Linkage trip failure prediction system and method based on disturbance trajectory features

Abstract

The application discloses a kind of linkage trip fault prediction methods based on perturbation trajectory characteristics, comprising the following steps: A, collection power grid historical operating condition data, extract the perturbation trajectory characteristics of each node, calculate path search parameter;B, construct power grid Voronoi diagram, generate the initial reference path of fault propagation, and carry out discretization processing and inflection point simplification to initial reference path;C, establish the fault potential field of fusion disturbance characteristics, respectively calculate the attractive force potential energy and repulsive force potential energy of node;D, based on ellipsoid sampling optimization algorithm and combined with fault potential field, path search is generated, and global candidate fault path set is generated;E, the total propagation cost of candidate fault path is corrected, and the five-dimensional risk composite entropy of path is calculated;F, all candidate fault paths are arranged in descending order according to risk composite entropy, and fault path is output according to preset requirement.The application can improve the deficiency of prior art, improve the prediction accuracy of linkage trip fault.

Description

Technical Field

[0001] This invention relates to the field of power system safety early warning technology, and in particular to a system and method for predicting tripping faults based on disturbance trajectory characteristics. Background Technology

[0002] Grid-linked tripping, also known as cascading tripping or chain-reaction tripping, is a typical chain-like propagation phenomenon of faults in power systems. Essentially, it is a chain process in which an initial disturbance triggers the dynamic transmission of power flow and voltage, cascadingly triggering equipment tripping at each level. The fault continues to propagate and amplify along the grid topology, eventually evolving into a large-scale blackout or even system-wide instability. Especially in recent years, the proportion of renewable energy integrated into my country's power grid has significantly increased, leading to a decrease in system inertia and an increase in operational uncertainty. This has significantly increased the triggering risk and propagation speed of cascading tripping, making it one of the core risks threatening the safe and stable operation of the power grid.

[0003] Existing interlocking tripping fault prediction technologies are mainly divided into two categories: physical simulation prediction and data-driven machine learning prediction. The former relies on the physical equations of the power system to simulate the fault, resulting in low computational efficiency, insufficient real-time performance, and poor matching with the actual operating conditions of the power grid; the latter heavily relies on high-quality samples for early training, resulting in poor generalization ability and robustness. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a system and method for predicting linkage tripping faults based on disturbance trajectory characteristics, which can overcome the shortcomings of the prior art and improve the prediction accuracy of linkage tripping faults.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.

[0006] A method for predicting tripping faults based on disturbance trajectory characteristics includes the following steps:

[0007] A. Collect historical power grid operating data, extract disturbance trajectory characteristics of each node, and calculate path search parameters;

[0008] B. Construct a Voronoi diagram of the power grid, generate an initial reference path for fault propagation in the Voronoi diagram, and discretize and simplify the inflection points of the initial reference path.

[0009] C. Establish a fault potential field with integrated disturbance characteristics, and calculate the attractive and repulsive potential energy of the nodes respectively.

[0010] D. Based on the ellipsoidal sampling optimization algorithm and combined with the fault potential field, a global candidate fault path set is generated for path search.

[0011] E. Correct the total propagation cost of the candidate failure path and calculate the five-dimensional risk composite entropy of the path;

[0012] F. Sort all candidate fault paths in descending order of risk composite entropy. According to preset requirements, select fault paths that meet the preset requirements in sequence, starting from the fault path with the highest risk composite entropy.

[0013] Preferably, in step A, the time of the initial disturbance is taken as the zero point of time, and the following four types of original trajectory features are calculated.

[0014] The maximum rate of decrease of the node voltage amplitude after the disturbance: , Let be the voltage magnitude at node x at time t. The sampling interval;

[0015] Power fluctuation amplitude: , Let be the total incoming power of node x at time t;

[0016] Disturbance propagation delay: , The moment when the voltage at node x first exceeds the limit. The time when the initial disturbance occurs;

[0017] Trajectory volatility: , for Standard deviation of time series data for The mean of time series data.

[0018] Preferably, in step A, the voltage time sequence of each node is scaled to a specific value. The coarsening process, the coarsening sequence at the ε-th scale is: , N is the total length of the original time series, and L is the maximum number of layers at the largest scale. For each scale of coarse-grained sequence, a sliding window is used to calculate the permutation entropy. w is the width of the sliding window, and m is the embedding dimension. Let x be the probability of the i-th permutation pattern occurring; take the weighted average of the permutation entropies of all scales and all sliding windows to obtain the perturbation trajectory entropy fingerprint of node x. , The weights for the ε-th scale are... ;

[0019] Path search parameters include,

[0020] Node perturbation sensitivity coefficient: , These are the weighting coefficients;

[0021] Disturbance amplification factor from line i to line j: .

[0022] Preferably, step B, generating the initial reference path for fault propagation, includes the following steps:

[0023] Based on historical records, the k nodes with the highest failure rates across the entire network are selected as seed nodes, resulting in a set of seed nodes. For all non-seed nodes and all seed nodes, calculate the electrical distance one by one. For each non-seed node, assign it to the Voronoi region of the corresponding seed node according to the minimum electrical distance criterion; the Voronoi region corresponding to the i-th seed node is defined as follows: , The electrical distance between two nodes, nodes and The method for calculating the electrical distance between them is as follows: , Let be the self-impedance of node i. Let J be the self-impedance of node j. Let be the mutual impedance between node i and node j;

[0024] If node x satisfies the condition that there are two distinct seed points , making Then node x is a region and The common boundary nodes are defined, where ε is the boundary determination tolerance; if two Voronoi regions have at least one common boundary node, the two regions are determined to be electrically adjacent, and a region adjacency matrix is ​​constructed. , ;

[0025] Using each Voronoi region as a vertex and the region adjacency matrix A as the connectivity basis, a weighted region graph is constructed; two adjacent regions and (Right now The edge weights between () are represented by the crossing cost. , Let $\frac{i}{j}$ be the cost of traversing from region $i$ to region $j$. The average protection margin coefficient for all common boundary nodes of the two regions is used. Dijkstra's algorithm is used to search for the shortest path from the starting region to the ending region to obtain the initial reference path at the region level. Within each region, the node with the highest failure probability is selected as a transfer point, connecting the starting point, the transfer points in each region, and the ending point to obtain the initial reference path at the node level. .

[0026] Preferably, step B, which discretizes the initial reference path and simplifies the inflection points, includes the following steps:

[0027] Calculate the initial reference path upper node To the node electrical length The total electrical length of the entire path is Where p is the number of nodes in the initial reference path; with a fixed electrical step size Uniform sampling along the path transforms the unevenly spaced original path nodes into equally spaced scattered points; the total number of discrete points is... The cumulative electrical distance corresponding to the kth discrete point is The position of the kth discrete point is calculated by linear interpolation. , where i satisfies Then, establish an inflection point set, removing discrete points that are not inflection point positions; add the starting node to the inflection point set, and set the index pointer at the starting node position; starting from the last discrete point, traverse backwards to find the discrete point farthest from the current pointer that meets the following conditions: the direct fault propagation probability between this discrete point and the index pointer is not lower than a set threshold, and the protection margin of all intermediate actual nodes on the path connecting this discrete point and the index pointer is lower than a set threshold; if a discrete point that meets the conditions is found, add this discrete point to the inflection point set, and move the index pointer to the position of the latest discrete point added to the inflection point set; repeat the above process until the index pointer traverses all discrete points to obtain the final inflection point set.

[0028] Preferably, step C, establishing the fault potential field that integrates the disturbance characteristics, includes the following steps:

[0029] Define the attractive potential energy at node x as follows:

[0030] in, The gain coefficient of the attraction potential field. Let x be the static failure probability of node x. Let x be the electrical distance from node x to the target node. Critical electrical distance threshold; attractive force ;

[0031] Define the repulsive potential energy at node x as follows: , The gain coefficient of the repulsive potential field. Let x be the electrical distance from node x to the nearest strongly protected node. This is the critical distance for repulsion. The protection margin coefficient for node x; repulsive force. .

[0032] Preferably, in step D, an asymmetric ellipsoidal sampling optimization algorithm is used to generate the global candidate fault path set. The method for establishing the asymmetric ellipsoid is as follows:

[0033] Select the k nodes with the highest failure probability from the inflection point set as beacon points. The beacon points are used as dividing points to split the entire path into k+1 independent search segments; an initial symmetric ellipsoid is constructed for the i-th segment. The initial center point of the ellipsoid is The initial metric matrix is , The direct electrical distance between the two focal points. The cost of the optimal path for the current segment. , , for and The probability of direct fault propagation between them;

[0034] The ellipsoid is asymmetrically corrected, and perturbation features are introduced; the method for correcting the major axis of the ellipsoid is as follows: , The length of the semi-major axis on the starting side. The length of the semi-major axis at the endpoint. For adjustment coefficients, The fault accumulation gain at normalized position s is given, with S=0 at the start point and S=1 at the end point; the center of the corrected ellipsoid is... , Corrected metric matrix .

[0035] Preferably, in step D, the global candidate fault path set includes the following steps.

[0036] Define the fault conductance at sampling point x , This represents the static failure probability of a node. To protect the margin factor, the path proximity factor ; Generate random sampling points within the unit sphere Decompose the metric matrix Obtain the scaling transformation matrix , making Sampling points within the ellipsoid in the world coordinate system are obtained through spatial transformation. ,in This is a coordinate rotation matrix used to rotate local ellipsoid points, aligning the major axis to the direction of the line connecting the two foci, matching the orientation of the world coordinate system; it calculates the fault conductance at each sampling point within the ellipsoid, if... If the sampling point is correct, retain it; otherwise, delete it. Correct the sampling point orientation to obtain the guide sampling point. Where η is the guiding step size, ;

[0037] Find the tree node closest to the guiding sampling point and designate it as the parent node. Using this tree node as the starting point, define the direction pointing to the guiding sampling point as the expansion direction. Starting from the current starting point, advance along the expansion direction by an initial step size to generate candidate nodes. If the direct propagation probability from the starting point to a candidate node is not lower than a set threshold, and the maximum protection margin of all actual nodes along the path connecting the two points is lower than a preset protection threshold, then update the current starting point as a candidate node, increase the step size, and proceed to the next round of the loop. When a candidate node does not meet the above restrictions or the step size reaches the preset maximum step size, stop the loop. The candidate node obtained in the last round is the new node. Using the new node as the center, search all existing tree nodes within the set neighborhood. If there exists a path cost between a tree node and the new node that is less than the path cost between the parent node and the new node, then update the tree node corresponding to the minimum path cost as the parent node of the new node.

[0038] Update using the path cost between the new node and its corresponding parent node. Then update the ellipsoid sampling parameters for the next round and repeat the following steps until the new node reaches the end of the current segment; when the search of all segments is completed, each segment will output an optimal sub-path; the optimal paths of each segment are spliced ​​together end to end according to the beacon point order to obtain a complete global candidate fault path.

[0039] Regenerate random sampling points within the unit sphere Repeat this step and each subsequent step until the number of selected fault paths in the global candidate fault path set reaches the preset value.

[0040] Preferably, in step E, the calculation of the risk composite entropy includes the following steps:

[0041] Risk composite entropy of path σ , Let j be the weight coefficient of the j-th dimension. The normalized risk value for the j-th dimension; probabilistic risk dimension. , The maximum path cost among all candidate failure paths; propagation time dimension. , Let σ be the total propagation time of the path. The maximum propagation time among all candidate fault paths; cross-sectional impact dimension. , Let σ be the number of weak cross sections traversed by path σ. The maximum number of weak sections among all candidate failure paths; load loss dimension. , This represents the load loss after all circuit breakers σ trip. The maximum load loss among all candidate fault paths; disturbance intensity dimension. , Let σ be the average trajectory intensity value of all nodes on path σ. The maximum average intensity of all candidate failure paths;

[0042] ,in The direct failure propagation rate between two points. , This is the corrected direct fault propagation rate. , The gain factor for the propagation of the k-th level fault. , The initial gain factor, The growth index is k, where k is the current fault level. denoted as the directional disturbance amplification factor for the k-th propagation path.

[0043] A system for predicting interlocking trip faults based on disturbance trajectory features, used to implement the aforementioned method for predicting interlocking trip faults based on disturbance trajectory features, includes,

[0044] The data acquisition module is used to collect historical power grid operating data and extract disturbance trajectory features and path search parameters;

[0045] The initial fault propagation path generation module is used to generate an initial reference path for fault propagation using a Voronoi diagram.

[0046] The fault potential field generation module is used to establish a fault potential field that integrates disturbance characteristics and calculates the attractive and repulsive potential energy of nodes.

[0047] The candidate fault path set generation module is used to generate a candidate fault path set using an asymmetric ellipsoidal sampling optimization algorithm.

[0048] The risk composite entropy calculation module is used to correct the total propagation cost of the path and calculate the five-dimensional risk composite entropy of the path.

[0049] The prediction output module is used to sort candidate fault paths in descending order of risk composite entropy and output the fault paths according to preset requirements.

[0050] The beneficial effects of adopting the above technical solution are as follows: This invention equates power grid fault propagation path prediction to an optimal path search problem with probabilistic constraints. A weighted region graph is constructed based on the Voronoi diagram structure, and an initial reference path is obtained through shortest path search. Then, the initial reference path is discretized, and a skip check method is used to simplify it, thereby improving subsequent computational efficiency. This invention is based on Informed-RRT. *Based on the algorithm, the search domain is asymmetrically modified and perturbation features are introduced to achieve dynamic matching between the sampling distribution and the risk distribution, improving the utilization of sampling resources and the model convergence speed. A cyclical search is performed using an incremental step size to enhance search robustness. Simultaneously, this invention introduces the concept of a fault potential field into the search process, guiding and strengthening the targeting of local searches and balancing the contradiction between "global exploration" and "local refinement." The five-dimensional risk composite entropy in this invention introduces a nonlinear accumulation mechanism of entropy increase along the fault propagation path, realizing a positive closed loop of "risk quantification → strategy adaptation → search efficiency improvement," which, along with the improved Informed-RRT,... * The algorithms work together to optimize risk indicators from "post-assessment" to "pre-assessment". Attached Figure Description

[0051] Figure 1 This is the system logic flowchart of the present invention. Detailed Implementation

[0052] In the following description of the embodiments, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0053] This invention deeply integrates the time-series characteristics of dynamic disturbances in the power grid with a static path search algorithm, solving the problem that traditional cascading fault prediction relies only on static parameters such as topology and equipment probability, ignoring the dynamic propagation characteristics of disturbances.

[0054] The first step is to collect historical power grid operating data and extract disturbance trajectory features and path search parameters.

[0055] Taking the moment of the initial disturbance as the zero point, the time series data within T seconds after the fault is extracted, and the following four types of original trajectory features are extracted for each node:

[0056] The maximum rate of decrease of the node voltage amplitude after the disturbance: , Let be the voltage magnitude at node x at time t. The sampling interval is denoted as .

[0057] Power fluctuation amplitude: , Let be the total incoming power of node x at time t.

[0058] Disturbance propagation delay: , The moment when the voltage at node x first exceeds the limit. This is the time when the initial disturbance occurs.

[0059] Trajectory volatility: , for Standard deviation of time series data for The mean of time series data.

[0060] For the voltage time series of each node, scale is... The coarsening process, the coarsening sequence at the ε-th scale is: , N represents the total length of the original time series, and L represents the maximum number of layers at the largest scale. For each scale of the coarse-grained sequence, a sliding window is used to calculate the permutation entropy. w is the width of the sliding window, and m is the embedding dimension. Let be the probability of the i-th permutation pattern occurring. The permutation entropy fingerprint of node x is obtained by taking the weighted average of the permutation entropies across all scales and all sliding windows. , The weights for the ε-th scale are... This invention introduces permutation entropy to describe the degree of oscillation disorder, non-stationarity, and evolutionary complexity of perturbation trajectories, thereby effectively identifying hidden risks in the system. The perturbation trajectory entropy fingerprint, through multi-scale weighting and normalization, can unify the permutation entropy of different types of data, thereby improving the accuracy and robustness of subsequent calculations.

[0061] Next, the original trajectory features and the perturbed trajectory entropy fingerprint are used to calculate the quantization parameters for path search.

[0062] Node perturbation sensitivity coefficient: , These are the weighting coefficients.

[0063] Disturbance amplification factor from line i to line j: .

[0064] Intensity value of node trajectory: .

[0065] The second step is to construct a Voronoi diagram and generate initial reference paths.

[0066] Based on historical records, the k nodes with the highest failure rates across the entire network are selected as seed nodes, resulting in a set of seed nodes. For all non-seed nodes and all seed nodes, calculate the electrical distance one by one. For each non-seed node, assign it to the Voronoi region of the corresponding seed node according to the minimum electrical distance criterion; that is, the node belongs to the electrical influence region of the nearest seed node. The Voronoi region corresponding to the i-th seed node is defined as follows: , The electrical distance between two nodes, nodes and The method for calculating the electrical distance between them is as follows: , Let be the self-impedance of node i. Let J be the self-impedance of node j. Let be the mutual impedance between node i and node j.

[0067] This invention identifies regional boundary nodes through distance tolerance. Boundary nodes are transitional nodes between two adjacent regions and are also critical channels for fault propagation across regions. If node x satisfies the condition that there are two distinct seed points... , making Then node x is a region and The common boundary nodes are defined, where ε is the boundary determination tolerance. If two Voronoi regions have at least one common boundary node, the two regions are determined to be electrically adjacent, and a region adjacency matrix is ​​constructed. , .

[0068] We construct a weighted region graph using each Voronoi region as a vertex and the region adjacency matrix A as the connectivity standard for edges. Two adjacent regions... and (Right now The edge weights between () are represented by the crossing cost. , Let $\frac{i}{j}$ be the cost of traversing from region $i$ to region $j$. A higher value indicates that the fault propagation across regions is more difficult. Let be the average protection margin coefficient for all common boundary nodes between the two regions. Dijkstra's algorithm is used to perform a shortest path search from the starting region to the ending region, obtaining the initial reference path at the region level. Here, we only perform path expansion between adjacent regions to avoid traversing all nodes and save computational resources. Then, within each region, the node with the highest failure probability is selected as a transfer point, connecting the starting point, the transfer points in each region, and the ending point to obtain the initial reference path at the node level. .

[0069] Calculate the initial reference path The i-th segment (node) To the node Electrical length The total electrical length of the entire path is , where p is the number of nodes in the initial reference path. A fixed electrical step size is used. Uniform sampling along the path transforms the unevenly spaced original path nodes into evenly spaced scattered points. The total number of discrete points is... The cumulative electrical distance corresponding to the kth discrete point is The position of the kth discrete point is calculated by linear interpolation. , where i satisfies Then, an inflection point set is established, removing discrete points that are not inflection points. The starting node is added to the inflection point set, and the index pointer is set at the starting node's position. Starting from the last discrete point, the process iterates backward, searching for the discrete point farthest from the current pointer that meets the following conditions: the direct fault propagation probability between this discrete point and the index pointer is not less than a set threshold; and the protection margin of all intermediate actual nodes on the path connecting this discrete point and the index pointer is less than a set threshold. If a discrete point that meets the conditions is found, it is added to the inflection point set, and the index pointer is moved to the position of the most recently added discrete point. This process is repeated until the index pointer has traversed all discrete points, resulting in the final inflection point set.

[0070] The third step is to establish a fault potential field that integrates the disturbance characteristics.

[0071] Define the attractive potential energy at node x as follows:

[0072] in, The gain coefficient of the attraction potential field. Let x be the static failure probability of node x. Let x be the electrical distance from node x to the target node. This represents the critical electrical distance threshold. (Attraction) .

[0073] Define the repulsive potential energy at node x as follows: , The gain coefficient of the repulsive potential field. Let x be the electrical distance from node x to the nearest strongly protected node. This is the critical distance for repulsion. This represents the protection margin coefficient for node x. Repulsive force. .

[0074] The fourth step is to calculate the global set of candidate fault paths through path optimization.

[0075] We generate a global candidate fault path set based on a symmetric ellipsoidal sampling optimization algorithm. From the inflection point set, we select the k nodes with the highest fault probability as beacon points. The beacon points are used as dividing points to split the entire path into k+1 independent search segments. An initial symmetric ellipsoid is constructed for the i-th segment. The initial center point of the ellipsoid is The initial metric matrix is , The direct electrical distance between the two focal points. The cost of the optimal path for the current segment. , , for and The probability of direct fault propagation between them.

[0076] However, traditional symmetric ellipsoidal sampling optimization algorithms have no directional bias in their path search; the path from the start point to the end point is equivalent to the path from the end point to the start point, and the sampling is evenly distributed upstream and downstream. In contrast, power grid cascading faults are a unidirectional cascading evolution process: starting from the initial disturbance node, the fault propagates sequentially downstream, and the deeper the fault level, the greater the power flow impact and the stronger the risk amplification effect; the probability of fault triggering at downstream nodes is significantly higher than that upstream. Therefore, traditional path search processes have significant adaptation defects in power grid cascading trip fault prediction scenarios.

[0077] To address the aforementioned shortcomings, we perform asymmetric correction on the ellipsoid and introduce perturbation features. The method for correcting the ellipsoid's major axis is as follows: , The length of the semi-major axis on the starting side. The length of the semi-major axis at the endpoint. For adjustment coefficients, The fault accumulation gain at position s is normalized, with S=0 at the start point and S=1 at the end point. The center of the corrected ellipsoid is... , Corrected metric matrix .

[0078] Define the fault conductance at sampling point x , This represents the static failure probability of a node. To protect the margin factor, the path proximity factor Generate random sampling points within the unit sphere. Decompose the metric matrix Obtain the scaling transformation matrix , making Sampling points within the ellipsoid in the world coordinate system are obtained through spatial transformation. ,in This is a coordinate rotation matrix used to rotate local ellipsoid points, aligning the major axis to the direction of the line connecting the two foci, matching the orientation of the world coordinate system. It calculates the fault conductance at each sampling point within the ellipsoid. If the sampling point is correct, retain it; otherwise, delete it. Correct the sampling point orientation to obtain the guide sampling point. Where η is the guiding step size, .

[0079] Find the tree node closest to the guiding sampling point and designate it as the parent node. Using this tree node as the starting point, define the direction pointing to the guiding sampling point as the expansion direction. Starting from the current starting point, advance along the expansion direction by an initial step size, generating candidate nodes. If the direct propagation probability from the starting point to a candidate node is not lower than a set threshold, and the maximum protection margin of all actual nodes along the path connecting the two points is lower than a preset protection threshold, then update the current starting point as a candidate node, increase the step size, and proceed to the next round of iteration. Stop the iteration when a candidate node no longer meets the above restrictions or the step size reaches the preset maximum step size. The candidate node obtained in the last round is the new node. Using the new node as the center, search all existing tree nodes within a set neighborhood. If the path cost between a tree node and the new node is less than the path cost between the parent node and the new node, then update the tree node with the minimum path cost as the parent node of the new node.

[0080] Update using the path cost between the new node and its corresponding parent node. Then, update the ellipsoid sampling parameters for a new round. Because of this update, The sampling range is gradually reduced, thus gradually shrinking the sampling domain. Then, step 4 is repeated until the new node reaches the end of the current segment (the next beacon point). After all segments have been searched, each segment will output an optimal sub-path. By concatenating the optimal paths of each segment in beacon point order, a complete global candidate fault path is obtained.

[0081] Regenerate random sampling points within the unit sphere Repeat the above steps until the number of selected fault paths in the global candidate fault path set reaches the preset value.

[0082] The fifth step is to correct the total path cost and calculate the risk compound entropy.

[0083] The corrected total path cost is calculated by introducing a fault propagation rate correction process. ,in The direct failure propagation rate between two points. , This is the corrected direct fault propagation rate. , The gain factor for the propagation of the k-th level fault. , The initial gain factor, The growth index is k, where k is the current fault level. denoted as the directional disturbance amplification factor for the k-th propagation path.

[0084] For the risk composite entropy, we calculate it using five dimensions: probabilistic risk, propagation duration, cross-sectional impact, load loss, and disturbance intensity. The risk composite entropy of path σ is... , Let j be the weight coefficient of the j-th dimension. This represents the normalized risk value for the j-th dimension. (Probabilistic risk dimension) , This represents the maximum path cost among all candidate failure paths. (Propagation duration dimension) , Let σ be the total propagation time of the path. This represents the maximum propagation time among all candidate fault paths. (Cross-sectional impact dimension) , Let σ be the number of weak cross sections traversed by path σ. This represents the maximum number of weak sections among all candidate failure paths. (Load loss dimension) , This represents the load loss after all circuit breakers σ trip. This represents the maximum load loss among all candidate fault paths. Disturbance intensity dimension. , Let σ be the average trajectory intensity value of all nodes on path σ. This represents the maximum average intensity of all candidate failure paths.

[0085] The sixth step is to sort all candidate fault paths in descending order of risk composite entropy, and then, according to preset requirements, select fault paths that meet the preset requirements in sequence, starting from the fault path with the highest risk composite entropy.

[0086] To implement the above prediction method, this invention establishes a corresponding prediction system, including:

[0087] The data acquisition module is used to collect historical power grid operating data and extract disturbance trajectory features and path search parameters.

[0088] The initial fault propagation path generation module is used to generate an initial reference path for fault propagation using a Voronoi diagram.

[0089] The fault potential field generation module is used to establish a fault potential field that integrates disturbance characteristics and calculates the attractive and repulsive potential energy of nodes.

[0090] The candidate fault path set generation module is used to generate a candidate fault path set using an asymmetric ellipsoidal sampling optimization algorithm.

[0091] The risk composite entropy calculation module is used to correct the total propagation cost of the path and calculate the five-dimensional risk composite entropy of the path.

[0092] The prediction output module is used to sort candidate fault paths in descending order of risk composite entropy and output the fault paths according to preset requirements.

[0093] The prediction method of this invention was compared with traditional power grid cascading failure path search using Monte Carlo random sampling and traditional power grid cascading failure path search using an RNN prediction model through simulation experiments. The simulation scenarios were the IEEE 39-bus system and the IEEE 118-bus system, with an Intel Core Ultra 9 285K processor, 64GB of RAM, and an NVIDIA GeForce RTX 4090D graphics card. Fifty experiments were conducted for each simulation scenario, and the average results were taken. The experimental results are as follows:

[0094] Table 1 IEEE 39-node system

[0095] This invention 1.53 100 Monte Carlo random sampling 13.44 96 RNN prediction model 8.54 100

[0096] Table 2 IEEE 118-node system

[0097] This invention 3.98 100 Monte Carlo random sampling 32.57 92 RNN prediction model 13.12 98

[0098] Therefore, it is evident that this invention significantly outperforms traditional algorithms in both prediction speed and prediction accuracy.

[0099] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0100] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0101] In various embodiments, the hardware implementation of the technology can directly utilize existing smart devices, including but not limited to industrial control computers, PCs, smartphones, handheld devices, and floor-standing devices. Its input device preferably uses an on-screen keyboard, its data storage and computing modules utilize existing memory, calculators, and controllers, its internal communication modules utilize existing communication ports and protocols, and its remote communication utilizes existing GPRS networks, the World Wide Web, etc.

[0102] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0103] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0104] In the various embodiments of this invention, the functional units can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units can be implemented in hardware or as software functional units. If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in computer-readable media may be appropriately added to or subtracted from the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, computer-readable media may not include electrical carrier signals and telecommunication signals, in accordance with legislation and patent practice.

[0105] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0106] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting interlocking tripping faults based on disturbance trajectory characteristics, characterized in that... Includes the following steps: A. Collect historical power grid operating data, extract disturbance trajectory characteristics of each node, and calculate path search parameters; B. Construct a Voronoi diagram of the power grid, generate an initial reference path for fault propagation in the Voronoi diagram, and discretize and simplify the inflection points of the initial reference path. C. Establish a fault potential field with integrated disturbance characteristics, and calculate the attractive and repulsive potential energy of the nodes respectively. D. Based on the ellipsoidal sampling optimization algorithm and combined with the fault potential field, a global candidate fault path set is generated for path search. E. Correct the total propagation cost of the candidate failure path and calculate the five-dimensional risk composite entropy of the path; F. Sort all candidate fault paths in descending order of risk composite entropy. According to preset requirements, select fault paths that meet the preset requirements in sequence, starting from the fault path with the highest risk composite entropy.

2. The method for predicting interlocking tripping faults based on disturbance trajectory features according to claim 1, characterized in that: In step A, taking the moment of the initial disturbance as the zero point of time, the following four types of original trajectory features are calculated: The maximum rate of decrease of the node voltage amplitude after the disturbance: , Let be the voltage magnitude at node x at time t. The sampling interval; Power fluctuation amplitude: , Let be the total incoming power of node x at time t; Disturbance propagation delay: , The moment when the voltage at node x first exceeds the limit. The time when the initial disturbance occurs; Trajectory volatility: , for Standard deviation of time series data for The mean of time series data.

3. The method for predicting linked tripping faults based on disturbance trajectory characteristics according to claim 2, characterized in that: In step A, the voltage time series of each node is scaled to a specific value. The coarsening process, the coarsening sequence at the ε-th scale is: , N is the total length of the original time series, and L is the maximum number of layers at the largest scale. For each scale of coarse-grained sequence, a sliding window is used to calculate the permutation entropy. w is the width of the sliding window, and m is the embedding dimension. Let be the probability of the i-th permutation pattern occurring; The perturbation trajectory entropy fingerprint of node x is obtained by taking the weighted average of the permutation entropy of all scales and all sliding windows. , The weights for the ε-th scale are... ; Path search parameters include, Node perturbation sensitivity coefficient: , These are the weighting coefficients; Disturbance amplification factor from line i to line j: .

4. The method for predicting interlocking tripping faults based on disturbance trajectory characteristics according to claim 3, characterized in that: Step B, generating the initial reference path for fault propagation, includes the following steps: Based on historical records, the k nodes with the highest failure rates across the entire network are selected as seed nodes, resulting in a set of seed nodes. For all non-seed nodes and all seed nodes, calculate the electrical distance one by one. For each non-seed node, assign it to the Voronoi region of the corresponding seed node according to the minimum electrical distance criterion. The Voronoi region corresponding to the i-th seed point is defined as , The electrical distance between two nodes, nodes and The method for calculating the electrical distance between them is as follows: , Let be the self-impedance of node i. Let J be the self-impedance of node j. Let be the mutual impedance between node i and node j; If node x satisfies the condition that there are two distinct seed points , making Then node x is a region and The common boundary nodes are defined, where ε is the boundary determination tolerance; if two Voronoi regions have at least one common boundary node, the two regions are determined to be electrically adjacent, and a region adjacency matrix is ​​constructed. , ; Using each Voronoi region as a vertex and the region adjacency matrix A as the connectivity basis, a weighted region graph is constructed; two adjacent regions and (Right now The edge weights between () are represented by the crossing cost. , Let $\frac{i}{j}$ be the cost of traversing from region $i$ to region $j$. The average protection margin coefficient for all common boundary nodes of the two regions; Dijkstra's algorithm is used to search for the shortest path from the starting region to the ending region to obtain the initial reference path at the region level. Then, within each region, the node with the highest failure probability is selected as a transfer point, connecting the starting point, the transfer points in each region, and the ending region to obtain the initial reference path at the node level. .

5. The method for predicting interlocking tripping faults based on disturbance trajectory characteristics according to claim 4, characterized in that: Step B, which involves discretizing the initial reference path and simplifying inflection points, includes the following steps: Calculate the initial reference path upper node To the node electrical length The total electrical length of the entire path is Where p is the number of nodes in the initial reference path; with a fixed electrical step size Uniform sampling along the path transforms the unevenly spaced original path nodes into equally spaced scattered points; the total number of discrete points is... The cumulative electrical distance corresponding to the kth discrete point is The position of the kth discrete point is calculated by linear interpolation. , where i satisfies Then, establish an inflection point set, removing discrete points that are not inflection point positions; add the starting node to the inflection point set, and set the index pointer at the starting node position; starting from the last discrete point, traverse backwards to find the discrete point farthest from the current pointer that meets the following conditions: the direct fault propagation probability between this discrete point and the index pointer is not lower than a set threshold, and the protection margin of all intermediate actual nodes on the path connecting this discrete point and the index pointer is lower than a set threshold; if a discrete point that meets the conditions is found, add this discrete point to the inflection point set, and move the index pointer to the position of the latest discrete point added to the inflection point set; repeat the above process until the index pointer traverses all discrete points to obtain the final inflection point set.

6. The method for predicting linked tripping faults based on disturbance trajectory characteristics according to claim 5, characterized in that: In step C, establishing the fault potential field with fused perturbation characteristics includes the following steps: Define the attractive potential energy at node x as follows: in, The gain coefficient of the attraction potential field. Let x be the static failure probability of node x. Let x be the electrical distance from node x to the target node. Critical electrical distance threshold; attractive force ; Define the repulsive potential energy at node x as follows: , The gain coefficient of the repulsive potential field. Let x be the electrical distance from node x to the nearest strongly protected node. The critical distance for repulsion. The protection margin coefficient for node x; repulsive force. .

7. The method for predicting linked tripping faults based on disturbance trajectory characteristics according to claim 6, characterized in that: In step D, an asymmetric ellipsoidal sampling optimization algorithm is used to generate a global candidate fault path set. The method for establishing the asymmetric ellipsoid is as follows: Select the k nodes with the highest failure probability from the inflection point set as beacon points. The beacon points are used as dividing points to split the entire path into k+1 independent search segments; an initial symmetric ellipsoid is constructed for the i-th segment. The initial center point of the ellipsoid is The initial metric matrix is , The direct electrical distance between the two focal points. The cost of the optimal path for the current segment. , , for and The probability of direct fault propagation between them; The ellipsoid is asymmetrically corrected and perturbation features are introduced. The method for correcting the major axis of the ellipsoid is as follows: , The length of the semi-major axis on the starting side. The length of the semi-major axis at the endpoint. For adjustment coefficients, The fault accumulation gain at position s is normalized, with S=0 at the start point and S=1 at the end point. The corrected ellipsoid center is , Corrected metric matrix .

8. The method for predicting interlocking tripping faults based on disturbance trajectory features according to claim 7, characterized in that: In step D, the global candidate fault path set includes the following steps: Define the fault conductance at sampling point x , This represents the static failure probability of a node. To protect the margin factor, the path proximity factor ; Generate random sampling points within the unit sphere Decompose the metric matrix Obtain the scaling transformation matrix , making Sampling points within the ellipsoid in the world coordinate system are obtained through spatial transformation. ,in This is a coordinate rotation matrix used to rotate local ellipsoid points, aligning the major axis to the direction of the line connecting the two foci, matching the orientation of the world coordinate system; it calculates the fault conductance at each sampling point within the ellipsoid, if... If the sampling point is positive, then retain it; otherwise, delete it. Correct the sampling point orientation to obtain the guide sampling point. Where η is the guiding step size, ; Find the tree node closest to the guiding sampling point and use it as the parent node. Starting from this tree node, define the direction pointing to the guiding sampling point as the expansion direction. Starting from the current starting point, advance the initial step size along the expansion direction to generate candidate nodes. When the direct propagation probability from the starting point to the candidate node is not lower than the set threshold, and the maximum protection margin of all actual nodes along the path connecting the two points is lower than the preset protection threshold, then update the current starting point as a candidate node, increase the step size, and proceed to the next round of the loop. When a candidate node does not meet the above restrictions or the step size reaches the preset maximum step size, the loop stops; the candidate node obtained in the last round is the new node. With the new node as the center, all existing tree nodes are searched within the set neighborhood. If there is a path cost between a tree node and the new node that is less than the path cost between the parent node and the new node, the tree node corresponding to the minimum path cost is updated as the parent node of the new node. Update using the path cost between the new node and its corresponding parent node. Then update the ellipsoid sampling parameters for the next round and repeat the following steps until the new node reaches the end of the current segment; when the search of all segments is completed, each segment will output an optimal sub-path; the optimal paths of each segment are spliced ​​together end to end according to the beacon point order to obtain a complete global candidate fault path. Regenerate random sampling points within the unit sphere Repeat this step and each subsequent step until the number of selected fault paths in the global candidate fault path set reaches the preset value.

9. The method for predicting linked tripping faults based on disturbance trajectory characteristics according to claim 8, characterized in that: In step E, the calculation of the risk composite entropy includes the following steps: Risk composite entropy of path σ , Let j be the weight coefficient of the j-th dimension. The normalized risk value for the j-th dimension; probabilistic risk dimension. , The maximum path cost among all candidate failure paths; propagation time dimension. , Let σ be the total propagation time of the path. The maximum propagation time among all candidate fault paths; cross-sectional impact dimension. , Let σ be the number of weak cross sections traversed by path σ. The maximum number of weak sections among all candidate failure paths; load loss dimension. , This represents the load loss after all circuit breakers σ trip. The maximum load loss among all candidate fault paths; disturbance intensity dimension. , Let σ be the average trajectory intensity value of all nodes on path σ. The maximum average intensity of all candidate failure paths; ,in The direct failure propagation rate between two points. , This is the corrected direct fault propagation rate. , The gain factor for the propagation of the k-th level fault. , The initial gain factor, The growth index is k, where k is the current fault level. denoted as the directional disturbance amplification factor for the k-th propagation path.

10. A system for predicting interlocking trip faults based on disturbance trajectory features, used to implement the method for predicting interlocking trip faults based on disturbance trajectory features as described in any one of claims 1 to 9, characterized in that: include, The data acquisition module is used to collect historical power grid operating data and extract disturbance trajectory features and path search parameters; The initial fault propagation path generation module is used to generate an initial reference path for fault propagation using a Voronoi diagram. The fault potential field generation module is used to establish a fault potential field that integrates disturbance characteristics and calculates the attractive and repulsive potential energy of nodes. The candidate fault path set generation module is used to generate a candidate fault path set using an asymmetric ellipsoidal sampling optimization algorithm. The risk composite entropy calculation module is used to correct the total propagation cost of the path and calculate the five-dimensional risk composite entropy of the path. The prediction output module is used to sort candidate fault paths in descending order of risk composite entropy and output the fault paths according to preset requirements.

Images