A power grid load fault identification method and system

Abstract

The application discloses a kind of power grid load fault identification method and system, comprising: obtaining the node operating quantity measurement data of each load node in target power grid, node state data and network topology data, and generate standardized operation sequence;According to the standardized operation sequence, construct operating state feature vector and normal disturbance feature vector, generate normal disturbance benchmark;Normal disturbance decoupling and fault residual enhancement are carried out based on normal disturbance benchmark, obtain fault saliency index and determine initial abnormal load node set;Fault identification model is inputed with fault residual enhancement result, and fault type identification result and fault confidence are obtained;Topological-timing consistency positioning calculation is carried out in combination with fault saliency index, fault confidence and network topology data, and target fault load node and fault influence range are obtained;And propagation consistency check is carried out, and fault identification result is generated.The application can improve the accuracy of load fault identification and the real-time performance of positioning under complex conditions.

Description

Technical Field

[0001] This invention relates to the field of power grid load fault identification technology, and in particular to a power grid load fault identification method and system. Background Technology

[0002] Power grid load fault identification is a crucial technical aspect of ensuring the safe and stable operation of the power grid. With the increasing variety of distribution network load types, continuous changes in operating states, and the expanding scale of electricity consumption data, load-side faults have become a significant factor affecting power supply reliability, power quality, and fault handling efficiency. Real-time monitoring, anomaly identification, and fault location of power grid load operating states enable the timely detection of potentially faulty loads, shortening fault investigation time, reducing the risk of fault propagation, and providing accurate data support for dispatch control, operation and maintenance, and load management. Therefore, improving the accuracy of load fault identification and the real-time performance of fault location in complex operating environments has become a key technical challenge in intelligent power grid operation and maintenance.

[0003] Existing power grid load fault identification methods typically extract features from load operation measurement data, status data, or historical sample data, and determine abnormal load states through threshold judgment, rule matching, or fault identification models. However, in actual power grid operation, load fluctuations, disturbance superposition, operating state switching, and changes in sampled data quality can couple fault features with normal operating disturbances, resulting in insufficient load fault feature identification. Fault identification results are easily affected by normal load changes, leading to misjudgments or omissions, making it difficult to meet the requirements for high-accuracy fault identification under complex operating conditions. Furthermore, when facing power grid scenarios with multiple nodes, multiple types of loads, and dynamically changing topology, existing methods often still rely on centralized data processing or fixed discrimination logic, failing to adequately utilize the topological relationships between load nodes and the timing of anomalies. When multiple load nodes simultaneously exhibit abnormal responses or fault effects propagate along the network topology, existing technologies struggle to distinguish between the truly faulty load node and affected associated nodes in a timely manner, resulting in insufficient real-time fault identification and location, impacting fault handling efficiency and power grid operational safety. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention provides a method and system for identifying power grid load faults.

[0005] On one hand, the present invention provides a method for identifying power grid load faults, the method comprising the following steps:

[0006] Step S1: Obtain node operation measurement data, node status data and network topology data of each load node in the target power grid, and perform synchronization processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node.

[0007] Step S2: Construct the operating status feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples;

[0008] Step S3: Based on the operating state feature vector, normal disturbance feature vector, and normal disturbance benchmark, perform normal disturbance decoupling and fault residual enhancement on each load node to obtain the fault saliency index of each load node, and determine the initial abnormal load node set based on the fault saliency index.

[0009] Step S4: Input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence for each abnormal load node.

[0010] Step S5: Based on the fault saliency index, fault confidence and network topology data, perform topology-time consistency location calculation on the initial abnormal load node set to obtain the target fault load node and the fault impact range.

[0011] Step S6: Taking the target faulty load node as the starting node, perform propagation consistency verification on the associated load nodes within the fault influence range along the fault propagation direction in the network topology, generate fault identification results based on the propagation consistency verification results, and write back the fault identification results after on-site confirmation to the sample library to update the normal disturbance benchmark and fault identification model.

[0012] Furthermore, step S1 specifically includes: using load nodes in the network topology as data organization units, acquiring node operation measurement data and node status data of each load node according to a unified time base, and establishing a mapping relationship between load node identifier, sampling time, and topology location; performing time-scale correction and sequence alignment on node operation measurement data and node status data of the same load node formed under different acquisition sources to obtain node time-series data that matches the network topology data; performing quality verification on the node time-series data based on operation boundary constraints, time-series continuity constraints, and topology association constraints, identifying and eliminating abnormal sampling records that are inconsistent with the power grid operation mechanism; performing missing data recovery, dimension normalization, and topology index binding on the quality-verified node time-series data to generate a standardized operation sequence that uniquely corresponds to the load node, sampling time, and topology location.

[0013] Furthermore, S201, based on the historical normal operation samples of each load node, extract the normal disturbance characteristics of the corresponding load node under different operating states;

[0014] S202. Calculate the normal disturbance reference for the corresponding load node based on the normal disturbance characteristics. The normal disturbance reference includes the normal disturbance coupling matrix, the normal residual mean vector, and the normal residual covariance matrix.

[0015] S203. Based on the operating state corresponding to the current sampling time, retrieve the normal disturbance reference that matches the current operating state from the normal disturbance reference;

[0016] S204. When the fault-free operation data confirmed on-site meets the update conditions, the fault-free operation data is incorporated into the historical normal operation sample to correct the normal disturbance reference of the corresponding load node.

[0017] Furthermore, step S3 includes the following steps:

[0018] S301, according to the... Each load node at the sampling time The standardized operating sequence is used to construct the operating state feature vector and the normal perturbation feature vector:

[0019]

[0020] in, Indicates the first Each load node at the sampling time The feature vector of the running state; Indicates the first Each load node at the sampling time The A standardized operational status characteristic, ; Dimensions representing operational status characteristics; Indicates the first Each load node at the sampling time The normal perturbation feature vector; Indicates the first Each load node at the sampling time The A normal disturbance characteristic ; Dimensions representing the characteristics of normal perturbations;

[0021] S302, according to the... Calculate the normal disturbance coupling matrix using historical normal operation samples of each load node:

[0022]

[0023] in, Indicates the first Normal disturbance coupling matrix of each load node; Indicates the first The operational status feature matrix of each load node in the historical normal operation sample; Indicates the first Normal disturbance feature matrix of each load node in historical normal operation samples; This indicates the sampling time corresponding to the historical normal operation sample; This indicates the number of historically normal operating samples; Represents the regularization coefficient; Represents the identity matrix;

[0024] S303. Decouple the current operating state feature vector from the normal disturbance coupling matrix to obtain the fault residual vector:

[0025]

[0026] in, Indicates the first Each load node at the sampling time The fault residual vector is used to characterize the fault-related features retained after removing the effects of normal load disturbances from the current operating state features;

[0027] S304. Enhance the fault residual vector based on the characteristic reliability to obtain the enhanced fault residual vector:

[0028]

[0029] in, Indicates the first Each load node at the sampling time The fault residual enhancement vector; Indicates the first Each load node at the sampling time The feature reliability weight matrix; Indicates the first Each load node before the current sampling time The normal residual mean vector within each sampling period; Indicates the length of the sliding statistics window;

[0030] S305. Calculate the fault significance index based on the fault residual enhancement vector:

[0031]

[0032] Indicates the first Each load node at the sampling time Fault significance index; Indicates the first Each load node before the current sampling time Normal residual covariance matrix within each sampling period; Indicates the first Each load node at the sampling time Data quality penalty coefficient; when When the value is greater than the adaptive saliency threshold of the corresponding load node, the th node will be... Each load node is added to the initial set of abnormal load nodes.

[0033] Furthermore, step S4 includes:

[0034] S401. Construct an anomaly identification input vector based on the fault residual enhancement vector, the fault significance index, and the time-series change in fault significance.

[0035] S402. Input the anomaly identification input vector into the fault identification model and output the probability distribution of the fault type corresponding to the abnormal load node;

[0036] S403. Determine the fault type corresponding to the highest probability in the fault type probability distribution as the fault type identification result, and determine the highest probability as the fault confidence level;

[0037] S404. When the fault confidence level is lower than the preset confidence threshold, the abnormal state of the abnormal load node is retained, and the fault identification result is corrected in step S5 by combining the topology-time consistency location calculation result.

[0038] Furthermore, step S5 includes the following steps:

[0039] S501. Construct a set of candidate fault nodes based on the fault significance index and fault confidence.

[0040] S502. Calculate the topology-timing deviation of candidate node j in the candidate fault node set;

[0041] S503. Calculate the fault location score of candidate node j based on the topology-time series deviation, fault significance index and fault confidence.

[0042] S504. Determine the target faulty load node based on the fault location score, and generate the fault impact range.

[0043] Furthermore, the anomaly weight is determined based on the fault significance index and fault confidence of the candidate node.

[0044] Furthermore, step S6 includes:

[0045] S601. Link and store the target faulty load node, fault type identification result, fault confidence, fault impact range and on-site handling result to generate a closed-loop fault sample.

[0046] S602. When the on-site handling results confirm that the fault identification results are correct, the corresponding fault residual enhancement vector, fault significance index, fault type identification results and target fault load node are added to the fault sample library.

[0047] S603. When the on-site handling results confirm that there is a deviation in the fault identification results, the corresponding fault residual enhancement vector, actual fault type and actual fault load node are added to the correction sample library.

[0048] S604. According to the preset update cycle, update the fault identification model based on the fault sample library and the correction sample library, and update the normal disturbance benchmark based on the fault-free operation data confirmed on site.

[0049] Furthermore, before generating the fault identification result, step S6 also includes:

[0050] Starting with the target faulty load node, the propagation consistency of associated load nodes within the fault's influence range is verified along the fault propagation direction in the network topology. If the abnormal starting order of the associated load nodes is inconsistent with the fault propagation direction in the network topology, or if the fault salience of the associated load nodes does not show a decreasing trend from the target faulty load node outwards, the location reliability of the target faulty load node is downgraded, and step S5 is called again to perform topology-temporal consistency location calculation. If the abnormal starting order of the associated load nodes is consistent with the fault propagation direction in the network topology, and the fault salience of the associated load nodes shows a decreasing trend from the target faulty load node outwards, the target faulty load node is confirmed as the final faulty load node.

[0051] On the other hand, the present invention provides a power grid load fault identification system applied to the above-mentioned method, the system comprising:

[0052] The acquisition module is used to acquire node operation measurement data, node status data and network topology data of each load node in the target power grid, and to perform synchronous processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node.

[0053] The benchmark module is used to construct the operating state feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples.

[0054] The decoupling module is used to perform normal disturbance decoupling and fault residual enhancement on each load node according to the operating state feature vector, normal disturbance feature vector and normal disturbance benchmark, to obtain the fault saliency index of each load node, and to determine the initial abnormal load node set according to the fault saliency index.

[0055] The identification module is used to input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence levels corresponding to each abnormal load node.

[0056] The location module is used to perform topology-time consistency location calculation on the initial abnormal load node set based on the fault saliency index, fault confidence and network topology data, so as to obtain the target fault load node and the fault impact range.

[0057] The verification module is used to perform propagation consistency verification on associated load nodes within the fault influence range along the fault propagation direction in the network topology, starting from the target fault load node. Based on the propagation consistency verification result, it generates a fault identification result and writes the fault identification result confirmed on-site back to the sample library to update the normal disturbance benchmark and fault identification model.

[0058] Compared with the prior art, the present invention has the following advantages: On the one hand, by constructing operating state feature vectors, normal disturbance feature vectors, and normal disturbance benchmarks, and using the normal disturbance coupling matrix to decouple the current operating state features from normal disturbances, the present invention can reduce the interference of normal load fluctuations, operating state switching, and data quality changes on fault features from the load operating state. At the same time, by using fault residual enhancement vectors and fault saliency indicators to quantify the decoupled fault-related features, the weak fault features that were originally masked by normal operating disturbances are highlighted, thereby improving the identification of load fault features under complex operating conditions, reducing the probability of misjudgment and missed judgment, and improving the accuracy and stability of fault identification. On the other hand, after obtaining the initial set of abnormal load nodes, this invention performs topology-time consistency location calculation by combining fault saliency indicators, fault confidence, and network topology data. It can distinguish between the real faulty load nodes and the affected associated load nodes based on the anomaly start time, topological electrical distance, and fault propagation relationship. Furthermore, it uses propagation consistency verification to determine whether the anomaly start order and fault saliency decay trend of the associated load nodes conform to the fault propagation law, and confirms or corrects the reliability of the location results. This improves the accuracy and real-time performance of determining the target faulty load node, reduces the scope of fault impact, and enhances the efficiency of power grid load fault handling. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 A flowchart of a power grid load fault identification method provided in an embodiment of the present invention;

[0061] Figure 2 A flowchart of step S3 of a power grid load fault identification method provided in an embodiment of the present invention;

[0062] Figure 3 This is a flowchart of step S5 of a power grid load fault identification method provided in an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely one unit of embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0064] As one embodiment of the present invention, such as Figure 1 As shown, this embodiment provides a method for identifying power grid load faults, the method including the following steps:

[0065] Step S1: Obtain node operation measurement data, node status data and network topology data of each load node in the target power grid, and perform synchronization processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node.

[0066] Step S2: Construct the operating status feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples;

[0067] Step S3: Based on the operating state feature vector, normal disturbance feature vector, and normal disturbance benchmark, perform normal disturbance decoupling and fault residual enhancement on each load node to obtain the fault saliency index of each load node, and determine the initial abnormal load node set based on the fault saliency index.

[0068] Step S4: Input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence for each abnormal load node.

[0069] Step S5: Based on the fault saliency index, fault confidence and network topology data, perform topology-time consistency location calculation on the initial abnormal load node set to obtain the target fault load node and the fault impact range.

[0070] Step S6: Taking the target faulty load node as the starting node, perform propagation consistency verification on the associated load nodes within the fault influence range along the fault propagation direction in the network topology, generate fault identification results based on the propagation consistency verification results, and write back the fault identification results after on-site confirmation to the sample library to update the normal disturbance benchmark and fault identification model.

[0071] Furthermore, step S1 specifically includes: using load nodes in the network topology as data organization units, acquiring node operation measurement data and node status data of each load node according to a unified time base, and establishing a mapping relationship between load node identifier, sampling time, and topology location; performing time-scale correction and sequence alignment on node operation measurement data and node status data of the same load node formed under different acquisition sources to obtain node time-series data that matches the network topology data; performing quality verification on the node time-series data based on operation boundary constraints, time-series continuity constraints, and topology association constraints, identifying and eliminating abnormal sampling records that are inconsistent with the power grid operation mechanism; performing missing data recovery, dimension normalization, and topology index binding on the quality-verified node time-series data to generate a standardized operation sequence that uniquely corresponds to the load node, sampling time, and topology location.

[0072] Specifically, by uniformly binding node operation measurement data, node status data, and network topology data to the same load node and the same sampling time, time scale deviation and topology index deviation between multi-source data can be eliminated; by performing quality verification through operation boundary constraints, time sequence continuity constraints, and topology association constraints, data anomalies that do not conform to the power grid operation mechanism can be eliminated, reducing misidentification caused by data acquisition errors; by generating standardized operation sequences, a unified and reliable data foundation can be provided for subsequent normal disturbance benchmark construction, fault residual enhancement, and topology-time sequence consistency positioning.

[0073] Furthermore, S201, based on the historical normal operation samples of each load node, extract the normal disturbance characteristics of the corresponding load node under different operating states;

[0074] S202. Calculate the normal disturbance reference for the corresponding load node based on the normal disturbance characteristics. The normal disturbance reference includes the normal disturbance coupling matrix, the normal residual mean vector, and the normal residual covariance matrix.

[0075] S203. Based on the operating state corresponding to the current sampling time, retrieve the normal disturbance reference that matches the current operating state from the normal disturbance reference;

[0076] S204. When the fault-free operation data confirmed on-site meets the update conditions, the fault-free operation data is incorporated into the historical normal operation sample to correct the normal disturbance reference of the corresponding load node.

[0077] Specifically, establishing normal disturbance benchmarks for different load nodes enables subsequent fault identification processes to make judgments based on the normal operating characteristics of the load nodes themselves, rather than relying on a unified fixed threshold. By setting the normal disturbance benchmarks as dynamic benchmarks that match the operating status, the interference of normal load fluctuations, operating status switching, and data quality changes on fault identification results can be reduced. By updating the benchmark module through a sample library, the normal disturbance benchmarks can be continuously corrected as the load operating characteristics change, thereby improving adaptability in long-term operating scenarios.

[0078] Furthermore, such as Figure 2 As shown, step S3 includes the following steps:

[0079] S301, according to the... Each load node at the sampling time The standardized operating sequence is used to construct the operating state feature vector and the normal perturbation feature vector:

[0080]

[0081] in, Indicates the first Each load node at the sampling time The feature vector of the running state; Indicates the first Each load node at the sampling time The A standardized operational status characteristic, ; Dimensions representing operational status characteristics; Indicates the first Each load node at the sampling time The normal perturbation feature vector; Indicates the first Each load node at the sampling time The A normal disturbance characteristic ; Dimensions representing the characteristics of normal perturbations;

[0082] S302, according to the... Calculate the normal disturbance coupling matrix using historical normal operation samples of each load node:

[0083]

[0084] in, Indicates the first Normal disturbance coupling matrix of each load node; Indicates the first The operational status feature matrix of each load node in the historical normal operation sample; Indicates the first Normal disturbance feature matrix of each load node in historical normal operation samples; This indicates the sampling time corresponding to the historical normal operation sample; This indicates the number of historically normal operating samples; Represents the regularization coefficient; Represents the identity matrix;

[0085] S303. Decouple the current operating state feature vector from the normal disturbance coupling matrix to obtain the fault residual vector:

[0086]

[0087] in, Indicates the first Each load node at the sampling time The fault residual vector is used to characterize the fault-related features retained after removing the effects of normal load disturbances from the current operating state features;

[0088] S304. Enhance the fault residual vector based on the characteristic reliability to obtain the enhanced fault residual vector:

[0089]

[0090] in, Indicates the first Each load node at the sampling time The fault residual enhancement vector; Indicates the first Each load node at the sampling time The feature reliability weight matrix; Indicates the first Each load node before the current sampling time The normal residual mean vector within each sampling period; Indicates the length of the sliding statistics window;

[0091] S305. Calculate the fault significance index based on the fault residual enhancement vector:

[0092]

[0093] Indicates the first Each load node at the sampling time Fault significance index; Indicates the first Each load node before the current sampling time Normal residual covariance matrix within each sampling period; Indicates the first Each load node at the sampling time Data quality penalty coefficient; when When the value is greater than the adaptive saliency threshold of the corresponding load node, the th node will be... Each load node is added to the initial set of abnormal load nodes.

[0094] Specifically, by establishing the coupling relationship between the normal disturbance feature vector and the operating state feature vector through the normal disturbance coupling matrix, the normal operation disturbance component can be deducted from the current operating state features, making the fault residual vector more concentrated in representing fault-related changes. By correcting the mean of the normal residual through the fault residual enhancement vector and introducing feature reliability weights, the influence of low-reliability features on the identification results can be suppressed, highlighting stable and effective fault features. By quantifying the fault significance index in combination with the normal residual covariance matrix and the data quality penalty coefficient, the identification accuracy of weak fault features can be improved under conditions of load fluctuation, disturbance superposition, and data quality changes. Step S3 can significantly reduce misjudgments and missed judgments caused by normal operation disturbances.

[0095] Furthermore, step S4 includes:

[0096] S401. Construct an anomaly identification input vector based on the fault residual enhancement vector, the fault significance index, and the time-series change in fault significance.

[0097] S402. Input the anomaly identification input vector into the fault identification model and output the probability distribution of the fault type corresponding to the abnormal load node;

[0098] S403. Determine the fault type corresponding to the highest probability in the fault type probability distribution as the fault type identification result, and determine the highest probability as the fault confidence level;

[0099] S404. When the fault confidence level is lower than the preset confidence threshold, the abnormal state of the abnormal load node is retained, and the fault identification result is corrected in step S5 by combining the topology-time consistency location calculation result.

[0100] Specifically, by using the enhanced fault residual results as input to the fault identification model, the model can make judgments mainly based on the fault-related features after normal disturbance decoupling, reducing the interference of normal load disturbances on fault type identification. By outputting fault confidence, it can provide a basis for credibility weight for subsequent topology-time consistency localization. By retaining the abnormal state of low-confidence abnormal load nodes instead of directly excluding them, it can avoid the premature elimination of real fault nodes due to the uncertainty of single-node model judgment, thereby improving the robustness of the overall fault identification link.

[0101] Furthermore, such as Figure 3 As shown, step S5 includes the following steps:

[0102] S501. Construct a set of candidate fault nodes based on the fault significance index and fault confidence:

[0103]

[0104] in, Indicates the sampling time The corresponding set of candidate faulty nodes; Indicates the first Each load node at the sampling time Adaptive significance threshold; Indicates the first Fault confidence of each load node This indicates a pre-set lower confidence level;

[0105] S502. Calculate the topology-timing deviation of candidate node j in the candidate fault node set:

[0106]

[0107] in, Indicates candidate nodes At sampling time Topology-timing skewness; Indicates the first Abnormal weights of candidate nodes; Indicates the first The abnormal start time of each candidate node; Indicates candidate nodes The abnormal start time; Indicates candidate nodes To candidate nodes The shortest electrical distance in the network topology; This represents the equivalent speed at which fault disturbances propagate along the network topology;

[0108] S503. Based on the topology-time series deviation, fault significance index, and fault confidence, calculate the fault location score of candidate node j:

[0109]

[0110] in, Indicates candidate nodes At sampling time The fault location score indicates that the smaller the fault location score, the more likely it is to be a candidate node. The closer to the actual faulty load node; Represents the topology-time skew weights; Indicates the significance weight of the fault; Indicates the fault confidence weight;

[0111] S504. Determine the target faulty load node based on the fault location score, and generate the fault impact range:

[0112]

[0113] in, Indicates the sampling time The target faulty load node has been identified; Indicates the sampling time The corresponding scope of the fault impact; Represents any load node in the target power grid; Indicates the target faulty load node With load nodes The topological association identifier between them is set to 1 if a fault propagation association exists, and 0 if no fault propagation association exists. Indicates the target faulty load node To load node The shortest electrical distance; Indicates the time when the anomaly begins in the target faulty load node; This represents the correction factor for the scope of influence.

[0114] Specifically, by filtering out anomalous nodes that simultaneously meet the requirements of fault significance and fault confidence from the candidate fault node set, the interference of low-confidence anomalous points on the localization process can be reduced. By incorporating the anomalous start time, shortest electrical distance, and equivalent propagation speed into the localization calculation using topology-time sequence deviation, it is possible to determine whether the order of anomalous occurrence among candidate nodes conforms to the law of fault propagation along the network topology. By using anomalous weights, candidate nodes with high fault significance and high fault confidence have a greater impact on the localization results, improving the physical consistency of the localization score. By integrating topology-time sequence deviation, fault significance, and fault confidence into the fault localization score, it is possible to distinguish between the real faulty load node and the affected associated load node from multiple anomalous nodes. Step S5 can improve the speed and accuracy of determining the target faulty load node and narrow the scope of fault impact when multiple node anomalousness occurs simultaneously.

[0115] Furthermore, the anomaly weight is calculated according to the following formula:

[0116]

[0117] in, This represents any candidate node in the set of candidate faulty nodes.

[0118] Furthermore, step S6 includes:

[0119] S601. Link and store the target faulty load node, fault type identification result, fault confidence, fault impact range and on-site handling result to generate a closed-loop fault sample.

[0120] S602. When the on-site handling results confirm that the fault identification results are correct, the corresponding fault residual enhancement vector, fault significance index, fault type identification results and target fault load node are added to the fault sample library.

[0121] S603. When the on-site handling results confirm that there is a deviation in the fault identification results, the corresponding fault residual enhancement vector, actual fault type and actual fault load node are added to the correction sample library.

[0122] S604. According to the preset update cycle, update the fault identification model based on the fault sample library and the correction sample library, and update the normal disturbance benchmark based on the fault-free operation data confirmed on site.

[0123] Furthermore, before generating the fault identification result, step S6 also includes:

[0124] Starting with the target faulty load node, the propagation consistency of associated load nodes within the fault's influence range is verified along the fault propagation direction in the network topology. If the abnormal starting order of the associated load nodes is inconsistent with the fault propagation direction in the network topology, or if the fault salience of the associated load nodes does not show a decreasing trend from the target faulty load node outwards, the location reliability of the target faulty load node is downgraded, and step S5 is called again to perform topology-temporal consistency location calculation. If the abnormal starting order of the associated load nodes is consistent with the fault propagation direction in the network topology, and the fault salience of the associated load nodes shows a decreasing trend from the target faulty load node outwards, the target faulty load node is confirmed as the final faulty load node.

[0125] Specifically, through propagation consistency verification, the location result of the target faulty load node is verified a second time against the fault propagation direction, anomaly initiation sequence, and fault significance decay trend in the network topology. This avoids location deviations caused by relying solely on a single location score. When the propagation law is not satisfied, the topology-time consistency location calculation is called again, which can self-correct the initial location result and improve the credibility of the final faulty load node. When the propagation law is satisfied, the target faulty load node is confirmed as the final faulty load node, which can improve the output certainty of the fault identification result. This step further strengthens the location result of step S5, ensuring that the fault identification result satisfies both data feature judgment and network topology propagation law, thereby improving the reliability of fault identification and location in a dynamic power grid environment. By writing back the fault identification result confirmed on-site to the sample library, and then having the sample library feed back to the benchmark module, the normal disturbance benchmark can be continuously updated based on confirmed fault-free operating data and closed-loop fault samples, enhancing the system's adaptability to changes in load operating characteristics and reducing identification errors caused by benchmark drift during long-term operation.

[0126] Specifically, before writing the fault identification results confirmed on-site back to the sample library, a baseline freeze mark is set for the load nodes within the fault's influence range. During the validity period of the baseline freeze mark, the operating data corresponding to the load nodes within the fault's influence range does not participate in the update of the normal disturbance baseline. When the on-site confirmation results indicate that the corresponding load node has recovered to a fault-free operating state within multiple consecutive sampling periods, the baseline freeze mark is removed, and the removed fault-free operating data is used as a normal operating sample to participate in the normal disturbance baseline update. By setting the baseline freeze mark, this invention can prevent abnormal operating data during the fault occurrence or the initial stage of fault recovery from being mistakenly incorporated into historical normal operating samples, and prevent the normal disturbance baseline from being contaminated by fault samples, thereby improving the long-term stability of the normal disturbance baseline and the accuracy of subsequent fault identification.

[0127] As one embodiment of the present invention, this embodiment provides a power grid load fault identification system applied to the above-described method, the system comprising:

[0128] The acquisition module is used to acquire node operation measurement data, node status data and network topology data of each load node in the target power grid, and to perform synchronous processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node.

[0129] The benchmark module is used to construct the operating state feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples.

[0130] The decoupling module is used to perform normal disturbance decoupling and fault residual enhancement on each load node according to the operating state feature vector, normal disturbance feature vector and normal disturbance benchmark, to obtain the fault saliency index of each load node, and to determine the initial abnormal load node set according to the fault saliency index.

[0131] The identification module is used to input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence levels corresponding to each abnormal load node.

[0132] The location module is used to perform topology-time consistency location calculation on the initial abnormal load node set based on the fault saliency index, fault confidence and network topology data, so as to obtain the target fault load node and the fault impact range.

[0133] The verification module is used to perform propagation consistency verification on associated load nodes within the fault influence range along the fault propagation direction in the network topology, starting from the target fault load node. Based on the propagation consistency verification result, it generates a fault identification result and writes the fault identification result confirmed on-site back to the sample library to update the normal disturbance benchmark and fault identification model.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying power grid load faults, characterized in that, The method includes the following steps: Step S1: Obtain node operation measurement data, node status data and network topology data of each load node in the target power grid, and perform synchronization processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node. Step S2: Construct the operating status feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples; Step S3: Based on the operating state feature vector, normal disturbance feature vector, and normal disturbance benchmark, perform normal disturbance decoupling and fault residual enhancement on each load node to obtain the fault saliency index of each load node, and determine the initial abnormal load node set based on the fault saliency index. Step S4: Input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence for each abnormal load node. Step S5: Based on the fault saliency index, fault confidence and network topology data, perform topology-time consistency location calculation on the initial abnormal load node set to obtain the target fault load node and the fault impact range. Step S6: Taking the target faulty load node as the starting node, perform propagation consistency verification on the associated load nodes within the fault influence range along the fault propagation direction in the network topology, generate fault identification results based on the propagation consistency verification results, and write back the fault identification results after on-site confirmation to the sample library to update the normal disturbance benchmark and fault identification model.

2. The power grid load fault identification method according to claim 1, characterized in that, Step S1 specifically includes: Using load nodes in the network topology as data organization units, node operation measurement data and node status data of each load node are acquired according to a unified time base, and a mapping relationship between load node identifier, sampling time, and topological location is established. Time-scale correction and sequence alignment are performed on the node operation measurement data and node status data of the same load node generated from different acquisition sources to obtain node time-series data that matches the network topology data. Based on operational boundary constraints, temporal continuity constraints, and topological correlation constraints, the node time-series data undergoes quality verification, identifying and eliminating abnormal sampling records inconsistent with the power grid operation mechanism. Missing data recovery, dimension normalization, and topological index binding are performed on the quality-verified node time-series data to generate a standardized operation sequence uniquely corresponding to the load node, sampling time, and topological location.

3. The power grid load fault identification method according to claim 2, characterized in that, S201. Based on the historical normal operation samples of each load node, extract the normal disturbance characteristics of the corresponding load node under different operating conditions; S202. Calculate the normal disturbance reference for the corresponding load node based on the normal disturbance characteristics. The normal disturbance reference includes the normal disturbance coupling matrix, the normal residual mean vector, and the normal residual covariance matrix. S203. Based on the operating state corresponding to the current sampling time, retrieve the normal disturbance reference that matches the current operating state from the normal disturbance reference; S204. When the fault-free operation data confirmed on-site meets the update conditions, the fault-free operation data is incorporated into the historical normal operation sample to correct the normal disturbance reference of the corresponding load node.

4. The power grid load fault identification method according to claim 3, characterized in that, Step S3 includes the following steps: S301, according to the... Each load node at the sampling time The standardized operating sequence is used to construct the operating state feature vector and the normal perturbation feature vector: in, Indicates the first Each load node at the sampling time The feature vector of the running state; Indicates the first Each load node at the sampling time The A standardized operational status characteristic, ; Dimensions representing operational status characteristics; Indicates the first Each load node at the sampling time The normal perturbation feature vector; Indicates the first Each load node at the sampling time The A normal disturbance characteristic ; Dimensions representing the characteristics of normal perturbations; S302, according to the... Calculate the normal disturbance coupling matrix using historical normal operation samples of each load node: in, Indicates the first Normal disturbance coupling matrix of each load node; Indicates the first The operational status feature matrix of each load node in the historical normal operation sample; Indicates the first Normal disturbance feature matrix of each load node in historical normal operation samples; This indicates the sampling time corresponding to the historical normal operation sample; This indicates the number of historically normal operating samples; Represents the regularization coefficient; Represents the identity matrix; S303. Decouple the current operating state feature vector from the normal disturbance coupling matrix to obtain the fault residual vector: in, Indicates the first Each load node at the sampling time The fault residual vector is used to characterize the fault-related features retained after removing the effects of normal load disturbances from the current operating state features; S304. Enhance the fault residual vector based on the characteristic reliability to obtain the enhanced fault residual vector: in, Indicates the first Each load node at the sampling time The fault residual enhancement vector; Indicates the first Each load node at the sampling time The feature reliability weight matrix; Indicates the first Each load node before the current sampling time The normal residual mean vector within each sampling period; Indicates the length of the sliding statistics window; S305. Calculate the fault significance index based on the fault residual enhancement vector: Indicates the first Each load node at the sampling time Fault significance index; Indicates the first Each load node before the current sampling time Normal residual covariance matrix within each sampling period; Indicates the first Each load node at the sampling time Data quality penalty coefficient; when When the value is greater than the adaptive saliency threshold of the corresponding load node, the th node will be... Each load node is added to the initial set of abnormal load nodes.

5. The power grid load fault identification method according to claim 4, characterized in that, Step S4 includes: S401. Construct an anomaly identification input vector based on the fault residual enhancement vector, the fault significance index, and the time-series change in fault significance. S402. Input the anomaly identification input vector into the fault identification model and output the probability distribution of the fault type corresponding to the abnormal load node; S403. Determine the fault type corresponding to the highest probability in the fault type probability distribution as the fault type identification result, and determine the highest probability as the fault confidence level; S404. When the fault confidence level is lower than the preset confidence threshold, the abnormal state of the abnormal load node is retained, and the fault identification result is corrected in step S5 by combining the topology-time consistency location calculation result.

6. The power grid load fault identification method according to claim 5, characterized in that, Step S5 includes the following steps: S501. Construct a set of candidate fault nodes based on the fault significance index and fault confidence. S502. Calculate the topology-timing deviation of candidate node j in the candidate fault node set; S503. Calculate the fault location score of candidate node j based on the topology-time series deviation, fault significance index and fault confidence. S504. Determine the target faulty load node based on the fault location score, and generate the fault impact range.

7. The power grid load fault identification method according to claim 6, characterized in that, The anomaly weights are determined based on the fault significance index and fault confidence of the candidate nodes.

8. The power grid load fault identification method according to claim 7, characterized in that, Step S6 includes: S601. Link and store the target faulty load node, fault type identification result, fault confidence, fault impact range and on-site handling result to generate a closed-loop fault sample. S602. When the on-site handling results confirm that the fault identification results are correct, the corresponding fault residual enhancement vector, fault significance index, fault type identification results and target fault load node are added to the fault sample library. S603. When the on-site handling results confirm that there is a deviation in the fault identification results, the corresponding fault residual enhancement vector, actual fault type and actual fault load node are added to the correction sample library. S604. According to the preset update cycle, update the fault identification model based on the fault sample library and the correction sample library, and update the normal disturbance benchmark based on the fault-free operation data confirmed on site.

9. A method for identifying power grid load faults according to claim 8, characterized in that, Before generating the fault identification result, step S6 further includes: Starting with the target faulty load node, the propagation consistency of associated load nodes within the fault's influence range is verified along the fault propagation direction in the network topology. If the abnormal starting order of the associated load nodes is inconsistent with the fault propagation direction in the network topology, or if the fault salience of the associated load nodes does not show a decreasing trend from the target faulty load node outwards, the location reliability of the target faulty load node is downgraded, and step S5 is called again to perform topology-temporal consistency location calculation. If the abnormal starting order of the associated load nodes is consistent with the fault propagation direction in the network topology, and the fault salience of the associated load nodes shows a decreasing trend from the target faulty load node outwards, the target faulty load node is confirmed as the final faulty load node.

10. A power grid load fault identification system, applied to the method described in any one of claims 1-9, characterized in that, The system includes: The acquisition module is used to acquire node operation measurement data, node status data and network topology data of each load node in the target power grid, and to perform synchronous processing, quality verification and standardization processing on the node operation measurement data and node status data to obtain the standardized operation sequence corresponding to each load node. The benchmark module is used to construct the operating state feature vector and normal disturbance feature vector of each load node according to the standardized operating sequence, and generate the normal disturbance benchmark of the corresponding load node based on historical normal operating samples. The decoupling module is used to perform normal disturbance decoupling and fault residual enhancement on each load node according to the operating state feature vector, normal disturbance feature vector and normal disturbance benchmark, to obtain the fault saliency index of each load node, and to determine the initial abnormal load node set according to the fault saliency index. The identification module is used to input the fault residual enhancement results corresponding to each load node in the initial abnormal load node set into the fault identification model to obtain the fault type identification results and fault confidence levels corresponding to each abnormal load node. The location module is used to perform topology-time consistency location calculation on the initial abnormal load node set based on the fault saliency index, fault confidence and network topology data, so as to obtain the target fault load node and the fault impact range. The verification module is used to perform propagation consistency verification on associated load nodes within the fault influence range along the fault propagation direction in the network topology, starting from the target fault load node. Based on the propagation consistency verification result, it generates a fault identification result and writes the fault identification result confirmed on-site back to the sample library to update the normal disturbance benchmark and fault identification model.

Images