A power dispatch automation system fault tracing method based on convergent cross mapping

Abstract

The embodiment of the application provides a power dispatch automation system fault tracing method based on convergent cross mapping, comprising: collecting operation data before and after a fault of a power dispatch automation system, calculating permutation entropy of components before and after the fault, and measuring abnormality degree of the components according to the permutation entropy change; calculating convergent cross mapping coefficients between the components before and after the fault, constructing a causal diagram of the power dispatch automation system in a normal state, and obtaining a global causal relationship change degree of the components according to a change of the causal link; performing community division on the causal diagram in the normal state, calculating a local causal relationship change degree of the components in a tightly coupled community range, then combining the abnormality degree of the components, the global and local causal relationship change degrees to fit a fault degree measurement index, and positioning a fault root component according to a fault degree sorting result. The technical scheme provided by the embodiment of the application can improve the accuracy of power dispatch automation system fault tracing.

Description

[Technical Field]

[0001] This invention relates to a fault tracing method for solving faults with unknown topological relationships in the field of fault location, and particularly to a fault tracing method for power dispatch automation systems based on convergent cross mapping. [Background Technology]

[0002] With the rapid development of my country's economy and society, the demand for electricity in industry and daily life has increased significantly, placing increasingly higher demands on the safe, economical, and reliable operation of the power grid. As the command center of the power grid, the power dispatch automation system plays a crucial role in ensuring its safe and stable operation. This system consists of numerous components such as servers, storage devices, and network interface cards (NICs). When a component fails, due to the close coupling between components, the fault can propagate rapidly within a short time, causing widespread abnormal alarms. In situations where the complex topology cannot be obtained, timely and accurate location of the fault source, troubleshooting, and repair of faulty equipment are essential for ensuring the safe and stable operation of the power dispatch automation system. Existing fault tracing methods mostly rely on expert experience and topology rules to infer fault propagation paths. These methods are only suitable for analyzing faults in small systems with relatively low complexity and simple logical relationships. For a power dispatch automation system, statistically analyzing all process knowledge of the system is an extremely difficult task. Furthermore, the business functions of the power dispatch automation system are constantly being updated and upgraded, and the associated rules are also constantly changing. Modifying the corresponding topology rules for each business upgrade is impractical. Data-driven fault tracing methods based on Granger causality or transitive entropy are time-consuming to model and struggle to distinguish spurious causal relationships, resulting in poor accuracy. Therefore, this paper proposes a data-driven fault tracing algorithm based on convergent cross-mapping to model causal relationships between components. This algorithm can assist maintenance personnel in quickly locating the root cause of faults from a wide range of alarm information, which is of great significance for ensuring the stable operation of the power grid. [Summary of the Invention]

[0003] In view of this, this invention proposes a fault tracing method for power dispatch automation systems based on convergent cross-mapping. In the absence of logical topological relationships, the method analyzes the changes in the operating modes and patterns of components before and after a fault based on the operating data of the power dispatch automation system, thereby effectively locating the fault source that causes widespread alarms and improving the accuracy of fault tracing.

[0004] This invention proposes a fault tracing method for power dispatch automation systems based on convergent cross-mapping, comprising the following steps:

[0005] (1) Collect operational data of the power dispatch automation system before and after a fault, calculate the permutation entropy of the components before and after the fault, and measure the degree of abnormality of the components based on the change in permutation entropy. Specifically:

[0006] Collect resource usage data for CPU, memory, disk, network, and process components of all servers in the power dispatch automation system. For each component, resource usage includes I / O read / write activity, utilization rate, collision rate, and latency. This resource usage data is then used to create a component feature sequence set {x1, x2, ..., x...}. n As input to the fault tracing method, n represents the number of sequences, and each sequence contains l observations. Phase space reconstruction is performed on each sequence, with the embedding vector dimension set to m and the delay time set to τ, to obtain the corresponding phase space matrix Y. i :

[0007]

[0008] Where, x i (k) represents the numerical value of the i-th sequence at time k, where i = 1, 2, ..., n, k = 1, 2, ..., l. Each row of the matrix represents a reconstructed component. Arranging the elements of each component in ascending order yields a set of position index sequences. Theoretically, there are m! possible index sequences. The probability of the j-th index sequence is denoted as... The permutation entropy PE of the component feature sequence can then be calculated using the following formula. i :

[0009]

[0010] Based on the above formula, the permutation entropy of the component feature sequence before and after the failure can be calculated respectively. and The exception level of the component itself, Self_Error, can then be calculated using the following formula. i :

[0011]

[0012] Where Normalize(·) represents the normalization process, Self_Error i It is an important indicator for measuring changes in the operating mode of components within a power dispatch automation system. The higher the value, the higher the degree of abnormality of the component itself.

[0013] (2) Calculate the convergence cross-mapping coefficients between components before and after the fault, construct the causal graph of the power dispatch automation system under normal conditions, and obtain the degree of change in the global causal relationship of the components based on the changes in causal links, specifically:

[0014] For the feature sequence {x1,x2,...,x} of the components of a power dispatch automation system n Any two sequences x in} i ={x i(1),x i (2),...,x i (l)} and x j ={x j (1),x j (2),...,x j (l)}, to determine whether component j drives component i to work, first, it is necessary to set the embedding dimension E and the sampling interval ε to construct the sequence x. i Embedding vector at each time step Obtain the shadow manifold matrix M i :

[0015]

[0016] Secondly, find the E+1 nearest neighbors of each point within the shadow manifold, and use... This indicates that the sequence x used for prediction is determined. j The weight w of each point k :

[0017] w k =u k / ∑u p p = 1, 2, ..., E+1

[0018] Where k = 1, 2, ..., E+1, d[·] is the Euclidean distance metric between two vectors, which allows us to calculate the estimate of the observation of component j at time t, fitted from the historical data of component i.

[0019]

[0020] Finally, based on the correlation between the estimated and true values, the convergent cross-mapping coefficient ρ, which characterizes the magnitude of the causal relationship from component j to component i, can be calculated. j→i :

[0021]

[0022] Where Cov(·) is the formula for calculating covariance, and Var(·) is the formula for calculating variance. If ρ j→i If the value is greater than the threshold Θ, a causal relationship from j to i is considered to exist. Whether a causal relationship exists from i to j can be determined using the same principle. The magnitude of the causal relationship between every two components of the power dispatch automation system under normal conditions is calculated. Using components as nodes and causal links between components as edges, a causal graph G = (V, ζ) describing the state of the power dispatch automation system can be constructed, where V is the set of component nodes and ζ is the set of edges. ΔAbs_Causality is then used to... iLet Global_Causality represent the degree of change in global causality of component i, where Global_Causality represents the absolute value of the change in in-degree and out-degree of node i after a failure occurs, and Total_Causality represents the total number of causal links in the causal graph. i It can be expressed by the following formula:

[0023]

[0024] (3) Divide the causal graph under normal conditions into communities, calculate the degree of change in local causal relationships of components within tightly coupled communities, and then fit a fault severity metric by combining the component's own anomaly severity, global and local causal relationship changes. Based on the fault severity ranking results, locate the root cause component of the fault, specifically:

[0025] First, each node in the causal graph G = (V, ζ) is taken as its initial community label. Then, for each node, its neighboring nodes are searched, and attempts are made to update the node's community label to the neighboring node's community label. The community metric Q is calculated to determine if it produces a gain. If it does, the update is retained; otherwise, the node's original community label is restored. This process is iterated until moving any node in the graph no longer improves the value of the community metric Q, where the formula for calculating the community metric Q is shown below:

[0026] Q=αD(q|| <q>)

[0027]

[0028] Where α is the total number of edges in the graph, q is the proportion of edges within a community, and its numerator is the sum of the number of internal edges in all communities. c This indicates the number of edges within the community c to which component i belongs. <q>D(q||) represents the ratio of the number of internal edges to the total number of edges in a fully connected state. <q>) represents q and <q>KL divergence between components. Through community partitioning, the community corresponding to each component can be found, thus obtaining the metric Local_Causality, which measures the degree of change in local causal relationships within the component. i :

[0029]

[0030] Among them, Community_Causality c Local_Causality is the total number of edges contained in the community c to which component i belongs. i A higher value indicates a greater likelihood of the component failing within a tightly coupled local area. Next, a final failure severity metric, Score, is fitted based on the component's own anomaly level, the degree of change in global causality, and the degree of change in local causality. i :

[0031]

[0032] The above formula can be used to calculate the degree of failure of each component. The failure degree ranking list is obtained by sorting the components by their numerical values ​​in descending order, thus obtaining the recommended list of components with root causes of failure.

[0033] In step (1) of the above method, the embedding vector dimension m is 7 and the delay time τ is 1.

[0034] In step (2) of the above method, the embedding dimension E is 3. Considering the time delay of causal transmission between components, the sampling interval ε of the shadow manifold is searched in the range of [1,3], and the value corresponding to the maximum calculated convergence cross-mapping coefficient is taken.

[0035] In step (2) of the above method, the threshold Θ of the convergence cross-mapping coefficient used to determine whether there is a causal relationship between components is set to 0.5.

[0036] The proposed fault tracing method for power dispatch automation systems improves the performance of fault tracing in power dispatch automation systems.

[0037] As can be seen from the above technical solutions, the present invention has the following beneficial effects:

[0038] In the technical solution implemented by this invention, by performing convergent cross-mapping modeling and entropy analysis on component operation data, a component failure degree index is fitted that considers the changes in the component's own pattern and the changes in its causal relationship in the local and global ranges. In the absence of logical topological relationships, the root cause of the fault that causes large-scale alarms can be analyzed based on the component failure degree ranking results, thereby improving the accuracy of fault tracing in power dispatch automation systems based on data-driven methods. [Attached Image Description]

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the fault tracing method for power dispatch automation systems based on convergent cross-mapping proposed in this embodiment of the invention.

[0041] Figure 2 This is a flowchart of the fault tracing method framework for power dispatch automation systems based on convergent cross mapping proposed in this embodiment of the invention.

Detailed Implementation Methods

[0042] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0044] This invention provides a fault tracing method for power dispatch automation systems based on convergent cross-mapping. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating the fault tracing method for power dispatch automation systems based on convergent cross-mapping proposed in this embodiment of the invention. Figure 1 As shown, the method includes the following steps:

[0045] Step 101: Collect operational data of the power dispatch automation system before and after the fault, calculate the permutation entropy of the component before and after the fault, and measure the degree of abnormality of the component itself based on the change in permutation entropy.

[0046] Specifically, resource usage data for CPU, memory, disk, network, and process components of all servers in the power dispatch automation system is collected. For each component, resource usage includes I / O read / write activity, utilization rate, collision rate, and latency. This resource usage data is then used to create a component feature sequence set {x1, x2, ..., x...}. n As input to the fault tracing method, n represents the number of sequences, and each sequence contains l observations. Phase space reconstruction is performed on each sequence, with the embedding vector dimension set to m and the delay time set to τ, where m is 7 and τ is 1, to obtain the corresponding phase space matrix Y. i :

[0047]

[0048] Where, x i (k) represents the numerical value of the i-th sequence at time k, where i = 1, 2, ..., n, k = 1, 2, ..., l. Each row of the matrix represents a reconstructed component. Arranging the elements of each component in ascending order yields a set of position index sequences. Theoretically, there are m! possible index sequences. The probability of the j-th index sequence is denoted as... The permutation entropy PE of the component feature sequence can then be calculated using the following formula. i :

[0049]

[0050] Based on the above formula, the permutation entropy of the component feature sequence before and after the failure can be calculated respectively. and The exception level of the component itself, Self_Error, can then be calculated using the following formula. i :

[0051]

[0052] Where Normalize(·) represents the normalization process, Self_Error i It is an important indicator for measuring changes in the operating mode of components within a power dispatch automation system. The higher the value, the higher the degree of abnormality of the component itself.

[0053] Step 102: Calculate the convergence cross-mapping coefficients between components before and after the fault, construct the causal graph of the power dispatch automation system under normal conditions, and obtain the degree of change in the global causal relationship of the components based on the changes in causal links.

[0054] Specifically, for the feature sequence {x1,x2,...,x} of the power dispatch automation system components... n Any two sequences x in} i ={x i (1),x i (2),...,x i (l)} and x j ={x j (1),x j (2),...,x j (l)}, to determine whether component j drives component i to work, first, set the embedding dimension E and sampling interval ε to construct sequence x. i Embedding vector at each time step Where E is 3 and τ is determined by searching within the range [1,3], the shadow manifold matrix M is obtained. i :

[0055]

[0056] Secondly, find the E+1 nearest neighbors of each point within the shadow manifold, and use... This indicates that the sequence x used for prediction is determined. j The weight w of each point k :

[0057] w k =u k / ∑u p p = 1, 2, ..., E+1

[0058] Where k = 1, 2, ..., E+1, This is a Euclidean distance metric between two vectors, which allows us to calculate the estimate of the observation of component j at time t, fitted from the historical data of component i.

[0059]

[0060] Finally, based on the correlation between the estimated and true values, the convergent cross-mapping coefficient ρ, which characterizes the magnitude of the causal relationship from component j to component i, can be calculated. j→i :

[0061]

[0062] Where Cov(·) is the covariance calculation formula, and Var(·) is the variance calculation formula. The convergence cross-mapping coefficient threshold is set to 0.5. If ρ j→i If the value is greater than the threshold Θ, a causal relationship from j to i is considered to exist. Whether a causal relationship exists from i to j can be determined using the same principle. The magnitude of the causal relationship between every two components of the power dispatch automation system under normal conditions is calculated. Using components as nodes and causal links between components as edges, a causal graph G = (V, ζ) describing the state of the power dispatch automation system can be constructed, where V is the set of component nodes and ζ is the set of edges. ΔAbs_Causality is then used to... i Let Global_Causality represent the degree of change in global causality of component i, where Global_Causality represents the absolute value of the change in in-degree and out-degree of node i after a failure occurs, and Total_Causality represents the total number of causal links in the causal graph. i It can be expressed by the following formula:

[0063]

[0064] Step 103: Divide the causal graph under normal conditions into communities, calculate the degree of change of local causal relationship of components within tightly coupled communities, and then fit the failure degree measurement index by combining the component's own abnormality degree, global and local causal relationship change degree, and locate the root cause component of the failure based on the failure degree ranking result.

[0065] Specifically, each node in the causal graph G = (V, ζ) is first used as its initial community label. Then, for each node, its neighboring nodes are searched, and attempts are made to update the node's community label to match the neighboring nodes' community labels. The community metric Q is calculated to determine if it produces a gain. If it does, the update is retained; otherwise, the node's original community label is restored. This process is iterated until moving any node in the graph no longer improves the value of the community metric Q, where the formula for calculating the community metric Q is shown below:

[0066] Q=αD(q|| <q>)

[0067]

[0068] Where α is the total number of edges in the graph, q is the proportion of edges within a community, and its numerator is the sum of the number of internal edges in all communities. c This indicates the number of edges within the community c to which component i belongs. <q>D(q||) represents the ratio of the number of internal edges to the total number of edges in a fully connected state. <q>) represents q and <q>KL divergence between components. Through community partitioning, the community corresponding to each component can be found, thus obtaining the metric Local_Causality, which measures the degree of change in local causal relationships within the component. i :

[0069]

[0070] Among them, Community_Causality c Local_Causality is the total number of edges contained in the community c to which component i belongs. i A higher value indicates a greater likelihood of the component failing within a tightly coupled local area. Next, a final failure severity metric, Score, is fitted based on the component's own anomaly level, the degree of change in global causality, and the degree of change in local causality. i :

[0071]

[0072] The above formula can be used to calculate the degree of failure of each component. The failure degree ranking list is obtained by sorting the components by their numerical values ​​in descending order, thus obtaining the recommended list of components with root causes of failure.

[0073] Figure 2 This is a flowchart of the fault tracing method for power dispatch automation systems based on convergent cross-mapping proposed in this invention: First, the permutation entropy of components before and after the fault is calculated, and the degree of abnormality of the component itself is obtained based on the difference between the two. Second, the magnitude of the causal relationship between components is analyzed based on the convergent cross-mapping method, a causal graph of the power dispatch automation system under normal conditions is constructed, and the degree of change of the global causal relationship of the component is obtained based on the change of the causal relationship. Then, the community structure in the causal graph is divided according to the strength of the causal relationship between components, and the degree of change of the local causal relationship of the component is obtained within the community. Finally, the final fault degree measurement index is fitted by combining the degree of abnormality of the component itself, the degree of change of the global causal relationship, and the degree of change of the local causal relationship, and the root cause component of the fault is located based on the fault degree ranking result.

[0074] In a specific embodiment, a power dispatch automation system dataset is used to evaluate the performance of the fault tracing method. This dataset contains 718 component feature sequence data collected by the SCADA application of the power dispatch automation system, with a duration of 30 minutes and a sampling period of 1 second. The data for the first 25 minutes was collected during normal system operation, and a fault occurred in the last 5 minutes, generating a large-scale alarm. Table 1 shows the component and feature information of the power dispatch automation system.

[0075] Table 1. Components and characteristics of power dispatch automation systems

[0076] CPU IO wait time; user usage time; utilization rate disk Number of IO reads; Number of IO writes; Bits in IO reads; Bits in IO writes network Maximum file descriptor delay; IO bytes sent; IO packets sent Memory Collision rate; Usage rate process IO read count; IO write count; memory usage percentage; CPU usage percentage

[0077] To visually demonstrate the advancement of the proposed method, Table 2 lists the fault origination results of the comparative methods and the proposed method on the aforementioned dataset. For ease of description, components are numbered from 1 to 718, and the selected data-driven comparative methods are mRank, gRank, RCA, and Fault_degree. The last column of Table 2 represents the baseline for ranking component fault severity. If the ranking results given by the fault origination method include components in the baseline, they are marked with an asterisk. As can be seen from Table 2, the proposed method can accurately identify the 78th root cause component with the highest fault severity. Furthermore, the results also show that the proposed fault origination method outperforms the other four methods, and its component fault severity ranking results are closest to the actual situation.

[0078] Table 2 Fault Origin Results of Power Dispatch Automation System

[0079] 1 26 53 26 ★78 ★78 78 2 40 6 39 40 ★81 81 3 ★25 3 ★46 11 34 25 4 96 15 57 ★25 26 52 5 46 ★81 41 ★81 ★52 21 6 ★78 67 ★25 73 ★25 46 7 50 7 61 ★58 29 19 8 63 ★78 ★78 ★19 ★12 67 9 ★81 99 54 44 ★46 12 10 33 ★25 ★81 65 44 58

[0080] To verify the effectiveness of the proposed method, this embodiment of the invention also uses three metrics—precision, recall, and normalized discounted cumulative gain (nDCG)—to compare the performance of the proposed fault tracing method. Precision and recall are commonly used metrics to measure the accuracy of fault detection, while nDCG is often used to evaluate the quality of the ranking results; with the same number of ranked items, a higher nDCG value indicates a better ranking result.

[0081] For fault tracing, the smaller the number of ranked items, the more important the corresponding nDCG value. Therefore, this invention performs fault tracing performance analysis on the ranking lists of the top 15 fault components given by various methods, and the results are shown in Table 3. The fault tracing results of the proposed method are at the best level in terms of accuracy, recall, and nDCG index. Among them, the accuracy and nDCG value are significantly improved compared with other tracing methods, indicating that the proposed method can deduce the ranking of root cause fault components in the power dispatch automation system with high accuracy.

[0082] Table 3 shows the performance metrics of each method on the given list of the top 15 faulty components.

[0083] Precision 0.2706 0.2750 0.3512 0.3650 0.5400 Recall 0.0487 0.0400 0.0500 0.2422 0.2450 nDCG 0.2840 0.2855 0.3200 0.3958 0.5533

[0084] In summary, the embodiments of the present invention have the following beneficial effects:

[0085] In the technical solution of this invention, operational data of the power dispatch automation system before and after a fault are collected, the permutation entropy of the components before and after the fault is calculated, and the degree of abnormality of the components is measured based on the change in permutation entropy; the convergence cross-mapping coefficients between components before and after the fault are calculated, a causal graph of the power dispatch automation system under normal conditions is constructed, and the degree of change in the global causal relationship of the components is obtained based on the change in causal links; the causal graph under normal conditions is divided into communities, and the degree of change in the local causal relationship of the components is calculated within the tightly coupled community; then, a fault severity metric is fitted by combining the degree of abnormality of the components themselves, the degree of change in the global and local causal relationships, and the root cause component of the fault is located based on the fault severity ranking result. Compared with other data-driven fault tracing methods, the fault tracing results given by this method are closer to the actual fault situation and can be used to assist operation and maintenance personnel in quickly troubleshooting faulty components.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.< / q> < / q> < / q> < / q> < / q> < / q> < / q> < / q>

Claims

1. A fault source tracing method for power dispatch automation systems based on convergent cross-mapping, characterized in that, The method steps include: (1) Collect operational data of the power dispatch automation system before and after a fault, calculate the permutation entropy of the components before and after the fault, and measure the degree of abnormality of the components based on the change in permutation entropy. Specifically, collect resource usage data of CPU, memory, disk, network, and process components of all servers in the power dispatch automation system. The resource usage of each component includes IO read / write status, utilization rate, collision rate, and waiting time. Build a component feature sequence set {x1, x2, ..., x...} composed of these resource usage data. n As input to the fault tracing method, n represents the number of sequences, each sequence contains l observations. Phase space reconstruction is performed on each sequence, with the embedding vector dimension set to m and the delay time set to τ, to obtain the corresponding phase space matrix Y. i : Where, x i (k) represents the numerical value of the i-th sequence at time k, where i = 1, 2, ..., n, k = 1, 2, ..., l. Each row of the matrix represents a reconstructed component. Arranging the elements of each component in ascending order yields a set of position index sequences. Theoretically, there are m! possible index sequences. The probability of the j-th index sequence is denoted as... The permutation entropy PE of the component feature sequence can then be calculated using the following formula. i : Based on the above formula, the permutation entropy of the component feature sequence before and after the failure can be calculated respectively. and The exception level of the component itself, Self_Error, can then be calculated using the following formula. i : Where Normalize(·) represents the normalization process, Self_Error i It is an important indicator for measuring changes in the operating mode of components within a power dispatch automation system. The larger the value, the higher the degree of abnormality of the component itself. (2) Calculate the convergence cross-mapping coefficients between components before and after the fault, construct the causal graph of the power dispatch automation system under normal conditions, and obtain the degree of change in the global causal relationship of the components based on the changes in causal links. Specifically, for the component feature sequence {x1,x2,...,x...} of the power dispatch automation system... n Any two sequences x in} i ={x i (1),x i (2),...,x i (l)} and x j ={x j (1),x j (2),...,x j (l)}, to determine whether component j drives component i to work, first, it is necessary to set the embedding dimension E and the sampling interval ε to construct the sequence x. i Embedding vector at each time step Obtain the shadow manifold matrix M i : Secondly, find the E+1 nearest neighbors of each point within the shadow manifold, and use... This indicates that the sequence x used for prediction is determined. j The weight w of each point k : w k =u k / ∑u p ,p=1,2,...,E+1 Where k = 1, 2, ..., E+1, d[·] is the Euclidean distance metric between two vectors, which allows us to calculate the estimate of the observation of component j at time t, fitted from the historical data of component i. Finally, based on the correlation between the estimated and true values, the convergent cross-mapping coefficient ρ, which characterizes the magnitude of the causal relationship from component j to component i, can be calculated. j→i : Where Cov(·) is the formula for calculating covariance, Var(·) is the formula for calculating variance, and if ρ j→i If the value is greater than the threshold Θ, then a causal relationship from j to i is considered to exist. Whether there is a causal relationship from i to j can be determined by the same principle. The magnitude of the causal relationship between every two components of the power dispatch automation system under normal conditions is calculated. Using components as nodes and causal links between components as edges, a causal graph G = (V, ζ) describing the state of the power dispatch automation system can be constructed, where V is the set of component nodes and ζ is the set of edges. ΔAbs_Causality is used to... i Let Global_Causality represent the degree of change in global causality of component i, where Global_Causality represents the absolute value of the change in in-degree and out-degree of node i after a failure occurs, and Total_Causality represents the total number of causal links in the causal graph. i It can be expressed by the following formula: (3) Divide the causal graph under normal conditions into communities, calculate the degree of change in local causal relationships of components within tightly coupled communities, and then fit a fault degree metric by combining the component's own abnormality degree, the degree of change in global and local causal relationships. Locate the root cause component of the fault based on the fault degree ranking results. Specifically: First, take each node in the causal graph G=(V,ζ) as its own initial community label, then find the neighboring nodes of each node, try to update the community label of the node to the community label of the neighboring nodes, calculate whether the community metric function produces a gain, if it produces a gain, retain this update, otherwise restore the original community label of the node, and iterate this process until moving any node in the graph can no longer improve the value of the community metric function. The calculation formula of the community metric function Q is as follows: Q=αD(q|| <q>)< / q> Where α is the total number of edges in the graph, q is the proportion of edges within a community, and its numerator is the sum of the number of internal edges in all communities. c This indicates the number of edges within the community c to which component i belongs. <q>D(q||) represents the ratio of the number of internal edges to the total number of edges in a fully connected state. <q>) represents q and <q>By analyzing the KL divergence between components and dividing them into communities, we can find the community corresponding to each component, and thus obtain the metric Local_Causality, which measures the degree of change in local causal relationships of the components. i :< / q> < / q> < / q> Among them, Community_Causality c Local_Causality is the total number of edges contained in the community c to which component i belongs. i The higher the value, the greater the likelihood of the component failing within a tightly coupled local area. Next, a final failure severity metric, Score, is fitted based on the component's own anomaly level, the degree of change in global causal relationships, and the degree of change in local causal relationships. i : The above formula can be used to calculate the degree of failure of each component. The failure degree can be sorted in descending order of numerical value to obtain a recommended list of components with the root cause of failure.

2. The fault tracing method for a power dispatch automation system based on convergent cross-mapping as described in claim 1, characterized in that, In step (1), the embedding vector dimension m is 7 and the delay time τ is 1.

3. The fault tracing method for a power dispatch automation system based on convergent cross-mapping as described in claim 1, characterized in that, In step (2), the embedding dimension E is 3. Considering the time delay of causal transmission between components, the sampling interval ε of the shadow manifold is searched in the range of [1,3], and the value corresponding to the maximum calculated convergence cross-mapping coefficient is taken.

4. The fault tracing method for a power dispatch automation system based on convergent cross-mapping as described in claim 1, characterized in that, In step (2), the threshold Θ of the convergence cross-mapping coefficient used to determine whether there is a causal relationship between components is set to 0.5.

Images