, abs is the absolute value function, V_i is the real-time voltage amplitude of the i-th over-limit node, V_lim is the voltage limit benchmark corresponding to that node, and N_v is the total number of over-limit nodes. This calculation method outputs a dimensionless objective severity assessment value.

[0140] Step 603: Preset multiple severity ranges for exceeding limits and corresponding adjustment coefficient values ​​for each range.

[0141] To establish a non-linear mapping between severity and screening thresholds, the system divides the configuration parameter table in memory into multiple numerical ranges. The preset ranges can be divided into three levels: slightly exceeding the limit, moderately exceeding the limit, and severely exceeding the limit. The system assigns a fixed adjustment coefficient value to each level.

[0142] As an example, the system defines the range of values ​​below 0.02 as the slight over-limit range and assigns an adjustment coefficient of 0.5, the range of values ​​between 0.02 and 0.05 as the moderate over-limit range and assigns an adjustment coefficient of 1.0, and the range of values ​​not below 0.05 as the severe over-limit range and assigns an adjustment coefficient of 1.5.

[0143] Step 604: Identify the interval to which the average severity of exceeding the limit belongs and assign the corresponding adjustment coefficient value.

[0144] The system reads the calculated average severity of the exceedance and executes the comparison logic of the interval boundaries. If the value falls within a preset interval, the system extracts the coefficient value bound to that interval from the parameter table and assigns it to the adjustment coefficient variable in the current calculation thread.

[0145] Step 605, wherein the higher the average severity of exceeding the limit, the larger the adjustment coefficient value is assigned, so as to reduce the screening threshold and expand the screening range of candidate sources.

[0146] The aforementioned tiered coefficient allocation rule has a specific technical motivation. When a severe limit violation occurs, the grid voltage safety margin is extremely low, and the system needs to retain more nodes for investigation to prevent overlooking critical sources of responsibility. By allocating a larger adjustment coefficient value, the screening baseline will be lowered in subsequent subtraction operations, thereby accepting more candidate sources at the edge of correlation. Conversely, when the limit violation is minor, allocating a smaller adjustment coefficient value keeps the screening baseline high, and the system retains only a very small number of nodes with the highest correlation, thus minimizing computational overhead.

[0147] Step 606: Determine the adjustment coefficient based on the average severity of exceeding the limit, and calculate the screening threshold based on a linear combination of the mean, standard deviation, and adjustment coefficient.

[0148] After obtaining the necessary statistical variables and dynamic coefficients, the system performs adaptive threshold value generation calculation. The system multiplies the standard deviation by the adjustment coefficient, then subtracts the product value from the mean, using this linear combination result as the initial threshold index. To prevent the calculated threshold from being too low and causing screening failure when the correlation distribution is extremely discrete, the system compares the initial calculation result with the first quartile obtained in the previous steps, extracting the larger value as the final threshold index.

[0149] ρ_th_temp=μ_ρ-β σ_ρ; ρ_th = max(ρ_th_temp, Q_1); Wherein, ρ_th_temp is the initial threshold, μ_ρ is the mean of the causal correlation index, β is the adjustment coefficient, σ_ρ is the standard deviation of the causal correlation index, ρ_th is the final generated screening threshold, max is the maximum value function, and Q_1 is the first quartile of all non-zero causal correlation indices. This lower bound constraint logic ensures that the system can stably eliminate at least 25% of candidate sources with the lowest correlation ranking in any discrete distribution scenario, guaranteeing the basic filtering capability of the pre-screening process.

[0150] Example 7: Building upon Example 6 above, this example further details the screening verification and candidate source supplementation mechanisms involved in the process of selecting and outputting a subset of target candidate responsibility sources. Since a single statistical threshold may result in some nodes exceeding the limit having no matching candidate sources or overall over-screening, this example introduces a multi-level verification constraint system encompassing node-level, order-of-magnitude, and power-level constraints.

[0151] In one possible implementation, before calculating the power coverage of the initially selected candidate sources relative to the causal valid candidate source reference set, the screening and verification process includes the following steps: Step 701: For each out-of-limit node, if the candidate source selected in the initial screening is empty, the adaptive screening threshold for that out-of-limit node is locally reduced and re-screened. Alternatively, if the corresponding initial candidate source subset is empty, the screening threshold for the over-limit node is locally lowered for re-screening, and the re-screening result is updated to the initial candidate source subset. Specifically, after performing initial filtering based on an adaptive screening threshold, the system traverses all out-of-bounds nodes, checking whether each out-of-bounds node maps to at least one candidate source. If the candidate source set corresponding to an out-of-bounds node is found to be empty, it indicates that the tracing responsibility chain for that node has broken. The system then initiates a local compensation mechanism for that specific out-of-bounds node. The system extracts the maximum value among all non-zero causal correlation indicators corresponding to that out-of-bounds node and lowers the local adaptive screening threshold to a preset proportion of that maximum value. The preset proportion can be set to 50%. By lowering the threshold, the system re-includes several nodes with relatively high correlation into the candidate sequence, ensuring that each out-of-bounds node has at least one physical tracing object.

[0152] Step 702: If the total number of candidate sources initially selected globally is lower than the preset minimum number, the adaptive screening threshold is lowered globally for re-screening to ensure the topological integrity and scale minimum of the screening results.

[0153] Alternatively, if the total number of candidate sources in the global preliminary candidate source subset is lower than the preset minimum number, the global screening threshold is lowered for re-screening, and the results of the re-screening are updated to the preliminary candidate source subset to ensure the topological integrity and scale minimum of the screening results.

[0154] After completing the single-node level checks, the system aggregates all candidate sources to generate a global set and counts the total number of independent nodes in this set. The system then determines whether this total number is less than a preset minimum limit. The preset minimum limit can be the greater of either an absolute value of 3 or 10% of the initial total number of candidate sources. When the total number is below this minimum limit, it indicates that the pre-screening has excessively compressed the responsibility space, posing an engineering risk of overlooking major responsibility sources. The system then globally lowers the adaptive screening threshold, for example, by decreasing it in a fixed step of 0.05, and re-executes the filtering instructions until the preset minimum limit is met.

[0155] Step 703: Identify nodes with valid causal validity results from the initial candidate source set and construct a baseline set of causal valid candidate sources; based on the screening threshold, select nodes with causal correlation indices not lower than the screening threshold from the initial candidate source set to obtain a preliminary candidate source subset; In detail, after satisfying the topology integrity and scale lower limit, the system further performs a power representativeness check. Traditional coverage checks typically use all similar nodes across the entire network as a benchmark, but this can lead to invalid nodes that have already been physically determined to have power flow obstruction being pulled back into the benchmark pool, causing logical conflicts. Therefore, based on the aforementioned path causality validity judgment results, the system extracts the nodes determined to be valid to form a causal valid candidate source benchmark set. This benchmark set excludes nodes that do not have a physical transmission channel with the over-limit event, constituting a complete set of responsibility space with actual physical significance.

[0156] Step 704: Calculate the power coverage of the initially screened candidate sources relative to the baseline set of causal effective candidate sources; The system extracts the absolute value of the active power injection power of each node in the initially selected candidate source set and sums it. Simultaneously, the system extracts the absolute value of the active power injection power of each node in the baseline set of causal valid candidate sources and performs the same summation. The system then calculates the ratio of the two summed values.

[0157] R_cov=SUM(P_filtered_abs) / SUM(P_valid_abs); Where R_cov is the power coverage, SUM is the summation operator, P_filtered_abs is the absolute value of active power injected into nodes in the initially selected candidate source set, and P_valid_abs is the absolute value of active power injected into nodes in the baseline set of causal valid candidate sources. This ratio objectively quantifies the extent to which the currently retained candidate source set represents the overall physical driving force that generates out-of-limit events.

[0158] Step 705: If the power coverage rate is lower than the preset coverage threshold, candidate sources are added to the preliminary candidate source subset from the causal effective candidate source benchmark set in descending order of causal correlation index, until the coverage threshold is met, thus obtaining the target candidate responsible source subset. If the power coverage rate is not lower than the preset coverage threshold, the preliminary candidate source subset is used as the target candidate responsible source subset.

[0159] The system compares the calculated power coverage rate with a preset coverage threshold, which can be set to 0.8. If the comparison shows a coverage rate < 0.8, the system identifies remaining nodes that exist in the baseline set of valid causal candidate sources but were not included in the initial screening set. The system sorts these remaining nodes in descending order based on their causal correlation index. Subsequently, the system extracts and merges the remaining nodes into the initial screening set one by one according to their ranking, re-executing the power coverage rate calculation logic for each merged node. The system continues to perform this iterative replenishment operation until the recalculated coverage rate is >= 0.8. Through this elastic replenishment strategy, the final target candidate source subset output by the system maintains a high degree of physical responsibility representativeness while significantly reducing the computational load, avoiding the omission of major disaster-causing sources.

[0160] Example 8: Based on the same inventive concept as the above method embodiments, this embodiment further describes in detail the performance evaluation step after outputting the subset of candidate responsibility sources for the target.

[0161] In one possible implementation, after outputting the subset of target candidate responsibility sources, the following steps are also included: Step 801: Calculate the ratio of the number of nodes in the target candidate source subset to the total number of nodes in the initial candidate source set, and the ratio of the squares of the two, respectively, and use them as the node compression ratio and the computation reduction ratio. Alternatively, calculate the node compression ratio by subtracting the ratio of the number of nodes in the target candidate source subset to the total number of nodes in the initial candidate source set; calculate the computational cost reduction ratio by subtracting the ratio of the squares of the two above.

[0162] Alternatively, calculate the ratio of the number of nodes in the target candidate source subset to the total number of nodes in the initial candidate source set, and use this as the node compression ratio; calculate the ratio of the square of the number of nodes in the target candidate source subset to the square of the total number of nodes in the initial candidate source set, and use this as the computational cost reduction ratio. Specifically, to quantify the burden reduction effect of this pre-screening method on the subsequent fine-grained source tracing stage, the system counts in memory the total number of nodes in the initially constructed candidate source set and the number of nodes in the final retained target candidate source subset after causal validity determination, multi-factor association quantification, and adaptive threshold verification. The system calculates the ratio of these two quantities and then subtracts this ratio from one to obtain the node compression ratio.

[0163] R_compress=1-(N_filtered / N_initial); Where R_compress is the node compression ratio, N_filtered is the number of nodes in the target candidate source subset, and N_initial is the total number of nodes in the initial candidate source set.

[0164] Furthermore, since the complexity of fine calculation is usually positively correlated with the square of the number of candidate sources in typical distribution network source tracing calculations, the system further calculates the ratio of the squares of the two quantities mentioned above and subtracts the ratio by one to obtain the computational reduction ratio.

[0165] R_compute=1-(N_filtered^2 / N_initial^2); Where R_compute is the computation reduction ratio, N_filtered is the number of nodes in the target candidate source subset, and N_initial is the total number of nodes in the initial candidate source set.

[0166] The two indicators mentioned above objectively reflect the engineering efficiency of the pre-screening process in eliminating invalid computational redundancy.

[0167] Step 802: The node compression ratio, computation reduction ratio, and target candidate source of responsibility subset are combined as the pre-screening performance evaluation results and output to the subsequent fine-grained source tracing module.

[0168] In detail, the system appends the performance evaluation indicators calculated above to the data structure of the pre-screening result table. The fields of this pre-screening result table include the number of the node exceeding the limit, the type of exceeding the limit, the sequence of target candidate responsible source nodes, the causal correlation index value corresponding to each candidate source, the injection power value of each candidate source, and the node compression ratio and computational reduction ratio. The system transmits this complete data table to the subsequent fine-grained contribution calculation module through a system interface or message middleware. After receiving the data, the fine-grained contribution calculation module only needs to perform fine-grained sorting and source tracing analysis on the highly correlated nodes in this subset, forming a complete two-stage source tracing analysis framework closed loop.

[0169] Taking a real distribution network example with 50 load nodes and 10 photovoltaic nodes, when the system detects 3 nodes exceeding the lower limit, the initial candidate source set contains 50 load nodes. After causal validity determination and adaptive threshold screening by this method, the target candidate responsibility source subset is reduced to 12 nodes, corresponding to a node compression ratio of approximately 76% and a computational reduction ratio of approximately 94%. The results of the above embodiments show that the method of the present invention can effectively eliminate irrelevant nodes that do not have a physical causal relationship with the exceeding limit event, significantly reducing the computational scale of subsequent fine-grained tracing. Those skilled in the art will understand that the specific compression ratio may vary depending on the distribution network scale, topology, and operating conditions.

[0170] Example 9: In a certain scenario, the method of the present invention includes the following steps: S1 Distribution network operation data acquisition and over-limit event detection.

[0171] Read the topology parameters of the distribution network, including the set of node numbers, branch connection relationships, and resistance and reactance parameters of each branch. Based on this, construct a node adjacency matrix describing the network connection relationships and a branch parameter table storing branch impedance information.

[0172] The system collects real-time operating data of the distribution network at the current moment, including voltage amplitude measurements of each node, active and reactive power flow measurements of each branch, active power output of each photovoltaic node, and active power consumption of each load node, and integrates them to form a complete system operating status set.

[0173] Based on the voltage over-limit judgment criteria, the voltage of each node in the system operating state set is scanned and detected one by one. Nodes with voltage amplitude exceeding the upper limit threshold (1.07 pu) are identified as nodes exceeding the upper limit, and nodes with voltage amplitude below the lower limit threshold (0.90 pu) are identified as nodes exceeding the lower limit. A list of over-limit nodes and their corresponding over-limit type identifiers are generated.

[0174] A type mapping for candidate responsibility sources is established based on the over-limit type identifier: for over-limit events, all photovoltaic access nodes are extracted to form an initial candidate source set; for over-limit events, all load nodes are extracted to form an initial candidate source set. Simultaneously, the power injection value of each candidate source is recorded to form a candidate source power vector.

[0175] S2, Node topology path search and connected graph construction.

[0176] Based on the node adjacency matrix, a depth-first search algorithm is used to traverse all simple paths between each out-of-bounds node in the list of out-of-bounds nodes and each candidate source in the initial candidate source set. For each path, the sequence of intermediate nodes and branch sequences it passes through are recorded to form a complete path information database.

[0177] For each path in the full path information database, calculate its topology hop count (number of branches traversed) and cumulative impedance distance. The cumulative impedance distance is defined as the sum of the impedance magnitudes of all branches along the path: ; in, This represents the set of branches along the path. and These are the resistance and reactance of branch l (unit: ohms). This represents the cumulative impedance distance of the path (in ohms). The calculation results are appended to the path information to update the full path information database.

[0178] When multiple paths exist for the same over-limit node-candidate source node pair, the path with the smallest cumulative impedance distance is selected as the primary path, while the second shortest path is retained as an alternative path, thus constructing a preferred path set. This set provides the path basis for subsequent power flow direction analysis.

[0179] S3, Real-time Power Flow Direction Judgment and Path Causality Validity Marking

[0180] Active power flow measurements for each branch are extracted from the system's operational status database, and the power flow direction is determined based on the sign of the flow. Branch flow direction coding rules are defined as follows: using the lower-numbered end of the branch as the reference, if the active power flow moves from the lower-numbered end to the higher-numbered end, the direction code is +1; if the active power flow moves from the higher-numbered end to the lower-numbered end, the direction code is -1; if the absolute value of the active power flow is less than a set threshold (0.01 MW), the direction code is 0, indicating that the branch is in a light-load or balanced state. Direction coding is performed on all branches to generate branch flow direction vectors.

[0181] Based on the physical logic of out-of-bounds types and responsibility propagation, the expected power flow direction for each path is derived. The core principle is: if a candidate source is responsible for an out-of-bounds event, then power must be able to be propagated from the candidate source to the out-of-bounds node (or from the vicinity of the out-of-bounds node to the candidate source, depending on the out-of-bounds type).

[0182] For events exceeding the upper limit, the active power injected by the photovoltaic node causes a voltage rise at nodes along the path. Therefore, the desired power flow direction is from the photovoltaic node (candidate source) to the node exceeding the upper limit. For each path in the preferred path set connecting the photovoltaic candidate source and the node exceeding the upper limit, the desired direction is determined for each branch: if the order of the branches in the path is from the candidate source side to the node exceeding the upper limit side, the desired direction is encoded as +1; otherwise, it is -1.

[0183] For a lower limit event, the active power consumed by the load node causes a voltage drop at nodes along the path, and the desired power flow direction is from the power source to the load node (candidate source). For each path connecting the load candidate source and the lower limit node, the desired direction is determined for each branch: if the branch direction points to the load node, the desired direction is encoded as +1; otherwise, it is -1.

[0184] The expected direction encoding sequence of each path is stored as a path expected direction matrix.

[0185] For each path in the preferred path set, the actual power flow direction code (from the branch power flow direction vector) of each branch it passes through is compared with the expected direction code (from the path expected direction matrix), and the direction consistency score of a single branch is calculated: ; in, Encode the actual power flow direction of branch l. Encode the desired power flow direction for branch l. The score is given for the directional consistency of this branch.

[0186] A score of 1 indicates that the power flow direction is completely consistent with the responsibility transmission direction; a score of 0.5 indicates that the branch is in a light-load state (neither hindering nor supporting the responsibility transmission); and a score of 0 indicates that the power flow direction is opposite to the responsibility transmission direction (there is power flow obstruction).

[0187] The overall power flow consistency index of the path is calculated based on the directional consistency scores of each branch along the path. Considering that responsibility transfer requires all branches along the path to meet the directional conditions, a weighted geometric average is used for aggregation. ; in, The weight of branch l is defined as the proportion of the impedance magnitude of that branch to the total path impedance: ; The physical meaning of this weighting design is that branches with higher impedance have a more significant impact on voltage, and the consistency of their power flow direction should account for a larger proportion in the overall evaluation. The value range is [0,1]. The closer the value is to 1, the more consistent the power flow direction and the responsibility transmission direction of the path are.

[0188] Each path is categorized by causal validity based on the overall trend consistency index. A validity threshold is then set. ,like If so, then mark the path as a "causal valid path"; if If so, the path is marked as a "causal invalid path".

[0189] The threshold is set based on the following: when more than 70% of the weighted branches on a path have power flow obstruction, the path can be considered to lack the physical conditions for responsibility transfer. All paths in the preferred path set are marked, generating a path validity mark table.

[0190] S4. Comprehensive Quantification of Causal Relationships Based on Multi-Factor Integration

[0191] For each pair of over-limit node-candidate source nodes, extract the cumulative impedance distance of its main path from the preferred path set. Convert it to a normalized electrical distance component: ; in, This represents the maximum cumulative impedance distance along the main path for all pairs of out-of-limit nodes and candidate source nodes. This is the normalized electrical distance component, with a value range of (0,1). The smaller the value, the closer the electrical distance and the higher the correlation.

[0192] Extract the overall power flow consistency index of the main path from each pair of out-of-limit nodes and candidate source nodes from the path validity label table. Directly used as a component of trend consistency : ; The value ranges from [0,1]. The larger the value, the more the power flow direction supports the transmission of responsibility and the higher the correlation.

[0193] Considering that power is split and lost when passing through intermediate nodes during transmission, the longer the path, the more significant the attenuation of responsibility transmission. Define a path complexity penalty component: ; in, The topological hop count of the main path (number of branch paths along the way). This is the attenuation coefficient (with a value of 0.2). This is the path complexity penalty component, with a value ranging from (0,1). The more hops there are, the smaller the penalty component becomes, and the lower its contribution to the correlation.

[0194] The power injection intensity of a candidate source directly affects its potential contribution to out-of-limit events. The absolute value of the active power injection of candidate source j is extracted from the candidate source power vector. Calculate the power intensity weighted components: ; in, For the initial candidate source set, The normalized power intensity component has a value range of (0,1). The larger the power injection of a candidate source, the stronger its potential responsible contribution capability.

[0195] By fusing the above four components, a comprehensive index of causal correlation between the over-limit node i and the candidate source j is constructed: ; in, To prevent small positive numbers from being divided by zero, It is a comprehensive index of causal correlation (dimensionless).

[0196] For the cross-limit node-candidate source node pairs marked as "causal invalid path" in the path validity labeling table, their causal correlation degree is directly set to 0, and the above calculation is not required.

[0197] For all valid cross-limit node-candidate source node pairs, calculate the causal correlation degree and summarize them to form a causal correlation degree matrix.

[0198] S5. Adaptive Determination of Relevance Threshold and Candidate Source Selection

[0199] Perform statistical analysis on all non-zero elements in the causal correlation matrix to calculate its mean, standard deviation, and quantile characteristics: ; ; in, This is the set of cross-boundary node-candidate source node pairs with non-zero causal correlation. The number of elements in the set. The mean of the correlation degree The standard deviation of the correlation is given.

[0200] Simultaneously calculate the quartiles of the correlation, including the 25th percentile. , median and the 75th percentile It is used to describe the central tendency and dispersion of the correlation distribution.

[0201] Quantify the severity of current limit violations as the basis for threshold adjustment. Define the limit violation severity index:

[0202] in, For the set of nodes that exceed the limit, The number of nodes exceeding the limit, The voltage amplitude at the over-limit node i, The corresponding voltage limits are (1.07 pu for the upper limit and 0.90 pu for the lower limit). The average severity of exceeding the limit (dimensionless).

[0203] Based on the correlation distribution characteristics and the severity of exceeding the limit, the screening threshold is adaptively determined:

[0204] Among them, the adjustment coefficient Based on the severity of exceeding the limit Dynamically determined:

[0205] The design logic of this adaptive mechanism is as follows: When the limit is slightly exceeded, a higher threshold is used. Only the candidate sources with the highest relevance are retained, thus improving the screening accuracy.

[0206] When the limit is severely exceeded, a lower threshold is used. Expand the range of candidate sources to avoid overlooking important sources of responsibility.

[0207] To prevent filtering from failing due to excessively low thresholds, a lower threshold constraint is set: ; That is, the threshold must not be lower than the 25th percentile of the correlation distribution to ensure that at least the 25% of candidate sources with the lowest correlation are screened out.

[0208] Based on the calculated adaptive threshold The causal correlation matrix is ​​then filtered. For each node i that exceeds the threshold, candidate sources with a correlation degree not lower than the threshold are selected: ; The filtered candidate source sets corresponding to each out-of-limit node are summarized. If the correlation between the same candidate source and multiple out-of-limit nodes exceeds the threshold, the candidate source is retained only once in the final result, forming a filtered candidate source set.

[0209] The validity of the filtered candidate source set is validated to ensure that the filtering results meet the following constraints: Constraint 1: Each out-of-bounds node must correspond to at least one candidate source. If an out-of-bounds node has no corresponding candidate source after filtering, the threshold for that out-of-bounds node will be reduced to 50% of its maximum correlation value for re-filtering.

[0210] Constraint 2: The number of candidate sources after filtering shall not be less than the preset minimum value (whichever is greater, either 3 or 10% of the initial number of candidate sources). If this condition is not met, the threshold shall be lowered globally until it is met.

[0211] Constraint 3: The total power coverage of candidate sources after screening shall not be less than 80%. ; If the requirement is not met, candidate sources will be added in descending order of relevance until the requirement is met.

[0212] After the verification is passed, the final set of candidate sources after filtering is output.

[0213] S6 Pre-screening Results Output and Performance Evaluation

[0214] The filtered candidate source set and its causal correlation value with each out-of-limit node are structured and organized to generate a pre-screening result table, which includes the following fields: out-of-limit node number, out-of-limit type, list of filtered candidate source numbers, causal correlation value of each candidate source, and power injection value of each candidate source.

[0215] Calculate the performance metrics for pre-screening, including candidate source compression ratio and computational cost reduction ratio: ; ; in, The compression ratio of the candidate source. This represents the computational reduction ratio (assuming the computational complexity of subsequent contribution calculations is proportional to the square of the number of candidate sources). The performance metrics are then appended to the pre-screening results table.

[0216] The pre-screening results table is passed to the subsequent fine contribution calculation module, which only performs contribution analysis and sorting on the screened candidate sources, thus completing the pre-screening stage of the "pre-screening-fine calculation" two-stage source tracing framework.

[0217] According to one aspect of this application, the process for determining the consistency of power flow direction in a single branch can also be as follows: Considering that qualitative consistency in power flow direction alone cannot distinguish the differences in causal transmission intensity under different power carrying levels, the ratio of active power flow amplitude of a branch to the injected power of candidate sources is introduced into the consistency evaluation, and a branch consistency score with enhanced power transmission intensity is defined.

[0218] For each path in the preferred path set, the active power flow amplitude of each branch it passes through is extracted from the system operating status set. Extract the corresponding candidate source from the candidate source power vector The absolute value of active injection Calculate the power transmission intensity ratio of each branch: ; in MW is used to prevent division by zero of small positive numbers. The physical meaning of this ratio is: the ratio of the actual active power carried by the branch to the power injected by the candidate source. If... This indicates that the power carried by the branch is not less than the amount injected by the candidate source, and it has sufficient power transmission capability; if This indicates that the branch power is lower than the candidate source injection amount, and the power of the candidate source has been attenuated due to shunting or loss before reaching the branch, thus weakening the causal transmission strength.

[0219] Based on the comparison results of the actual power flow direction coding and the expected direction coding of the branch, the consistency score of the branch with enhanced power transmission strength is defined as follows:

[0220] in This is the basic score for the lightly loaded branch.

[0221] When the actual direction matches the desired direction, the score is the smaller of the power transmission intensity ratio and 1. The physical basis of this design is: if the branch's carrying power is much smaller than the candidate source's injected power (e.g., a candidate PV injects 5 MW but the branch only has 0.5 MW of forward power flow), it indicates that most of the candidate source's power has been diverted upstream or consumed by loads along the route, with only a small amount of power transmitted through the branch. The causal relationship is weak, and the score is... If the branch carrying power is not lower than the candidate source injection power (the branch gathers the power of multiple sources), it means that the branch fully meets the conditions for transmitting the candidate source power, and the score is taken as the upper limit of 1.

[0222] When lightly loaded ( The score is a fixed value. The reason for reducing the light load score compared to the original 0.5 is that there is almost no effective power transmission on light load branches, and their support for causal transmission is extremely weak. A score of 0.5 would lead to an excessively high consistency index for paths containing multiple light load branches.

[0223] When the direction is reversed, the score is 0.

[0224] The range of values ​​for this score remains [missing information]. It is fully compatible with the weighted geometric mean aggregation formula and requires no modification to subsequent steps. Compared with the discrete three-level scoring of the original scheme, the continuous scoring makes the "bottleneck identification" effect of the geometric mean more refined: when the power carrying ratio of a certain branch on the path is low, the continuous low scores of the branch are transmitted to the overall score of the path through the geometric mean, which naturally reflects the weakest link in power transmission on the path, rather than the rough judgment of "full score as long as the direction is consistent" in the original scheme.

[0225] According to one aspect of this application, the expectation calculation of cross-boundary events in topology location awareness can also be: For events that exceed the lower limit, since the voltage influence mechanism of candidate loads on the nodes that exceed the limit depends on their relative positions in the radial topology, a segmented expected direction derivation method based on the lowest common ancestor analysis is adopted.

[0226] Assume the distribution network has substation nodes as root nodes. A radial tree topology. For candidate load nodes... and nodes that exceed the lower limit First, determine the lowest common ancestor node of the two nodes in the tree topology. The definition of the lowest common ancestor is: At the same time and Ancestor node (located from root to ) and root to On the path), and is the furthest from the root node among all common ancestor nodes. In a tree topology, the LCA can be determined by: from... and Backtrack to the root node and record the set of nodes along the path. and The LCA is the node farthest from the root in the intersection of two sets.

[0227] According to LCA and , Based on this relationship, the expected direction of events exceeding the lower limit can be deduced into three scenarios: Scenario A: LCA = (The node that exceeds the limit is the ancestor node of the candidate load) In this case, the node exceeds the limit. Located from root to candidate load On the path, that is exist Upstream. The physical mechanism is: load. The absorbed current flows from the power supply side through Flow direction ,exist to A voltage drop occurs on the branch, causing... The voltage decreases. This path is from... arrive The main path (unique in the tree) is expected to have the power flow direction from... Flow direction (That is, flow from the over-limit node side to the load side). Regarding the path... For each branch on the path, the desired direction is encoded as follows: ; in, branch road (Path forward direction encoding). Confidence weight coefficients of all branches on the path. (Complete confidence) indicates that the causal transmission mechanism is clear in this case.

[0228] Scenario B: LCA = (The candidate load is the ancestor node of the overload node)

[0229] In this case, the candidate load Located from the root to the over-limit node On the path, that is exist Upstream. The physical mechanism is: load. Power consumption leads to The node voltage drops, and the voltage drop edge comes from arrive The path spreads downstream, causing The voltage decreases synchronously. At this time, the desired power flow direction is from... Flow direction (That is, the power supply flows from the upstream node to the downstream node, which is the normal power supply direction). Regarding the path... For each branch on the path, the desired direction is encoded as follows: ; Confidence weight coefficients of all branches on the path The reason for using 0.8 instead of 1.0 is that the causal mechanism of scenario B is voltage propagation rather than direct current conduction, and the intensity of the influence is regulated by the shunting effect of the load along the way and reactive power compensation. The causal relationship is slightly weaker than that of scenario A.

[0230] Case C: LCA = ( and (The two are located in different branches)

[0231] In this case, the node exceeds the limit. and candidate load Located in common ancestor Within different subtrees, the path is divided into two segments: to Duan He to The physical mechanism is: load. The absorption current causes the current to flow from the root to the surface. The current and voltage drop increase on the common branch. The point voltage decreases, and this decrease propagates to The branch it belongs to.

[0232] right to The first half of the path (segment) is expected to have a current flow direction from... Flow direction (Current flows towards the load): ; right to The latter half of the path (the segment where the causal mechanism is voltage propagation rather than direct current conduction) has a desired direction from... Flow direction (Normal power supply direction): ; Note this. Need to be to Recalculate the path direction of the segment (i.e., from the direction of the path) arrive The direction is the direction of the path forward.

[0233] to Confidence weight coefficient of segment This reflects the indirectness and uncertainty of cross-branch causal transmission. to Confidence weight coefficient of segment .

[0234] In the weighted geometric mean calculation, confidence weights are incorporated into the branch weights. The corrected combined branch weights are:

[0235] in The original impedance is proportionally weighted. This is the confidence weight coefficient for the path segment to which the branch belongs. (The part from step S3 is missing.) Replace with The rest of the calculation process remains unchanged. This design reduces the weight of path segments with weaker causal mechanisms (such as cross-branch segments in case C) in the geometric mean, thereby naturally reducing the consistency index of the entire path and reflecting its weaker causal relationship.

[0236] According to one aspect of this application, the calculation of exponential proximity multiplicative causal correlation can also be: Considering the mathematical defect that the correlation formula of the division structure has no upper bound on the value range, and that the sensitivity of the node voltage to the injected power at the far end in the distribution network has an exponential decay characteristic with electrical distance, the inverse proportional relationship of electrical distance is replaced by an exponentially decaying electrical proximity, and a causal correlation formula of the full multiplication structure is constructed.

[0237] In calculating the normalized electrical distance component Based on this, the electrical proximity component is further defined:

[0238] in The distance attenuation sensitivity coefficient is set to 2.0. Its physical basis is that in a radial distribution network, the sensitivity of node voltage to branch injected power decreases approximately exponentially with respect to electrical distance. When... (Very close) time ;when (Medium distance) ;when (Furthest) time The dynamic range of proximity is approximately 7:1, which effectively distinguishes candidate sources at different distances.

[0239] Replace the division structure of the comprehensive index of causal correlation with a full multiplication structure:

[0240] All components of this formula are normalized to... ,therefore Its range is strictly bounded.

[0241] If any component is zero, the correlation is zero, reflecting the causal judgment criterion that power flow consistency, path simplicity, power intensity, and electrical proximity must be satisfied simultaneously.

[0242] The degree to which the trend supports the transmission of responsibility. The degree to which path length diminishes the impact of responsibility transfer. The potential strength of the influence of candidate source power on exceeding the limit. This refers to the extent to which electrical distance affects voltage.

[0243] For node pairs marked as "causal invalid path" in the path validity labeling table, the processing method is the same as the original solution: directly remove... Set to 0.

[0244] In another embodiment, a pre-screening system for the causal correlation of candidate sources of voltage limit violation is used to implement the pre-screening method for the causal correlation of candidate sources of voltage limit violation, comprising: The data acquisition and identification module acquires the operating status data and topology data of the distribution network, identifies voltage over-limit nodes, and constructs an initial candidate source set. The transmission path determination module determines the transmission path between the over-limit node and the initial candidate source set based on the topology data. The transmission path causal validity determination module obtains the expected power flow direction of the transmission path based on the operating status data, and compares it with the actual power flow characteristics to determine the causal validity of the transmission path. The causal correlation index construction module constructs a causal correlation index based on physical topology characteristics and actual power flow characteristics for transmission paths with valid causal validity. The filtering module dynamically generates a filtering threshold based on the distribution characteristics of the causal correlation index, and filters and outputs a subset of target candidate responsibility sources from the initial candidate source set according to the generated filtering threshold.

[0245] In another embodiment, a computer storage medium stores a computer program thereon, wherein when the computer executes the computer program, it implements the pre-screening method for causal correlation of voltage over-limit candidate responsibility sources as described in any of the preceding embodiments.

[0246] The specific implementation method is the one described above, and will not be repeated here.

[0247] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for pre-screening causal correlation of candidate sources of voltage exceeding limits, characterized in that, Includes the following steps: Acquire the operating status data and topology data of the distribution network, identify voltage over-limit nodes, and construct an initial candidate source set; Based on the topology data, determine the transmission path between the over-limit node and the initial candidate source set; The expected power flow direction of the transmission path is obtained based on the operating status data, and compared with the actual power flow characteristics to determine the causal validity of the transmission path. For transmission paths with valid causal validity, a causal correlation index is constructed based on physical topology characteristics and actual power flow characteristics. Based on the distribution characteristics of the causal correlation index, a screening threshold is dynamically generated, and a subset of target candidate responsibility sources is selected from the initial candidate source set according to the generated screening threshold.

2. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, Determining the transmission path between the out-of-limit node and the initial candidate source set includes: Construct node connection relationships and branch impedance magnitudes based on the topology data; Traverse all simple paths between the out-of-bounds node and the nodes in the initial candidate source set; During the traversal, if the search depth exceeds the preset maximum topology jump limit, the search of the current branch will be stopped; The path that meets the hop count limit is taken as the transmission path, and the impedance modulus of each branch on the path is accumulated to obtain the cumulative impedance distance.

3. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, Obtaining the desired power flow direction of the transmission path and comparing it with the actual power flow characteristics includes: The physical logic for responsibility transfer is determined based on the type of over-limit, and the physical logic for responsibility transfer is a causal driving relationship between node injection or absorption power and voltage fluctuations determined based on circuit theory. Based on the responsibility transfer physical logic determined by the out-of-limit type, assign the desired power flow direction code to each branch on the transmission path; The active power sign and amplitude of each branch power flow are extracted from the operating status data as actual power flow characteristics; By comparing the active power symbol of each branch power flow with the expected power flow direction code, the direction consistency determination result at the branch level is obtained.

4. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 3, characterized in that, Assigning desired power flow direction codes to each branch on the transmission path includes: Determine the lowest common ancestor node of the out-of-bounds node and the initial candidate source nodes in the initial candidate source set in the radial topology; If the out-of-limit node is the lowest common ancestor node, the expected power flow direction is determined to be from the out-of-limit node to the initial candidate source node, and the corresponding path segment is assigned a first confidence weight. If the initial candidate source node is the lowest common ancestor node, the expected power flow direction is determined to be from the initial candidate source node to the over-limit node, and the corresponding path segment is assigned a second confidence weight. Conversely, the transmission path is divided into a first path segment from the initial candidate source to the lowest common ancestor, and a second path segment from the lowest common ancestor to the over-limit node, and the expected power flow direction and corresponding confidence weight of each path segment are determined respectively.

5. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 4, characterized in that, Determining the causal validity of the transmission path includes: The ratio of the amplitude of the branch power flow to the active power injection intensity of the initial candidate sources in the initial candidate source set is used as the power transmission intensity ratio. If the direction consistency determination result of the branch is consistent, then the smaller value between the power transmission strength ratio and the predetermined value is taken as the consistency score of the corresponding branch. If the absolute value of the branch power flow is less than the preset light load threshold, then the preset light load base score is assigned. If the direction consistency judgment results of the branch are opposite, then zero points are assigned; The causal validity determination result of the transmission path is determined based on the consistency score of each branch on the transmission path.

6. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 5, characterized in that, The determination of the causal validity of the transmission path also includes aggregating the consistency scores of each branch to obtain power flow direction consistency features. These power flow direction consistency features include: Calculate the proportion of the impedance magnitude of each branch to the total impedance of the transmission path to obtain the initial weight of each branch; The initial weights are corrected using the confidence weights corresponding to each path segment to obtain the comprehensive weights; The consistency scores of each branch on the transmission path are aggregated using a weighted geometric mean based on the comprehensive weights to obtain the power flow direction consistency characteristics.

7. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, The causal correlation index is constructed based on the physical topological characteristics and the actual power flow characteristics, including: Extract the trend consistency component based on the trend direction consistency characteristics; The path complexity penalty component and electrical proximity feature component are extracted based on the physical topology attributes of the transmission path. The power intensity weighted component is extracted based on the power injection intensity of the nodes in the initial candidate source set; By integrating the power flow consistency component, the path complexity penalty component, the electrical proximity feature component, and the power intensity weighted component, a causal correlation index is generated.

8. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, Dynamically generated filtering thresholds include: The mean and standard deviation of all non-zero causal correlation indices are calculated as distribution characteristics; The average severity of the out-of-limit node is assessed based on the node voltage deviation values ​​in the operational status data. An adjustment coefficient is determined based on the average severity of exceeding the limit, and a screening threshold is calculated based on a linear combination of the mean, the standard deviation, and the adjustment coefficient.

9. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, The subset of candidate sources of responsibility for the filtered output includes: Nodes whose causal validity determination results are valid are identified from the initial candidate source set, and a causal valid candidate source benchmark set is constructed; Based on the screening threshold, nodes with a causal correlation index not lower than the screening threshold are selected from the initial candidate source set to obtain a preliminary candidate source subset; Calculate the power coverage of the candidate sources in the preliminary candidate source subset relative to the causal effective candidate source reference set; If the power coverage rate is lower than the preset coverage threshold, then candidate sources are added from the causal effective candidate source benchmark set in descending order of causal correlation index until the coverage threshold is met, thus obtaining a subset of target candidate responsible sources.

10. The pre-screening method for causal correlation of candidate sources of voltage exceeding limits according to claim 1, characterized in that, After filtering and outputting a subset of candidate sources of responsibility, the following steps are also included: Calculate the ratio of the number of nodes in the target candidate source subset to the total number of nodes in the initial candidate source set, and the ratio of the squares of these two terms, respectively, and use them as the node compression ratio and the computation reduction ratio. The node compression ratio, the computational cost reduction ratio, and the target candidate responsibility source subset are combined and output as the pre-screening performance evaluation results.

11. A pre-screening system for the causal correlation of candidate sources of voltage exceeding limits, characterized in that, A method for pre-screening the causal correlation of candidate sources of voltage exceeding limits as described in any one of claims 1-10 includes: The data acquisition and identification module acquires the operating status data and topology data of the distribution network, identifies voltage over-limit nodes, and constructs an initial candidate source set. The transmission path determination module determines the transmission path between the over-limit node and the initial candidate source set based on the topology data. The transmission path causal validity determination module obtains the expected power flow direction of the transmission path based on the operating status data, and compares it with the actual power flow characteristics to determine the causal validity of the transmission path. The causal correlation index construction module constructs a causal correlation index based on physical topology characteristics and actual power flow characteristics for transmission paths with valid causal validity. The filtering module dynamically generates a filtering threshold based on the distribution characteristics of the causal correlation index, and filters and outputs a subset of target candidate responsibility sources from the initial candidate source set according to the generated filtering threshold.

12. A computer storage medium having a computer program stored thereon, characterized in that, When the computer executes the computer program, it implements the pre-screening method for causal correlation of voltage over-limit candidate responsibility sources as described in any one of claims 1-10.