Power system fault tracing method and system based on data analysis

Abstract

The application discloses a power system fault tracing method and system based on data analysis, and relates to the technical field of fault detection, comprising the following steps: collecting multi-source operation data in a target power dispatching system, performing abnormal mode preprocessing, and generating local abnormal feature sequences of each subsystem; constructing an abnormal feature correlation matrix based on the local abnormal feature sequences, and based on the causal correlation analysis of the multi-source operation data, identifying matrix features to distinguish homologous abnormalities from pseudo-synchronous abnormalities; based on the identification result, performing fault boundary identification and abnormal tracing, outputting abnormal levels of each subsystem and corresponding fault source nodes; and generating fault response indication information for dispatching control according to the abnormal levels and fault source nodes. Through feature correlation and causal analysis of subsystem abnormalities, the application realizes fault tracing and response indication generation, so as to improve abnormal judgment accuracy and solve the problems of inaccurate tracing and untimely response of traditional methods.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and more specifically, to a method and system for tracing the source of power system faults based on data analysis. Background Technology

[0002] With the expansion of power dispatching systems and the increase in operational complexity, a large amount of multi-source operational data exists within these systems, including information on voltage, current, power, and frequency from dispatching centers, substations, and transmission and distribution terminals. Existing fault detection methods often rely on a single data source or anomaly identification based on local features, such as threshold judgment or feature similarity analysis. However, in actual operation, systems may experience chain disturbances, such as upstream voltage flicker triggering downstream frequency fluctuations, leading to similar fluctuation patterns in multiple subsystems at different times. In such cases, traditional feature similarity-based detection methods struggle to distinguish between co-origin anomalies and pseudo-synchronous anomalies (anomalies with different causes but similar forms), potentially leading to incorrect system-level fault assessments and impacting fault tracing and dispatch recovery decisions.

[0003] Furthermore, the differences in sampling frequencies, communication delays, and measurement noise in multi-source data make it impossible for anomaly identification methods that rely solely on signal shape or amplitude to accurately reflect the true correlation between subsystems. These factors present direct technical obstacles, including: failure to trace fault origins due to the non-uniqueness of anomaly patterns, unclear fault boundaries, and the possibility that scheduling and control commands may deviate from the actual operating state.

[0004] Therefore, there is an urgent need for a fault detection method that can make full use of multi-source operational data and comprehensively consider the correlation and causal relationship of abnormal features, so as to effectively distinguish between anomalies of the same source and pseudo-synchronization anomalies, accurately trace the source of faults, and provide reliable scheduling and control support.

[0005] The above-disclosed technical solutions have at least the following technical problems: existing fault detection methods based on multi-source data analysis do not consider the problem of pseudo-anomaly determination caused by time drift between multi-source heterogeneous data, resulting in failure of anomaly tracing. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a power system fault tracing method and system based on data analysis. By performing feature correlation and causal analysis on subsystem anomalies, fault tracing and response indication generation are achieved, thereby improving the accuracy of anomaly determination and solving the problems of inaccurate tracing and untimely response in traditional methods.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] On the one hand, the data analysis-based power system fault tracing method includes the following steps: collecting multi-source operation data from the target power dispatching system, performing abnormal mode preprocessing, and generating local abnormal feature sequences for each subsystem; constructing an abnormal feature correlation matrix based on the local abnormal feature sequences, and identifying matrix features based on causal correlation analysis of multi-source operation data to distinguish between co-source anomalies and pseudo-synchronization anomalies; based on the identification results, performing fault boundary identification and anomaly tracing, and outputting the anomaly level of each subsystem and the corresponding fault source node; and generating fault response indication information for dispatch control based on the anomaly level and fault source node.

[0009] In a preferred embodiment, the step of performing abnormal mode preprocessing to generate local abnormal feature sequences for each subsystem specifically involves: performing time-domain alignment on multi-source operating data to establish a synchronization alignment relationship; evaluating the response consistency of each source data based on the synchronization alignment relationship and screening sensitive monitoring signals; performing residual decomposition on the monitoring signals to obtain disturbance components; clustering the disturbance components based on fluctuation trends and spectral characteristics to form abnormal mode clusters; extracting the voltage offset rate, frequency drift rate, and active power fluctuation gradient of each abnormal mode cluster to construct local feature time-series vectors; and serializing and concatenating the local feature time-series vectors to form a local abnormal feature sequence.

[0010] In a preferred embodiment, the step of clustering the disturbance components based on fluctuation trends and spectral characteristics to form anomaly pattern clusters specifically involves: calculating the instantaneous fluctuation gradient and power spectral density function within a time window based on the disturbance components obtained from residual decomposition; constructing a dynamic similarity matrix based on the instantaneous fluctuation gradient and power spectral density function; performing time-series weighting processing on the dynamic similarity matrix and determining the number and boundaries of clusters through adaptive density clustering; and extracting the time-series evolution path of the disturbance signal within each cluster based on the clustering results to form anomaly pattern index.

[0011] In a preferred embodiment, the step of constructing an anomaly feature correlation matrix based on local anomaly feature sequences specifically involves: dividing the subsystem into several logical groups according to the power grid topology, control loop, or scheduling strategy; comparing the local anomaly feature vectors of each subsystem within each logical group, and generating dynamic correlation indicators based on the consistency of feature change trends and peak time synchronization; and filling the dynamic correlation indicators into the corresponding positions in the matrix to form the anomaly feature correlation matrix.

[0012] In a preferred embodiment, the causal correlation analysis based on multi-source operational data, identifying matrix features to distinguish between homogeneous anomalies and pseudo-synchronous anomalies, includes: performing mutual information analysis on the anomaly responses of subsystems in the anomaly feature correlation matrix to obtain causal strength indices; matching highly coupled subsystem pairs in the anomaly feature correlation matrix with their corresponding causal strength indices to select subsystem pairs that simultaneously possess strong response coupling and causal relationships; determining causal correlation based on the subsystem pairs to distinguish anomaly types; and marking the anomaly type of each subsystem according to the determination results.

[0013] In a preferred embodiment, the step of performing mutual information analysis on the abnormal responses of subsystems in the abnormal feature correlation matrix to obtain a causal strength index specifically involves: dividing the local abnormal feature sequences of each pair of subsystems into several time windows; estimating the probability distribution of the abnormal feature sequences within each time window; and weighting and integrating the mutual information values ​​of each time window in conjunction with the abnormal response features to obtain the causal strength index.

[0014] In a preferred embodiment, the step of matching highly coupled subsystem pairs in the anomaly feature correlation matrix with corresponding causal strength indices to screen out subsystem pairs that simultaneously possess strong response coupling and causal relationships specifically involves: extracting subsystem pairs with coupling strength higher than a preset coupling threshold from the anomaly feature correlation matrix; determining causal relationships based on the corresponding causal strength indices, combined with the anomaly triggering time sequence and response direction; and identifying the subsystem pair as a common-origin anomaly pair when the causal strength is within the threshold range and the correlation is stable across multiple scheduling cycles.

[0015] In a preferred embodiment, the step of identifying fault boundaries and tracing anomalies based on the identification results, and outputting the anomaly level of each subsystem and the corresponding fault source node, specifically involves: establishing an anomaly propagation relationship diagram between subsystems based on the classification results of co-origin anomaly pairs and pseudo-synchronous anomaly pairs; calculating the causal dominance coefficient of each node, marking nodes with a coefficient higher than a preset threshold as causal dominance nodes, and tracing downstream nodes along the causal direction to form an anomaly propagation path; determining the fault propagation boundary and classifying node anomaly levels based on the propagation length of the anomaly propagation path, the response delay between nodes, and the intensity of anomaly impact; designating nodes that are at the starting position in multiple anomaly paths or have the highest causal dominance coefficient as fault source nodes; and generating a diagnostic result set containing the anomaly level of each subsystem and the corresponding fault source node.

[0016] In a preferred embodiment, generating fault response indication information for scheduling control based on the anomaly level and fault source node specifically involves: constructing a fault response priority queue based on the anomaly level and fault source node, and evaluating feasible response strategies for each node in conjunction with the subsystem topology and available control measures; generating fault response indications for each fault source node, including the target node, control measure type, expected response time, and scope of impact; performing conflict analysis and optimization on the fault indications, adjusting the execution order, and merging or decomposing conflicting indications.

[0017] On the other hand, the power system fault tracing system based on data analysis includes the following modules: an anomaly feature sequence generation module: used to collect multi-source operating data from the target power dispatching system, perform anomaly mode preprocessing, and generate local anomaly feature sequences for each subsystem; causal analysis module: used to construct an anomaly feature correlation matrix based on the local anomaly feature sequences, and identify matrix features based on the causal correlation analysis of multi-source operating data to distinguish between homogeneous anomalies and pseudo-synchronous anomalies; fault boundary identification: used to perform fault boundary identification and anomaly tracing based on the identification results, and output the anomaly level of each subsystem and the corresponding fault source node; and fault response indication generation module: used to generate fault response indication information for dispatch control based on the anomaly level and fault source node.

[0018] The technical effects and advantages of the data analysis-based power system fault tracing method and system of this invention are as follows:

[0019] This invention collects and processes multi-source operational data from the power dispatching system to form local anomaly feature sequences for each subsystem. By combining the correlation and causal relationship analysis of anomaly features, it effectively distinguishes between anomalies originating from the same source and pseudo-synchronization anomalies. At the same time, it accurately determines the anomaly level and fault source node of each subsystem through fault boundary identification and anomaly tracing. Based on the anomaly level and fault source information, it generates fault response indications for dispatching control, thereby improving the accuracy and robustness of anomaly identification, reducing the false alarm rate, improving the accuracy of fault tracing, and providing a reliable basis for the rapid and intelligent response of the dispatching system. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the power system fault tracing method based on data analysis according to the present invention.

[0021] Figure 2 This is a schematic diagram of the power system fault tracing system based on data analysis according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 This invention presents a data analysis-based method for tracing the source of power system faults, comprising the following steps:

[0024] S1: Collect multi-source operation data from the target power dispatching system, perform abnormal mode preprocessing, and generate local abnormal feature sequences for each subsystem.

[0025] Multi-source operating data includes: voltage, power, frequency, and relay protection action information.

[0026] Local anomaly feature sequences are used to describe the dynamic response characteristics of each subsystem in the power dispatching system under local disturbances.

[0027] In this embodiment, the abnormal pattern preprocessing to generate local abnormal feature sequences for each subsystem specifically involves:

[0028] The collected multi-source operation data is aligned with time domain and operating condition labels. Based on the timestamp differences between the dispatch center, substation and power transmission and distribution terminal, a synchronization alignment matrix for multi-source data is established to eliminate data misalignment caused by different sampling frequencies and communication delays.

[0029] The coupling sensitivity between the data sources is calculated based on the synchronization alignment matrix. The coupling sensitivity is used to characterize the response consistency of different data sources within the same scheduling period, thereby identifying the set of monitoring signals most sensitive to system disturbances.

[0030] The monitoring signal set is decomposed into mode residuals, and the signal sequence is decomposed into steady-state components and disturbance components, and pseudo-anomaly components caused by measurement noise or communication jitter are removed.

[0031] Dynamic similarity clustering is performed on the disturbance components after residual decomposition, and abnormal pattern clusters are divided according to fluctuation trend and spectral characteristics to distinguish different anomaly types caused by equipment instability, communication packet loss or scheduling strategy switching.

[0032] Within each anomalous mode cluster, a local feature time-series vector of the subsystem is generated. The time-series vector consists of voltage offset rate, frequency drift rate, and active power fluctuation gradient, which is used to characterize the dynamic response characteristics of the subsystem during the anomalous triggering process.

[0033] The steps for obtaining the voltage offset rate are as follows: perform steady-state baseline correction on the voltage time series data within the abnormal mode cluster, use the average voltage value during the normal operation phase of the subsystem as the reference baseline, calculate the voltage offset magnitude relative to the baseline at each moment, and serialize the offset magnitude to characterize the changing trend of voltage stability during the abnormal period.

[0034] The steps for obtaining the frequency drift rate are as follows: perform sliding difference operation on the frequency data in the mode cluster in chronological order, calculate the frequency change amplitude and rate of change in adjacent time periods, so as to reflect the dynamic frequency modulation response capability of the system when an anomaly is triggered, and to identify the persistence and abrupt change characteristics of frequency anomalies.

[0035] The steps for obtaining the active power fluctuation gradient are as follows: perform first-order derivative fitting on the active power time series data, calculate the active power fluctuation gradient, and use it to reflect the energy response intensity of the subsystem during the abnormal process.

[0036] The local feature time-series vectors corresponding to each abnormal pattern cluster are serialized and concatenated to form a local abnormal feature sequence.

[0037] The coupling sensitivity is specifically calculated using the following formula:

[0038]

[0039] in, This represents the coupling sensitivity (the closer to 1, the stronger the coupling). , These are the i-th and j-th data source signals after being corrected by the synchronization alignment matrix, respectively. Let covariance be the variance of the two signals within the same scheduling period. , They are respectively , standard deviation This represents the time offset between the two signals. This refers to the system's scheduling cycle.

[0040] The dynamic similarity clustering of the disturbance components after residual decomposition, and the division of abnormal pattern clusters according to fluctuation trends and spectral characteristics, specifically involves:

[0041] Based on the disturbance components obtained from residual decomposition, the instantaneous fluctuation gradient and power spectral density function within the time window are calculated to characterize the response differences of each disturbance signal in the time and frequency domains.

[0042] A dynamic similarity matrix is ​​constructed based on the instantaneous fluctuation gradient and the power spectral density function. The dynamic similarity matrix is ​​used to characterize the amplitude coupling degree and frequency distribution consistency of different disturbance signals within a specific time period.

[0043] The dynamic similarity matrix is ​​subjected to time-series weighting. Based on the rate of similarity change between adjacent time windows, the response mutation signal is given a higher weight to enhance the sensitivity of distinguishing sudden anomalies during the clustering process.

[0044] An adaptive density clustering algorithm is executed based on the weighted similarity matrix to dynamically determine the number and boundaries of clusters, so as to avoid the problem of abnormal pattern mixing caused by traditional fixed threshold clustering.

[0045] The clustering results are fitted with anomaly evolution trajectories. The dominant frequency migration path and fluctuation peak distribution of the perturbation signal are extracted within each cluster to characterize the temporal evolution features of this type of anomaly.

[0046] The abnormal evolution trajectory is used as a pattern index to generate a corresponding abnormal pattern cluster identifier table, and provides a basis for the classification of abnormal types for subsequent local feature extraction steps.

[0047] The adaptive density clustering algorithm, based on the weighted similarity matrix, dynamically determines the number and boundaries of clusters. Specifically:

[0048] Based on the similarity between each signal in the weighted similarity matrix, the cumulative similarity value of each perturbation signal in its neighboring signals is calculated to characterize the degree of clustering of the signal in the similarity space.

[0049] The aggregation degree of all disturbance signals is statistically analyzed to obtain its distribution trend as the number of signals changes, and the inflection point of the aggregation degree change is identified to reflect the aggregation level of abnormal signals within the current time window.

[0050] Disturbance samples with significantly higher clustering density than adjacent signals are marked as candidate cluster centers. The number of candidate centers is controlled according to the degree of abrupt change in clustering density, increasing when abnormal fluctuations are severe and decreasing when the disturbance distribution is uniform, without the need to preset a fixed number of clusters.

[0051] For non-central signals, find the signal with the highest similarity and higher clustering and use it as the target of the signal, so that all signals converge to the corresponding central candidate along the increasing path of similarity and clustering.

[0052] Signals with a common direction of affiliation are grouped into the same cluster, and the clustering distribution within each cluster is analyzed. The locations where there are obvious jumps in the distribution are recorded in order to identify whether there are sub-patterns within the cluster that need to be split.

[0053] If there is a significant density jump within a cluster, the cluster is automatically split into two sub-clusters at the jump location; if the difference in aggregation between two adjacent clusters is small, they are merged into one cluster, thereby avoiding too many clusters or too wide cluster boundaries.

[0054] After adaptive splitting and merging, the number of stable and clearly defined clusters under the current time window is obtained, and the perturbation signal range corresponding to each cluster is determined.

[0055] The specific formula for calculating the instantaneous fluctuation gradient is as follows:

[0056]

[0057] The power spectral density function is specifically:

[0058]

[0059] in, Let be the instantaneous fluctuation gradient of the i-th time window. Let k be the disturbance signal value at the kth sampling point. It is a frequency domain signal (obtained by performing a Fourier transform on the disturbance component signal within the same time window). Let f be the power density at frequency f.

[0060] S2, construct an anomaly feature correlation matrix based on local anomaly feature sequences, and identify matrix features based on causal correlation analysis of multi-source operational data to distinguish between homologous anomalies and pseudo-synchronization anomalies;

[0061] It should be noted that the anomaly feature correlation matrix is ​​a matrix used to characterize the degree of correlation between local anomaly features of various subsystems in a target power dispatching system. Each row and column of the matrix corresponds to the local feature time-series vector of different subsystems or monitoring nodes in the system. The matrix element values ​​are obtained by calculating the distance or correlation between local anomaly feature sequences, which can reflect the spatial distribution and pattern consistency of abnormal behavior at the system level.

[0062] In this embodiment, the construction of the abnormal feature correlation matrix based on the local abnormal feature sequence specifically involves:

[0063] The subsystem is divided into several logical groups based on the power grid topology, control loop, or dispatching strategy.

[0064] Within each logical group, the local anomaly feature vectors of each subsystem are compared, and corresponding dynamic correlation indicators are generated based on the consistency of feature change trends and the synchronization of peak time.

[0065] The dynamic correlation indicators are filled into the corresponding positions in the matrix to form an abnormal feature correlation matrix.

[0066] It should be noted that the consistency of the change trend is obtained by the following method: For the local abnormal feature vector of each subsystem, its feature value change curve is plotted in chronological order; the upward / downward trends of the feature value curves between subsystems are compared, and the proportion of the time period of change in the same direction to the total time window is counted as the consistency of the change trend.

[0067] The peak time synchronization is specifically obtained as follows: identify the peak or extreme point of the local feature vector within each time window; calculate the time difference of the peak occurrence of different subsystems, and use the proportion of the number of time differences less than a preset threshold to the total number of peaks as the peak time synchronization index.

[0068] In this embodiment, the causal correlation analysis based on multi-source operational data, identifying matrix features to distinguish between homogeneous anomalies and pseudo-synchronization anomalies, specifically includes:

[0069] Based on multi-source operational data, mutual information analysis is performed on the abnormal responses of each pair of subsystems in the abnormal feature correlation matrix to obtain the causal strength index of each pair of subsystems.

[0070] The highly coupled subsystem pairs in the anomaly feature correlation matrix are matched with the corresponding causal strength index to screen out subsystem pairs that have both strong response coupling and causal relationship.

[0071] Based on the subsystem, causal correlation is determined for abnormal responses of each subsystem in order to distinguish different types of abnormalities;

[0072] The identification results are mapped back to the subsystem level to generate an anomaly type identifier table for each subsystem, providing a basis for subsequent fault tracing and scheduling control.

[0073] Based on multi-source operational data, mutual information analysis is performed on the abnormal responses of each pair of subsystems in the abnormal feature correlation matrix to obtain the causal strength index of each pair of subsystems, specifically:

[0074] The local anomaly feature sequence of each pair of subsystems is divided into several time windows to capture the local dynamic characteristics of the anomaly response.

[0075] For the anomalous feature sequences within each time window, the probability distribution function is obtained through kernel density estimation;

[0076] Calculate the mutual information value of each pair of subsystems within the corresponding time window to quantify the statistical dependency of their anomalous responses;

[0077] The mutual information values ​​of each time window are weighted and integrated, and the contribution of abnormal response amplitude or disturbance peak value is combined to obtain the overall causal strength index.

[0078] The mutual information value is specifically:

[0079]

[0080] in, For mutual information value, abnormal feature sequence and Simultaneous value acquisition within the time window and The joint probability, For subsystem The abnormal characteristics in the value The probability, For subsystem The abnormal characteristics in the value The probability of.

[0081] The process of matching highly coupled subsystem pairs in the anomaly feature correlation matrix with their corresponding causal strength indices to filter out subsystem pairs that simultaneously possess strong response coupling and causal relationships specifically involves:

[0082] Based on the coupling strength distribution of each subsystem pair in the anomaly feature correlation matrix, extract the set of subsystem pairs that are higher than the preset coupling threshold;

[0083] In the set of subsystem pairs, the corresponding causal strength index is retrieved, and the consistency of the time sequence of abnormal triggering and the response direction between subsystems is used to determine whether the pair of subsystems meets the causal association condition.

[0084] When the causal strength index of a pair of subsystems is within the threshold range and remains stably correlated over multiple scheduling cycles, the pair of subsystems is determined to have a causal relationship.

[0085] This allows us to identify anomalous pairs that exhibit both strong coupling and causal dependence, providing a basis for subsequent fault tracing and boundary identification.

[0086] The determination of causal relationships for abnormal responses of each subsystem based on subsystems specifically involves:

[0087] Subsystem pairs that simultaneously exhibit high coupling and high causal strength are labeled as homologous anomalies; subsystem pairs that exhibit high coupling but low causal strength are labeled as pseudo-synchronization anomalies.

[0088] S3, based on the identification results, performs fault boundary identification and anomaly tracing, and outputs the anomaly level of each subsystem and the corresponding fault source node;

[0089] In this embodiment, the step of identifying fault boundaries and tracing anomalies based on the identification results, and outputting the anomaly level of each subsystem and the corresponding fault source node, specifically involves:

[0090] Based on the classification of homologous anomaly pairs and pseudo-synchronous anomaly pairs, an anomaly propagation relationship diagram between subsystems is established;

[0091] In the propagation relationship diagram, the causal strength index between any pair of nodes is calculated, and based on the directionality and asymmetry of the causal strength, the causal dominance coefficient of the node is defined. Its value is the ratio of the causal output strength to the causal input strength of the node, which is used to characterize the dominant role of the node in the propagation of anomalies.

[0092] Nodes with a causal dominance coefficient greater than a preset threshold are marked as causal dominant nodes, and downstream nodes are traced along the causal propagation direction starting from these nodes to form an abnormal propagation path.

[0093] Calculate the propagation length, inter-node response delay, and anomaly impact intensity for each anomaly propagation path, and determine the fault propagation boundary based on the propagation termination condition (impact intensity below a threshold or causal chain breakage).

[0094] Based on the node's hierarchical position in the propagation path, cumulative response delay, and the intensity of the impact, the node's anomaly level is calculated comprehensively. The anomaly levels, from high to low, correspond to: fault source node, directly affected node, and indirectly affected node.

[0095] When a node is at the starting position or has the highest causal dominance coefficient in multiple abnormal paths, the node is identified as the fault source node.

[0096] Finally, a diagnostic result set containing the anomaly level of each subsystem and the corresponding fault source node is generated, which is used for fault response decision-making of the scheduling and control system.

[0097] The steps for calculating the propagation length, inter-node response delay, and anomaly impact intensity for each anomaly propagation path are as follows:

[0098] For the identified abnormal propagation paths, starting from the fault source node, the trigger time records of each subsystem on the path are obtained sequentially according to the causal triggering order;

[0099] Based on the triggering order of the subsystems in the abnormal event, the number of nodes appearing in the path is counted, and this number is used as the propagation length of the abnormal propagation path to characterize the range of the abnormal spread across subsystems.

[0100] Between adjacent subsystems, the response delay between nodes is determined based on the difference in their trigger times, and this delay is used to characterize the speed characteristics of anomaly propagation along the path.

[0101] The characteristic quantities reflecting the abnormal amplitude in the local anomaly feature sequence of each subsystem are extracted, including voltage offset amplitude, frequency drift rate change and active power fluctuation gradient. Between adjacent subsystems, the amplification or attenuation of the abnormal amplitude between nodes is determined according to the degree of change of the characteristic quantities, and the degree of change is used as the anomaly transmission strength between nodes. The anomaly transmission strength between nodes is cumulatively evaluated along the anomaly propagation path to reflect the overall gain or loss of the anomaly in the cross-subsystem propagation process, thereby obtaining the anomaly influence strength of the propagation path.

[0102] The anomaly level of the node is specifically as follows:

[0103]

[0104] in, The anomaly level of the node. , , The weighting coefficients are determined empirically based on system type (e.g., power dispatching, distribution terminals, etc.). The influence intensity of node i (obtained from the amplitude of the anomalous signal). Let be the response delay of node i (representing the average response delay of this node relative to the fault source node. The larger the delay, the longer the propagation chain and the weaker the impact). The depth of node i (reflects the node's depth in the propagation path (the source node is at level 0). The deeper the level, the more likely it is to be a passively affected node). , , These are the maximum values ​​of the corresponding indicators in the path, used for normalization.

[0105] S4 generates fault response indication information for scheduling control based on the anomaly level and the fault source node.

[0106] In this embodiment, the step of generating fault response indication information for scheduling control based on the anomaly level and the fault source node specifically includes:

[0107] A fault response priority queue is constructed based on the anomaly level and the fault source node, and the feasible response strategy for each node is evaluated in combination with the subsystem topology and available control measures.

[0108] A fault response indication is generated for each fault source node, the indication including the target node, the type of recommended control measures, the expected response time, and the scope of impact;

[0109] Perform conflict analysis and optimization on candidate instructions, adjust the execution order, and merge or decompose conflicting instructions.

[0110] The fault response indication further includes:

[0111] Calculate the confidence level of each instruction and classify the instructions into two categories based on the confidence level: automatic execution and manual confirmation.

[0112] The final instruction is formatted into a command that can be issued according to the scheduling system interface specification, and corresponding monitoring and rollback rules are generated.

[0113] During instruction execution, rollback or upgrade measures are triggered based on monitoring data feedback, and execution results and diagnostic information are recorded for subsequent analysis and optimization.

[0114] The calculation of the confidence level for each indication is specifically as follows:

[0115]

[0116] in, For confidence level, It is an abnormal level. The causal dominance coefficient This represents the maximum value among all subsystem causality dominance coefficients in the current anomalous event. To ensure path coverage, , , This is an adjustment coefficient, set based on system experience or scheduling strategy.

[0117] The response path coverage rate is obtained through the following steps: expanding the abnormal propagation path of the fault source node obtained based on causal dominance analysis to obtain a propagation path set that includes each subsystem within the influence range of the fault source; counting the number of subsystems covered by the current fault response indication information in the propagation path set, and comparing it with the total number of subsystems in the propagation path set to determine the coverage degree of the current fault response indication information on the abnormal propagation range; using the coverage degree as the response path coverage rate to characterize the effectiveness of scheduling instructions in blocking the abnormal propagation chain.

[0118] Example 2, Figure 2 The present invention provides a power system fault tracing system based on data analysis, comprising the following modules:

[0119] Anomaly feature sequence generation module: used to collect multi-source operation data from the target power dispatching system, perform anomaly mode preprocessing, and generate local anomaly feature sequences for each subsystem;

[0120] Causal analysis module: used to construct an anomaly feature correlation matrix based on local anomaly feature sequences, and to identify matrix features based on causal correlation analysis of multi-source operational data to distinguish between homologous anomalies and pseudo-synchronization anomalies;

[0121] Fault boundary identification: Based on the identification results, it is used to identify fault boundaries and trace the source of anomalies, and output the anomaly level of each subsystem and the corresponding fault source node;

[0122] Fault Response Indicator Generation Module: Used to generate fault response indication information for scheduling and control based on the anomaly level and the fault source node.

[0123] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0124] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0125] Those skilled in the art will recognize that the modules 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 implementation should not be considered beyond the scope of this application.

[0126] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0128] In conclusion, 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 protection scope of the present invention.

Claims

1. A method for power system fault tracing based on data analysis, characterized in that, The method comprises the following steps: Collecting multi-source operation data in a target power dispatching system, performing abnormal mode preprocessing, and generating local abnormal feature sequences of each subsystem; Based on the local abnormal feature sequences, an abnormal feature correlation matrix is constructed, and based on the causal correlation analysis of the multi-source operation data, the matrix features are identified to distinguish homogenous abnormalities and pseudo-synchronous abnormalities, including: mutual information analysis of the abnormal feature correlation matrix for abnormal response of the subsystem pairs to obtain causal strength indicators; matching the high-coupling subsystem pairs in the abnormal feature correlation matrix with the corresponding causal strength indicators to screen out subsystem pairs that have strong response coupling and causal relationship at the same time; performing causal correlation determination based on the subsystem pairs to distinguish abnormal types; and marking the abnormal types of each subsystem according to the determination result; The matching of the high-coupling subsystem pairs in the abnormal feature correlation matrix with the corresponding causal strength indicators to screen out the subsystem pairs that have strong response coupling and causal relationship at the same time is specifically: extracting the subsystem pairs with coupling strength higher than a preset coupling threshold from the abnormal feature correlation matrix; judging the causal correlation based on the corresponding causal strength indicators, combined with the abnormal trigger time sequence and the response direction; when the causal strength is in the threshold interval and the stable correlation is maintained within multiple dispatching cycles, the subsystem pair is determined as a homogenous abnormal pair; the subsystem pairs with high coupling but low causal strength are marked as pseudo-synchronous abnormalities; Based on the identification result, fault boundary identification and abnormal tracing are performed, and the abnormal level and the corresponding fault source node of each subsystem are output; According to the abnormal level and the fault source node, fault response indication information for dispatching control is generated.

2. The data analytics based power system fault tracing method as claimed in claim 1 wherein, The abnormal mode preprocessing and the generation of the local abnormal feature sequences of each subsystem are specifically: Time domain alignment of the multi-source operation data is performed to establish a synchronous alignment relationship; Based on the synchronous alignment relationship, the response consistency of each source data is evaluated, and sensitive monitoring signals are screened; Residual decomposition is performed on the monitoring signals to obtain disturbance components; Based on the fluctuation trend and the spectral characteristics, the disturbance components are clustered to form abnormal mode clusters; Voltage deviation rate, frequency drift rate and active power fluctuation gradient of each abnormal mode cluster are extracted to construct a local feature time sequence vector; The local feature time sequence vector is serialized and spliced to form a local abnormal feature sequence.

3. The data analytics based power system fault origination method as claimed in claim 2, wherein, The clustering of the disturbance components based on the fluctuation trend and the spectral characteristics to form abnormal mode clusters is specifically: Based on the disturbance components obtained by residual decomposition, the instantaneous fluctuation gradient and the power spectral density function in the time window are calculated; A dynamic similarity matrix is constructed according to the instantaneous fluctuation gradient and the power spectral density function; The dynamic similarity matrix is subjected to time sequence weighting processing, and the number and boundary of the clustering clusters are determined through adaptive density clustering; According to the clustering result, the time sequence evolution path of the disturbance signal in each cluster is extracted to form an abnormal mode index.

4. The data analytics based power system fault origination method as claimed in claim 1, wherein, The construction of the abnormal feature correlation matrix based on the local abnormal feature sequences is specifically: According to the grid topology, control loop or dispatching strategy, the subsystems are divided into several logical groups; In each logical group, the local abnormal feature vectors of each subsystem are compared, and dynamic correlation indicators are generated according to the feature change consistency and peak time synchronization; The dynamic correlation index is filled in the corresponding position of the matrix to form an abnormal feature correlation matrix.

5. The data analytics based power system fault origination method as claimed in claim 1, wherein, The mutual information of the abnormal response in the abnormal feature correlation matrix is analyzed to obtain a causality strength index, specifically: The local abnormal feature sequence of each pair of subsystems is divided into several time windows. In each time window, the probability distribution of the abnormal feature sequence is estimated. The mutual information value of each time window is weighted and integrated in combination with the abnormal response features to obtain the causality strength index.

6. The data analytics based power system fault origination method as claimed in claim 1, wherein, Based on the identification result, fault boundary identification and abnormal tracing are performed, and the abnormal level of each subsystem and the corresponding fault source node are output, specifically: According to the division results of the homologous abnormal pair and the pseudo-synchronous abnormal pair, an abnormal propagation relationship diagram between subsystems is established. The causality dominance coefficient of each node is calculated, the nodes higher than the preset threshold are marked as causality dominant nodes, and the downstream nodes are tracked along the causality direction to form an abnormal propagation path. According to the propagation length of the abnormal propagation path, the response time delay between nodes, and the abnormal influence strength, the fault propagation boundary is determined and the node abnormal level is divided. The nodes that are in the starting position or have the highest causality dominance coefficient in multiple abnormal paths are taken as fault source nodes. A diagnosis result set containing the abnormal level of each subsystem and the corresponding fault source node is generated.

7. The data analytics based power system fault origination method as claimed in claim 1, wherein, According to the abnormal level and the fault source node, fault response indication information for scheduling control is generated, specifically: Based on the abnormal level and the fault source node, a fault response priority queue is constructed, and the feasible response strategy of each node is evaluated in combination with the subsystem topology and available control measures; For each fault source node, a fault response indication including the target node, the control measure type, the expected response time, and the influence range is generated; The fault indication is analyzed and optimized for conflict, the execution order is adjusted, and the conflicting indications are merged or decomposed.

8. A system using the data analytics based power system fault tracing method as claimed in any one of claims 1 to 7, characterized in that, The following modules are included: Abnormal feature sequence generation module: used for collecting multi-source operation data in the target power dispatching system, performing abnormal mode preprocessing, and generating local abnormal feature sequences of each subsystem; Causality analysis module: used for constructing an abnormal feature correlation matrix based on the local abnormal feature sequence, and identifying the matrix features to distinguish homologous abnormalities and pseudo-synchronous abnormalities based on the causality correlation analysis of multi-source operation data; Fault boundary identification: used for fault boundary identification and abnormal tracing based on the identification result, outputting the abnormal level of each subsystem and the corresponding fault source node; Fault response indication generation module: used for generating fault response indication information for scheduling control according to the abnormal level and the fault source node.

Images