A knowledge graph-based power data processing method and system

Abstract

The application relates to the technical field of computers and provides a power data processing method and system based on a knowledge graph, which comprises the following steps: establishing a data source binary tree according to data sources, and associating and mapping power data with entities in a power knowledge graph; first data features are extracted from a root node of the data source binary tree in a hierarchical order, preliminary anomaly determination is carried out based on the first data features and entity attributes corresponding to each data source in the power knowledge graph, and suspected abnormal data are determined; the upper layer node and the associated nodes in the same layer are determined by backtracking from the nodes corresponding to the suspected abnormal data in the data source binary tree; secondary anomaly determination is carried out based on the second data features of the upper layer node and the third data features of the associated nodes in the same layer and the association relationship between the fault modes and the fault propagation paths among the entities in the power knowledge graph, and a final anomaly determination result is determined. The application improves the accuracy and reliability of power data anomaly detection.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a power data processing method and system based on a knowledge graph. Background Art

[0002] During power system operation, power data anomaly detection is a critical step in ensuring stable, secure, and efficient grid operation. Existing power data anomaly detection methods typically rely on statistical analysis of historical data, identifying anomalous data by setting fixed thresholds or employing simple clustering algorithms. For example, a data fluctuation range is set based on the mean and standard deviation of historical data, and data outside this range is considered anomaly. However, due to the complex and ever-changing operating conditions of power systems, equipment parameters, environmental factors, and other factors all affect power data. Fixed thresholds are difficult to adapt to the changing data characteristics in different scenarios, and simple clustering algorithms are unable to effectively mine the complex correlations between data. This results in a large amount of normal data being misclassified as anomalies, or genuine anomaly data being missed. This severely impacts the accuracy and reliability of anomaly detection and fails to meet the power system's requirements for high-precision and real-time anomaly detection. Summary of the Invention

[0003] The present invention provides a power data processing method and system based on knowledge graph, aiming to improve the accuracy and reliability of power data anomaly detection.

[0004] In a first aspect, the present invention provides a method for processing power data based on a knowledge graph, comprising:

[0005] Establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records data source information and corresponding data;

[0006] Performing hierarchical extraction from the root node of the binary tree of the data sources in hierarchical order to obtain a first data feature under each data source, and performing preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data;

[0007] Based on the data logical relationship, backtrack from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper layer node and the same layer associated node of the suspected abnormal data;

[0008] Based on the second data feature of the upper-level node and the third data feature of the same-level associated node, a secondary abnormality judgment is performed with the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph to determine the final abnormality judgment result.

[0009] In a second aspect, the present invention further provides a power data processing system based on a knowledge graph, which is applied to the power data processing method based on a knowledge graph as described in the first aspect; the power data processing system based on a knowledge graph comprises:

[0010] A data processing module is used to establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records data source information and corresponding data;

[0011] a first anomaly determination module, configured to perform hierarchical extraction from the root node of the binary tree of the data sources in hierarchical order to obtain a first data feature under each data source, and perform preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data;

[0012] An abnormal data backtracking module is used to backtrack from the node corresponding to the suspected abnormal data in the data source binary tree based on the data logical relationship, and determine the upper layer node and the same layer associated node of the suspected abnormal data;

[0013] The second abnormality judgment module is used to perform secondary abnormality judgment based on the second data feature of the upper-level node and the third data feature of the same-level associated node, and the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph to determine the final abnormality judgment result.

[0014] In a third aspect, the present invention also provides an electronic device, comprising: a memory for storing a computer software program; a processor for reading and executing the computer software program, thereby implementing any of the above-mentioned knowledge graph-based power data processing methods.

[0015] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium, in which a computer software program is stored. When the computer software program is executed by a processor, it implements any of the above-mentioned knowledge graph-based power data processing methods.

[0016] In a fifth aspect, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described power data processing methods based on knowledge graphs.

[0017] The power data processing method based on the knowledge graph provided by the embodiment of the present invention can flexibly adjust the basis for abnormality judgment according to different operating scenarios, fault modes and fault propagation paths through the knowledge graph in the power field that contains rich equipment operation rules and fault characteristics, thereby improving the adaptability to complex scenarios. At the same time, by establishing a binary tree of data sources to sort out the hierarchical relationship of data, in the process of feature extraction and backtracking analysis, the correlation features between data from the same source and different sources are deeply mined, and the complex connections between data can be analyzed more accurately. In the secondary judgment link, the fault logic of the knowledge graph and the backtracking results of the data binary tree are combined to analyze the data from multiple dimensions and multiple correlation perspectives, effectively avoiding the misjudgment of normal data and the omission of abnormal data. Therefore, the embodiment of the present invention improves the accuracy and reliability of power data anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a method for processing power data based on a knowledge graph provided by an embodiment of the present invention;

[0019] Figure 2 This is a structural diagram of a power data processing system based on a knowledge graph provided by an embodiment of the present invention;

[0020] Figure 3 An embodiment diagram of an electronic device provided by an embodiment of the present invention;

[0021] Figure 4 A diagram of an embodiment of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0024] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0025] Optional, see Figure 1 , Figure 1 This is a flow chart of the power data processing method based on the knowledge graph provided by the present invention. In the embodiment of the present invention, the execution subject of the power data processing method based on the knowledge graph is an anomaly detection system. Therefore, the power data processing method based on the knowledge graph includes:

[0026] Step 10: Establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records the data source information and the corresponding data.

[0027] Optionally, the anomaly detection system collects power data from the power system and analyzes its sources, identifying the source of each type of data, such as sensors, smart meters, and monitoring equipment. Furthermore, the anomaly detection system constructs a binary tree structure with data sources as nodes. Each node records specific information about the data source, including device type, installation location, and collection frequency, and is associated with the power data corresponding to that source to construct a data source binary tree.

[0028] Furthermore, the anomaly detection system matches the data of each node in the binary tree of the data source with the entities in the pre-built power knowledge graph, determines the knowledge graph entity corresponding to the data source through information such as data characteristics and equipment attributes, and realizes the association mapping between data and knowledge graph.

[0029] In one embodiment, power data from a regional power grid is collected, including temperature data from transformer A (installed at substation 1, model T-100), current data from line B (connecting substations 1 and 2), and voltage data from smart meter C (located in cell X). A binary tree of data sources is constructed, with the root node being "Regional Power Grid Data," the left child being "Substation 1 Data," and the right child being "Line and User Data." The left child of "Substation 1 Data" is "Transformer A Temperature Data," which records information such as the device type (transformer), installation location (substation 1), and collection frequency (once per minute). The right child is other device data (not discussed here). The left child of "Line and User Data" is "Line B Current Data," and the right child is "Smart Meter C Voltage Data." Next, transformer A is associated with the "T-100 Model Transformer" entity under the "Transformer" entity class in the power knowledge graph, line B is associated with the "B Line" entity under the "Transmission Line" entity class, and smart meter C is associated with the "C Model Smart Meter" entity under the "Metering Equipment" entity class to complete the mapping of data and knowledge graph.

[0030] Step 20: Perform hierarchical extraction from the root node of the binary tree of the data source in hierarchical order to obtain the first data feature under each data source, and make a preliminary anomaly judgment based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data.

[0031] Furthermore, the anomaly detection system extracts data features from each data source layer by layer starting from the root node according to the hierarchical order of the binary tree of data sources, namely, the first data features, where the first data features include the average value, variance, fluctuation range, time series trend, etc. of the data.

[0032] Furthermore, after the extraction is completed, the anomaly detection system compares and analyzes the first data feature of each data source with the entity attributes of the corresponding entity of the data source in the power knowledge graph, where the entity attributes in the power knowledge graph include information such as the normal range and standard characteristics of this type of equipment or data. Through comparison, if the first data feature exceeds the normal range of the entity attribute or is significantly different from the standard feature, the data is determined to be suspected abnormal data, as shown in the process from step 201 to step 204.

[0033] Step 30 , based on the data logical relationship, backtrack from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper layer node and the same layer associated nodes of the suspected abnormal data.

[0034] Furthermore, after identifying suspected abnormal data, the anomaly detection system backtracks from the node corresponding to the suspected abnormal data in the data source binary tree based on the logical relationships between data in the power system. During the backtracking process, the anomaly detection system needs to determine the parent node of the node corresponding to the suspected abnormal data, that is, the parent node of the data source, as well as the same-level associated nodes, that is, other data source nodes at the same level as the data source and logically related to the power system.

[0035] Optionally, the upper-level nodes in the embodiment of the present invention reflect the hierarchical relationship of data sources, and the associated nodes at the same level reflect the horizontal logical connection between devices or data in the power system.

[0036] In one embodiment, the transformer A temperature data and line B current data, initially identified as suspected abnormal data in step 20, are used as an example. For transformer A temperature data, the corresponding node in the data source binary tree is "Transformer A Temperature Data," with the upper node being "Substation 1 Data," and the associated nodes on the same level being other device data nodes under "Substation 1" (such as other transformer data and circuit breaker data for Substation 1). The anomaly detection system traces back to the "Substation 1 Data" node, obtains the other device data information under that node, and simultaneously locates other device data nodes on the same level as "Transformer A Temperature Data." For line B current data, the corresponding node is "Line B Current Data," with the upper node being "Line and User Data," and the associated nodes on the same level being "Smart Meter C Voltage Data," as well as other possible line data nodes (such as Line D Data). The system then traces back to the "Line and User Data" node and obtains information such as Smart Meter C voltage data on the same level.

[0037] Step 40: Based on the second data feature of the upper-level node and the third data feature of the associated node at the same level, and the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph, a secondary abnormality determination is performed to determine the final abnormality determination result.

[0038] Furthermore, the electric power knowledge graph in the embodiment of the present invention records the possible fault impact relationships between different equipment entities and the paths and modes of fault propagation. Therefore, the anomaly detection system obtains the second data feature of the upper-level node of the suspected abnormal data node and the third data feature of the same-level associated node, and performs secondary anomaly judgment based on the correlation between the second data feature of the upper-level node and the fault mode and fault propagation path between entities in the electric power knowledge graph, as well as the correlation between the third data feature of the same-level associated node and the fault mode and fault propagation path between entities in the electric power knowledge graph.

[0039] Further, the abnormality detection system analyzes whether the second data features of the upper layer node are abnormal, if the upper layer node is abnormal, it is considered that the suspected abnormal data may be caused by upper layer fault propagation; meanwhile, it analyzes whether the third data features of the associated nodes in the same layer are abnormal, if the associated nodes in the same layer are abnormal, it is considered that it may be caused by the fault in the same layer. Through comprehensive judgment, the final abnormality determination result is determined, that is, whether the suspected abnormal data belongs to misjudgment abnormality (actually normal), or is caused by upper layer fault propagation, or / and is caused by fault in the same layer, specifically as the processes of steps 401 to 404.

[0040] The embodiment of the application can flexibly adjust the abnormality determination basis according to different operation scenarios, fault modes and fault propagation paths, and improve the adaptability to complex scenarios, by containing the power field knowledge graph including rich device operation rules and fault features. Meanwhile, the data hierarchical relationship is sorted by establishing the data source binary tree, the associated features between the data of the same source and different sources are deeply mined in the feature extraction and backtracking analysis process, and the complex relationship between the data can be more accurately analyzed. In the secondary determination link, the data is analyzed from multiple dimensions and multiple association angles in combination with the fault logic of the knowledge graph and the backtracking result of the data binary tree, the normal data misjudgment and abnormal data omission are effectively avoided, and the accuracy and reliability of the power data abnormality detection are improved.

[0041] In an embodiment, steps 201 to 204 are described as follows:

[0042] In step 201, based on the normal operation state features and fault case feature modes of each data source corresponding entity in the power knowledge graph, the first data features under each data source are combined to construct a three-dimensional association network. Each network node in the three-dimensional association network represents a type of feature, and if the first data features and the normal operation state features or the fault case feature modes have direct or indirect semantic association paths in the power knowledge graph, an edge connection is established between the corresponding network nodes.

[0043] Optionally, the abnormality detection system extracts the normal operation state features (such as rated temperature, standard current range, etc.) and fault case feature modes (such as temperature mutation mode when overloaded, current surge mode when short-circuited, etc.) of each data source corresponding entity from the power knowledge graph, and these features and the first data features are jointly used as nodes of the three-dimensional association network. Further, the abnormality detection system judges whether the first data features and the normal operation features or the fault case features have direct or indirect semantic association paths (such as connected through “belongs to”, “affects”, “associates” and the like) through the semantic relationship of the power knowledge graph, if there is, an edge connection is established between the corresponding nodes, and the weight of the edge can be initially set as the association strength of the semantic path (based on the relationship weight or path length reciprocal of the power knowledge graph).

[0044] In one embodiment, taking transformer A's temperature data as an example, its first data features are "85°C average value," "variance 5," and "temperature rising trend." The normal operating state features for the corresponding entity "T-100 Transformer" in the power knowledge graph include "rated temperature 75°C," "temperature variance standard ≤ 3," and "stable normal trend." Fault case feature patterns include "temperature continues to rise and exceeds the rated value during an overload fault" and "variance increases during a heat dissipation fault." The anomaly detection system uses these features as nodes to construct a three-dimensional association network. The "85°C average value" is directly associated with "rated temperature 75°C" through the semantic relationship of "exceeds," while the "temperature rising trend" is indirectly associated with "temperature continues to rise during an overload fault" through the "matching fault pattern" relationship (path: temperature trend → fault feature → overload case). The "variance 5" is directly associated with "temperature variance standard ≤ 3" through "deviating from the standard" and is directly associated with "variance increases during a heat dissipation fault" through "fault feature matching." Edges are established between these semantically associated nodes, with edge weights temporarily set to 1 (subsequently adjusted through calculations in steps 202-203).

[0045] Step 202: Determine the feature deviation based on the three-dimensional association network, the semantic path length of the first data feature and the target normal operating status feature to which it is connected in the power knowledge graph, and the feature value of the first data feature and the feature value of the target normal operating status feature.

[0046] Furthermore, for each first data feature, the anomaly detection system finds all target normal operating state features that are connected to it in the three-dimensional association network, and calculates the semantic path length (i.e., the number of associated edges) between the first data feature and the target normal operating state feature in the knowledge graph. The path length is recorded as At the same time, calculate the degree of deviation of the characteristic value: if the characteristic is numerical (such as temperature, current), it is expressed as absolute deviation or relative deviation, and the formula is ( is the first data eigenvalue, is a normal characteristic value); if it is a trend or pattern characteristic (such as an upward trend), the deviation is calculated using the preset pattern matching algorithm (scope is a perfect match for the normal mode). Final feature deviation It is necessary to comprehensively consider the semantic path length and the feature value deviation. The calculation formula of the embodiment of the present invention is:

[0047] .in, is the mode deviation weight , set according to the feature type).

[0048] Continue transformer Temperature Average value" characteristic, and its target normal operating characteristic is "rated temperature ”, semantic path length (directly related), the eigenvalue deviation is , mode deviation (only numerical deviation), then For the “temperature rising trend” feature, the target normal feature is “normal trend stable”, and the semantic path length is (indirectly related through fault characteristics), pattern deviation (completely inconsistent with the stationary trend), the value deviates from 0. Assume ,but .

[0049] Step 203 : determining a feature matching degree based on the three-dimensional association network, based on the semantic similarity between the first data feature and the target fault case feature pattern to which it is connected, and the feature value of the first data feature and the feature value of the target fault case feature pattern.

[0050] Furthermore, the anomaly detection system finds the connection between the first data feature and the feature patterns of all target fault cases in the three-dimensional association network and calculates the semantic similarity between the two. , the embodiment of the present invention can use the cosine similarity of the electric power knowledge graph embedding vector, such as the cosine value of the vector angle generated by the TransE model. At the same time, calculate the matching degree of the feature value: for numerical features, use , is the typical range of fault characteristics; for pattern-type features, the embodiment of the present invention uses dynamic time warping (DTW) or hidden Markov model (HMM) to calculate the matching probability , therefore, the final feature matching Combining semantic similarity and feature value matching, the formula is:

[0051] .in is the pattern matching weight .

[0052] Continue with the above transformer Temperature The semantic similarity between the fault characteristics of "average value" and "temperature continues to rise during overload fault" (calculated by knowledge graph embedding vector), the typical range of fault characteristics is , then the eigenvalue matching is , pattern matching (Temperature rising trend matches overload mode), set ,but: For the “variance 5” feature, the semantic similarity with the “variance increases during heat dissipation failure” fault feature is , the fault variance typically ranges from 4 to 8, and the eigenvalue matching is , pattern matching set up ,but .

[0053] Step 204 : Perform preliminary abnormality determination based on the feature deviation and feature matching of each first data feature to determine suspected abnormal data.

[0054] Furthermore, the anomaly detection system performs preliminary anomaly determination based on the feature deviation and feature matching of each first data feature, and determines suspected abnormal data, as specifically described in steps 2041 to 2044 .

[0055] The embodiment of the present invention uses a knowledge graph in the power field that contains rich equipment operation rules and fault characteristics, combined with the first data feature under each data source to perform anomaly judgment, which can accurately identify suspected abnormal data and effectively avoid misjudgment of normal data and omission of abnormal data.

[0056] In one embodiment, steps 2041 to 2044 are described as follows:

[0057] In step 2041, a first data feature whose feature deviation is greater than a deviation threshold and whose feature matching is greater than a matching threshold is determined as a preliminary abnormal feature, and an abnormal feature association subgraph is constructed based on the association relationship between the preliminary abnormal features in the three-dimensional association network.

[0058] Optionally, a deviation threshold and a matching threshold are set. The anomaly detection system compares the feature deviation of each first data feature with the deviation threshold, and the feature matching with the matching threshold. First data features that satisfy both a feature deviation greater than the deviation threshold and a feature matching greater than the matching threshold are identified as preliminary anomaly features. Furthermore, based on the associations between these preliminary anomaly features in the three-dimensional association network, the anomaly detection system extracts a subgraph containing these preliminary anomaly feature nodes and the edges between them, and constructs an anomaly feature association subgraph. In the anomaly feature association subgraph, nodes represent preliminary anomaly features, and edges represent semantic association paths between features in the three-dimensional association network.

[0059] Continuing with the above embodiment, the deviation threshold is set to 1.0, and the matching threshold is set to 1.0. In the previous example, the three first data features of the transformer A temperature data, "85°C average value", "temperature rising trend", and "variance 5", all have feature deviations and feature matching greater than the threshold, so they are determined to be preliminary abnormal features. In the three-dimensional association network, "85°C average value" and "temperature rising trend" have an associated edge through "overload fault feature", "85°C average value" and "variance 5" have an associated edge through "heat dissipation fault feature", and "temperature rising trend" and "variance 5" also have an indirect associated edge through "equipment failure mode". The anomaly detection system extracts these three preliminary abnormal feature nodes and the associated edges between them to construct an abnormal feature association subgraph.

[0060] Step 2042: Determine the node degree of each preliminary abnormal feature based on the abnormal feature association subgraph, and determine the abnormal path dependency based on the first path number of the path between any two nodes in the abnormal feature association subgraph and the second path number of the path passing through the node corresponding to the preliminary abnormal feature.

[0061] Furthermore, in the abnormal feature associated subgraph, the anomaly detection system first calculates the node degree of each preliminary abnormal feature. The node degree refers to the number of edges connected to the node in the subgraph. Next, the anomaly detection system calculates the number of paths between any two nodes in the abnormal feature associated subgraph, recorded as the first path number. Then, the anomaly detection system calculates the number of paths passing through the node corresponding to each preliminary abnormal feature, recorded as the second path number. The abnormal path dependency is used to measure the path importance of a node in the subgraph. Its calculation formula is: For each preliminary abnormal feature node i, the abnormal path dependency:

[0062] . Where n is the number of nodes in the subgraph, The larger the value, the higher the path dependence of the node in the abnormal feature association subgraph.

[0063] In one embodiment, the constructed abnormal feature association subgraph contains three nodes: "85°C Average" (node ​​A), "Temperature Rising Trend" (node ​​B), and "Variance 5" (node ​​C). Node A has a node degree of 2 (connected to nodes B and C), node B has a node degree of 2 (connected to nodes A and C), and node C has a node degree of 2 (connected to nodes A and B). The paths between any two nodes in the subgraph are: there is one direct path from node A to node B, one direct path from node A to node C, and one direct path from node B to node C. Furthermore, there is an indirect path from node A to node C through node B. Therefore, the total number of first paths is: one path between nodes A and B, two paths (direct and indirect) between nodes A and C, and two paths (direct and indirect) between nodes B and C, for a total of 1 + 2 + 2 = 5. The number of second paths passing through node A: The indirect path from node B to node C passes through node A, so there is 1; the paths from node A to node B and node A to node C themselves pass through node A, so there are 1 (B to C through A) + 1 (A to B) + 2 (A to C) = 4. Then the abnormal path dependency of node A is =4 / 5=0.8. Similarly, the number of second paths for node B: the indirect path from node A to node C passes through node B, which is 1; the paths from node B to node A and node B to node C themselves pass through node B, so there are 1 (A to C through B) + 1 (B to A) + 2 (B to C) = 4, and the abnormal path dependency is =4 / 5=0.8. The number of second paths for node C: Does the indirect path from node A to node B pass through node C? In this subgraph, the direct path from node A to node B does not pass through node C. Does an indirect path exist? Assuming that there is only a direct path from node A to node B in the subgraph, the number of paths passing through node C is the path from node A to node C and then to node B, which is 1, plus the paths from node C to node A and node C to node B themselves pass through node C, so there are 1 (A to B through C) + 2 (C to A and C to B) = 3 paths, which is abnormal path dependency. =3 / 5=0.6.

[0064] Step 2043 : determining the number of neighbor nodes and the number of node edges between neighbor nodes of each preliminary abnormal feature based on the abnormal feature association subgraph, and determining the clustering coefficient based on the number of neighbor nodes and the number of node edges.

[0065] Furthermore, in the anomaly feature association subgraph, the anomaly detection system determines the number of neighbor nodes for each preliminary anomaly feature, that is, the number of nodes directly connected to the node. It also determines the number of node edges between neighbor nodes, that is, the number of edges that actually exist between the node's neighbor nodes.

[0066] The clustering coefficient in the embodiment of the present invention is used to measure the degree of connection between the neighboring nodes of a node, and its calculation formula is: for each preliminary abnormal feature node i, assuming that the number of its neighboring nodes is k(i), and the actual number of edges between the neighboring nodes is e(i), then the clustering coefficient is: C(i)=[2*e(i)] / [k(i)*(k(i)-1)].

[0067] Optionally, when k(i)<2, the clustering coefficient is defined as 0. The clustering coefficient represents the ratio of the actual number of connections between neighbor nodes to the maximum possible number of connections. The larger the ratio, the higher the degree of clustering between neighbor nodes, that is, the closer the feature correlation of the local area where the node is located.

[0068] Continuing with the above example, for node A ("85°C average"), its neighboring nodes are nodes B and C, the number of neighboring nodes k(A) = 2, and the number of actual edges between neighboring nodes e(A) = 1 (there is one edge between nodes B and C). Therefore, the clustering coefficient C(A) = (2*1) / [2*(2-1)] = 1. For node B ("Temperature Rising Trend"), its neighboring nodes are nodes A and C, k(B) = 2, e(B) = 1 (there is one edge between nodes A and C), and the clustering coefficient C(B) = (2*1) / [2*(2-1)] = 1. For node C ("Variance 5"), its neighboring nodes are nodes A and B, k(C) = 2, e(C) = 1 (there is one edge between nodes A and B), and the clustering coefficient C(C) = (2*1) / [2*(2-1)] = 1.

[0069] Step 2044 , performing preliminary anomaly determination based on the node degree, anomaly path dependency, clustering coefficient, feature deviation, and feature matching of each preliminary anomaly feature, and determining suspected anomaly data.

[0070] Furthermore, the anomaly detection system performs preliminary anomaly judgment based on the node degree, anomaly path dependency, clustering coefficient, feature deviation, and feature matching of each preliminary anomaly feature, and determines suspected anomaly data, as specifically described in the process from step 20441 to step 20445.

[0071] The embodiment of the present invention constructs an abnormal feature association subgraph based on preliminary abnormal features determined by feature deviation and feature matching, and then performs preliminary abnormality judgment based on the node degree, abnormal path dependency and clustering coefficient determined by the abnormal feature association subgraph. This enables the preliminary abnormality judgment to more accurately identify true abnormal data, reduce misjudgments and missed judgments, avoid misjudgment of normal data and omission of abnormal data, and thus improve the accuracy and reliability of power data anomaly detection.

[0072] In one embodiment, steps 20441 to 20445 are described as follows:

[0073] Step 20441: Determine the data corresponding to the preliminary abnormal features with node degrees greater than or equal to a preset threshold as suspected abnormal data. Or,

[0074] Optionally, embodiments of the present invention preset a node degree threshold to measure the importance of the connection of preliminary abnormal features in the abnormal feature association subgraph. The anomaly detection system compares the node degree of each preliminary abnormal feature with the threshold. If the node degree of a preliminary abnormal feature is greater than or equal to the preset threshold, the feature is considered to be in a critical connection position in the abnormal feature association subgraph, and the corresponding power data is more likely to be true abnormal data. Therefore, the data corresponding to the preliminary abnormal feature is determined to be suspected abnormal data.

[0075] Continuing with the above example, the preset node degree threshold is 2. In the previously constructed abnormality feature association subgraph for transformer A's temperature data, the node degrees of the three preliminary abnormality features, "85°C average," "temperature rising trend," and "variance 5," are all 2, which is greater than or equal to the preset threshold of 2. Therefore, the abnormality detection system identifies the transformer A temperature data corresponding to these three preliminary abnormality features as suspected abnormal data.

[0076] Step 20442: If the node degree of each preliminary abnormal feature is less than a preset threshold, then determine the abnormal betweenness value based on the abnormal path dependency and clustering coefficient of each preliminary abnormal feature. Or,

[0077] Furthermore, when the node degree of each preliminary anomaly feature is less than a preset threshold, it indicates that the direct connection importance of these features in the anomaly feature association subgraph is low, and further analysis is needed from the perspectives of path dependence and clustering characteristics. The anomaly betweenness value is used to comprehensively measure the mediating role of preliminary anomaly features in the anomaly propagation path and the degree of local clustering.

[0078] Therefore, for each preliminary abnormal feature node i, the calculation formula of the abnormal betweenness value in the embodiment of the present invention is: B(i)= *[1+C(i)], where is the anomaly path dependency, and C(i) is the clustering coefficient. This embodiment of the present invention combines the anomaly path dependency and the clustering coefficient. A higher path dependency indicates a higher proportion of nodes passed through in the anomaly propagation path. A higher clustering coefficient indicates a denser connection between the node's neighboring nodes and a higher degree of clustering of anomaly features in a local area. The combination of the two can more comprehensively reflect the importance of a node in anomaly propagation.

[0079] In one embodiment, the node degrees of the preliminary abnormal feature in the abnormal feature association subgraph are all less than a preset threshold of 2. For example, there are two preliminary abnormal feature nodes in the subgraph: node X and node Y, the node degree of node X is 1, and the node degree of node Y is 1. Calculate the abnormal path dependency of node X (X)=0.5, clustering coefficient C(X)=0 (because the number of neighbor nodes is 1, k(i)<2, clustering coefficient is defined as 0), then the abnormal betweenness value B(X)=0.5*(1+0)=0.5. Abnormal path dependence of node Y (Y)=0.6, clustering coefficient C(Y)=0, and abnormal betweenness value B(Y)=0.6*(1+0)=0.6.

[0080] Step 20443: Determine the data corresponding to the preliminary abnormal features whose abnormal betweenness value is greater than or equal to the preset abnormal threshold as suspected abnormal data. Or,

[0081] Furthermore, an abnormality threshold is preset to determine whether the abnormal betweenness value of the preliminary abnormal feature has reached a level that requires attention. The abnormality detection system compares the abnormal betweenness value of each preliminary abnormal feature with the preset abnormality threshold. If the abnormal betweenness value of a preliminary abnormal feature is greater than or equal to the preset abnormality threshold, it is considered that the feature exhibits a high abnormal importance in terms of abnormal propagation path and local clustering, and its corresponding power data may be abnormal data. Therefore, the data corresponding to the preliminary abnormal feature is determined as suspected abnormal data. In one embodiment, the preset abnormality threshold is 0.5. In the above example, the abnormal betweenness value of node X is 0.5, which is equal to the preset abnormality threshold; the abnormal betweenness value of node Y is 0.6, which is greater than the preset abnormality threshold. Therefore, the power data corresponding to node X and node Y are determined as suspected abnormal data.

[0082] In step 20444, if the anomaly betweenness value of each preliminary anomaly feature is less than the preset anomaly threshold, the anomaly propagation intensity is determined based on the feature deviation and feature matching of each preliminary anomaly feature.

[0083] Furthermore, when the anomaly betweenness value of each preliminary anomaly feature is less than the preset anomaly threshold, it indicates that the anomaly importance of these features is not yet clear from the perspective of path dependence and clustering characteristics, and it is necessary to further analyze the possibility of anomaly propagation from the perspective of the degree of deviation of the features themselves and the degree of matching with the fault mode. The anomaly propagation strength is used to comprehensively measure the role of the feature deviation and feature matching of the preliminary anomaly features in promoting anomaly propagation. For each preliminary anomaly feature node i, the calculation formula for the anomaly propagation strength of the embodiment of the present invention is: S(i)=D(i)*M(i)*(1+[D(i)+M(i)] / 2), where D(i) is the feature deviation and M(i) is the feature matching. Therefore, the embodiment of the present invention can highlight the enhanced effect on anomaly propagation when both the feature deviation and matching are high by multiplying the product of the feature deviation and feature matching by an enhancement factor related to the mean of the two, and more accurately reflect the propagation potential of the anomaly feature.

[0084] In one embodiment, the anomaly betweenness value of each preliminary anomaly feature in the anomaly feature association subgraph is less than a preset anomaly threshold. For example, there is a preliminary anomaly feature node Z with a feature deviation D(Z) = 1.2 and a feature matching degree M(Z) = 1.1. Therefore, the anomaly propagation strength S(Z) = 1.2*1.1*(1+[1.2+1.1] / 2)≈2.838.

[0085] Step 20445: determine the data corresponding to the preliminary abnormal features with abnormal propagation intensity greater than or equal to the preset propagation intensity threshold as suspected abnormal data.

[0086] Furthermore, a propagation strength threshold is preset to determine whether the abnormal propagation strength of the preliminary abnormal feature is sufficient to cause an abnormality. The abnormality detection system compares the abnormal propagation strength of each preliminary abnormal feature with the preset propagation strength threshold. If the abnormal propagation strength of a preliminary abnormal feature is greater than or equal to the preset propagation strength threshold, the feature is considered to have a high feature deviation and feature matching degree, with strong abnormal propagation potential. The corresponding power data may be abnormal data, and the data corresponding to the preliminary abnormal feature is determined to be suspected abnormal data.

[0087] In one embodiment, the preset propagation strength threshold is 2.0. In the above example, the abnormal propagation strength of node Z is 2.838, which is greater than the preset propagation strength threshold of 2.0. Therefore, the abnormality detection system determines the power data corresponding to node Z as suspected abnormal data.

[0088] The embodiment of the present invention performs preliminary anomaly determination based on the node degree, abnormal path dependency, clustering coefficient, feature deviation, and feature matching of each preliminary abnormal feature, so that the preliminary anomaly determination can more accurately identify true abnormal data, reduce the occurrence of misjudgments and missed judgments, effectively avoid misjudgment of normal data and omission of abnormal data, thereby improving the accuracy and reliability of power data anomaly detection.

[0089] In one embodiment, steps 401 to 404 are described as follows:

[0090] Step 401: Construct an entity fault mode propagation directed graph based on the fault modes and propagation paths between entities in the power knowledge graph. Each node in the entity fault mode propagation directed graph represents a power equipment entity, the edge represents the direction of fault propagation, and the edge attributes record the triggering conditions and impact types of fault propagation.

[0091] Optionally, the anomaly detection system extracts information about fault modes and propagation paths between entities from the power knowledge graph. This system constructs a directed graph with power equipment entities as nodes and fault propagation directions as directed edges. The attributes of each edge record the triggering conditions for fault propagation (e.g., voltage exceeding a threshold, abnormal temperature rise, etc.) and the type of impact (e.g., overload, short circuit, poor heat dissipation, etc.). This transforms the scattered fault association information in the power knowledge graph into a structured directed graph model, resulting in a directed graph of entity fault mode propagation.

[0092] Continuing with the example of a transformer A temperature anomaly, the power knowledge graph contains the following entities: transformer A, busbar, transformer B (another transformer in the same substation), and cooling fan. The failure modes and propagation paths between these entities are as follows:

[0093] Abnormal bus voltage → Transformer A overload (Trigger condition: Bus voltage falls 20% below rated value, Impact type: Overload). Transformer A overload → Temperature rise (Trigger condition: Overload lasts for more than 30 minutes, Impact type: Abnormal temperature).

[0094] Cooling fan failure → Transformer A temperature rise (Trigger condition: Fan outage, Impact type: Poor cooling). Transformer A temperature rise → Transformer B temperature rise (Trigger condition: Shared cooling system, Impact type: Heat conduction).

[0095] The abnormality detection system constructs an entity fault mode propagation directed graph according to the above information, the nodes are transformer A, bus, transformer B and cooling fan, and the directed edges include: bus → transformer A (trigger condition: bus voltage < 80% of rated value, influence type: overload), transformer A → itself (trigger condition: overload duration > 30 minutes, influence type: temperature anomaly), cooling fan → transformer A (trigger condition: fan shutdown, influence type: poor heat dissipation), and transformer A → transformer B (trigger condition: sharing heat dissipation system and temperature difference > 10℃, influence type: heat conduction).

[0096] In step 402, the second data features are converted into fault feature vectors corresponding to the entity attributes of the power knowledge graph, and a fault feature matrix is constructed based on the third data features. The vector elements in the fault feature vector represent parameter anomaly identifiers of the device operating state. The rows in the fault feature matrix represent different associated nodes at the same level, the columns represent parameter anomaly identifiers of the corresponding entity attributes, and each matrix element represents a fault correlation feature between the associated nodes at the same level.

[0097] Further, for the second data features of the upper-level node, the abnormality detection system converts them into fault feature vectors corresponding to the entity attributes of the power knowledge graph, where the vector elements are binary or numerical identifiers, indicating whether the device operating state parameters are abnormal (such as 1 indicating abnormality and 0 indicating normality, or a specific abnormality degree value). For the third data features of the associated nodes at the same level, a fault feature matrix is constructed, with rows corresponding to different associated nodes at the same level and columns corresponding to parameter anomaly identifiers of the entity attributes, and the matrix elements representing fault correlation features (such as similarity of abnormal features and correlation strength) between the nodes at the same level.

[0098] Continuing the above example, the upper-level node is “substation 1 data”, and its second data features include bus voltage data (voltage value is 70% of the rated value) and cooling fan operating state (fan shutdown). These features are converted into fault feature vectors: bus voltage anomaly identifier: 1 (because 70% < 80%, triggering the bus → transformer A fault propagation condition). Cooling fan fault identifier: 1 (fan shutdown). Other irrelevant attribute identifiers: 0.

[0099] Then the fault feature vector is [1, 0, 1, 0] (assuming the vector order is bus voltage, transformer A rated temperature, cooling fan state, and transformer B temperature).

[0100] The associated nodes at the same level are transformer B, and its third data feature is temperature rising to 80℃ (normal range 30℃-75℃). When constructing the fault feature matrix, the row corresponds to transformer B, the column corresponds to the transformer B temperature anomaly identifier, and the matrix element is the correlation strength between transformer B and transformer A temperature anomaly, which is obtained by calculating the temperature difference and historical data correlation, and the correlation strength is 0.8. Then the fault feature matrix is:

[0101] Same-layer associated nodes Transformer B temperature abnormality indicator Transformer B 0.8

[0102] Step 403 : Based on the fault feature vector, reasoning is performed in the entity fault mode propagation directed graph to determine the target fault mode triggered by the second data feature, and based on the fault feature matrix, the fault coupling relationship between the associated nodes at the same layer is analyzed to determine the coupling mode in which multiple nodes are abnormal at the same time.

[0103] Furthermore, based on the fault feature vectors, reasoning is performed within the entity fault mode propagation directed graph. This embodiment of the present invention employs forward or backward reasoning, starting from the existing anomaly features and following the propagation paths of directed edges to determine the target fault modes that may be triggered. Simultaneously, the fault feature matrix is ​​analyzed, and the fault coupling relationships between nodes associated with the same layer are determined based on the values ​​of the matrix elements. The coupling modes (e.g., series coupling, parallel coupling, mixed coupling, etc.) in which multiple nodes are simultaneously abnormal are determined.

[0104] Continuing with the above example, starting from the fault feature vector [1, 0, 1, 0], we infer in the entity fault mode propagation directed graph: bus voltage anomaly (1) triggers the bus → transformer A overload fault propagation path, and the target fault mode is "transformer A overload." Cooling fan fault (1) triggers the cooling fan → transformer A poor heat dissipation fault propagation path, and the target fault mode is "transformer A poor heat dissipation."

[0105] Furthermore, the anomaly detection system analyzed the fault feature matrix and found a correlation strength of 0.8 between the temperature anomalies of transformer B and transformer A. Combined with the heat conduction path from transformer A to transformer B in the directed graph (trigger condition: shared cooling system and temperature difference > 10°C; current transformer A temperature is 85°C, transformer B temperature is 80°C, temperature difference is 5°C, not meeting the trigger condition), the matrix showed a strong correlation, suggesting other coupling modes, such as synchronous overload caused by shared power supply.

[0106] Step 404 : Perform secondary abnormality determination based on the target fault mode and the coupling mode to determine a final abnormality determination result.

[0107] Furthermore, the anomaly detection system performs secondary anomaly determination based on the target fault mode and the coupling mode to determine a final anomaly determination result, as specifically described in the process from step 4041 to step 4045 .

[0108] The embodiments of the present invention achieve accurate tracing of abnormal data, and can clearly distinguish whether the anomaly is caused by upper-layer fault propagation, same-layer fault coupling, or multiple factors. It provides clear guidance for fault handling and system maintenance, thereby improving the interpretability of anomaly detection and the accuracy of fault location, reducing the troubleshooting time of operation and maintenance personnel, and improving the reliability and stability of the power system.

[0109] In one embodiment, steps 4041 to 4045 are described as follows:

[0110] Step 4041 : performing consistency comparison between the coupling mode type of each coupling mode and the failure mode types of the failure modes between entities in the power knowledge graph to determine valid coupling modes and invalid coupling modes.

[0111] Optionally, the anomaly detection system compares the type of each coupling mode (e.g., thermal conduction, electrical coupling, shared power supply, etc.) with the type of inter-entity failure modes in the power knowledge graph (e.g., overload, short circuit, poor heat dissipation, etc.) for consistency. If the coupling mode type is consistent with the inter-entity failure mode type recorded in the power knowledge graph, it is determined to be a valid coupling mode; if not, it is determined to be an invalid coupling mode. This consistency comparison can be achieved using a preset mode type mapping table or semantic similarity calculation.

[0112] Continuing with the transformer A temperature anomaly scenario, the coupling mode is "Transformer A and transformer B share a power source, resulting in a synchronous overload," and its type is "electrical coupling." The fault mode types between transformer entities in the power knowledge graph include "thermal conduction" and "electrical coupling." "Electrical coupling" is consistent with the fault mode type in the power knowledge graph, making this coupling mode valid. If another coupling mode exists, such as "Transformer A temperature anomaly affects transformer B via wireless signals," and this fault mode type is not found in the power knowledge graph, then the coupling mode is considered invalid.

[0113] Step 4042: Starting from the target fault pattern, a path search is performed in the entity fault pattern propagation directed graph to obtain the target propagation path from the entity corresponding to the upper-level node to the entity corresponding to the suspected abnormal data. The path satisfies the logical order of fault propagation.

[0114] Furthermore, the anomaly detection system uses the target fault pattern as a starting point and performs a path search (such as a depth-first search or breadth-first search) within the entity fault pattern propagation directed graph, looking for a path from the entity corresponding to the upper-level node to the entity corresponding to the suspected anomaly data. This path must satisfy the logical order of fault propagation, meaning that the fault pattern of the previous node triggers the anomaly of the next node. It should be noted that this embodiment of the present invention considers whether the triggering conditions of the edge meet the current data characteristics during the search.

[0115] Continuing with the above example, the target fault modes are "Transformer A overload" and "Transformer A poor heat dissipation." The upper-level nodes correspond to the busbar and cooling fan, and the entity corresponding to the suspected abnormal data is transformer A. Two target propagation paths are found in the directed graph: busbar (voltage abnormality) → transformer A (overload trigger, lasting for more than 30 minutes) → transformer A (temperature abnormality). Cooling fan (outage) → transformer A (poor heat dissipation) → transformer A (temperature abnormality). Both paths satisfy the fault propagation logic sequence and the trigger conditions (busbar voltage 70% < 80%, fan outage) are met.

[0116] Step 4043, inferring based on whether the suspected abnormal data belongs to an entity in the target propagation path to obtain a first inference result.

[0117] Furthermore, the anomaly detection system determines whether the entity corresponding to the suspected anomaly data belongs to an entity in the target propagation path. If so, the first inference result is "the anomaly may be caused by upper-layer fault propagation." If not, the first inference result is "the anomaly may be caused by other factors outside the target path."

[0118] Continuing with the above embodiment, the entity corresponding to the suspected abnormal data is transformer A. The end points of target propagation path 1 and path 2 are both transformer A. Therefore, transformer A belongs to the entity in the target propagation path, and the first inference result is "the anomaly may be caused by upper-layer fault propagation."

[0119] Step 4044, based on whether the suspected abnormal data has an associated fault mode with the node in the valid coupling mode, reasoning is performed to obtain a second reasoning result, and based on whether the suspected abnormal data has a non-standard association with the node in the invalid coupling mode, reasoning is performed to obtain a third reasoning result.

[0120] Furthermore, based on the second inference result, the anomaly detection system determines whether the entity corresponding to the suspected anomaly data has an associated fault mode with a node in the valid coupling mode. If so, the second inference result is "the anomaly may be caused by the valid coupling mode at the same layer." Otherwise, the second inference result is "the anomaly is not related to the valid coupling mode at the same layer."

[0121] Furthermore, regarding the third reasoning result, the anomaly detection system determines whether the entity corresponding to the suspected anomaly data has a non-standard association with a node in the invalid coupling pattern (i.e., an association not recorded in the power knowledge graph but potentially existing). If so, the third reasoning result is "the anomaly may be caused by a non-standard association"; otherwise, "the anomaly is not related to the invalid coupling pattern."

[0122] Continuing with the above example, let's assume the valid coupling mode is "Transformer A and Transformer B share a power supply, leading to synchronous overload." The node in the valid coupling mode is Transformer B. The suspected abnormal data corresponds to the shared power supply fault mode associated with Transformers A and B, so the second inference result is "The anomaly may be caused by the valid coupling mode at the same layer." For invalid coupling modes, such as "Radio signal impact," the nodes in the inference result are not actually associated, so the third inference result is "The anomaly is not related to the invalid coupling mode."

[0123] Step 4045 , performing a secondary abnormality determination based on the first inference result, the second inference result, and the third inference result of the suspected abnormal data to determine a final abnormality determination result.

[0124] Furthermore, the anomaly detection system performs secondary anomaly determination based on the first inference result, the second inference result, and the third inference result of the suspected abnormal data to determine the final anomaly determination result, as specifically described in the process from step 40451 to step 40456.

[0125] The embodiment of the present invention combines the fault logic of the knowledge graph and the backtracking results of the data binary tree to analyze data from multiple dimensions and multiple association perspectives, effectively avoiding misjudgment of normal data and omission of abnormal data, so that secondary abnormality judgment can accurately distinguish the source of the abnormality, clarify the influence of upper-layer propagation, same-layer coupling and non-standard factors, improve the accuracy and explainability of fault location, and further provide operation and maintenance personnel with clear fault handling priorities, such as giving priority to upper-layer bus and cooling fan faults, and at the same time troubleshooting same-layer power supply coupling problems, thereby effectively improving the operation and maintenance efficiency and stability of the power system.

[0126] In one embodiment, steps 40451 to 40456 are described as follows:

[0127] In step 40451, if the first reasoning result is that the entity belongs to the entity, or the second reasoning result is that there is an associated fault mode, or the third reasoning result is that there is a non-standard association, then the final abnormality judgment result is determined to be that the abnormality is caused by the upper-layer fault propagation alone, or by the same-layer effective fault coupling alone, or by the same-layer invalid fault coupling alone.

[0128] Optionally, when the first inference result indicates that the entity corresponding to the suspected abnormal data belongs to an entity in the target propagation path, or the second inference result indicates the presence of an associated fault mode, or the third inference result indicates the presence of a non-standard association, the anomaly detection system determines that the anomaly may be caused by a single factor. In this case, it is necessary to further determine, based on the different inference results, whether the anomaly is caused solely by upper-layer fault propagation, valid fault coupling at the same layer, or invalid fault coupling at the same layer.

[0129] Continuing with the above example, let's consider scenario 1: assuming the upper-level bus voltage is abnormal, the target propagation path is bus → transformer A, the first inference result is "belongs to entity," the second inference result is "no associated fault mode," and the third inference result is "no non-standard association." The final abnormality determination result is "upper-level fault propagation alone caused the abnormality," meaning the abnormal temperature of transformer A is caused by the abnormal bus voltage propagation.

[0130] Scenario 2: If a fault occurs at the associated node transformer B on the same layer and there is an effective coupling mode with transformer A (e.g., shared power supply), the first reasoning result is "not an entity," the second reasoning result is "an associated fault mode exists," and the third reasoning result is "no non-standard association exists." This results in the determination that "effective fault coupling on the same layer alone caused the anomaly," meaning that the temperature anomaly of transformer A was caused by the fault of transformer B on the same layer through the effective coupling mode.

[0131] Scenario 3: If an invalid coupling pattern exists (such as the hypothetical influence of wireless signals), the first reasoning result is "not an entity," the second reasoning result is "no associated fault pattern," and the third reasoning result is "non-standard association." This results in a determination of "invalid fault coupling at the same layer alone causing the anomaly." However, further verification of the rationality of this non-standard association requires expert knowledge.

[0132] In step 40452, if the first reasoning result is that the entity belongs, the second reasoning result is that there is an associated fault mode, and the third reasoning result is that there is no non-standard association, then the final abnormality judgment result is determined to be that the abnormality is caused by the coupling of upper-layer fault propagation and same-layer effective faults.

[0133] Furthermore, if the first inference result indicates that the entity corresponding to the suspected abnormal data belongs to an entity in the target propagation path, the second inference result indicates that an associated fault pattern exists, and the third inference result indicates that no non-standard association exists, the anomaly is determined to be caused by both upper-layer fault propagation and effective fault coupling at the same layer. In this case, both the upper-layer fault propagation path and the effective coupling pattern at the same layer have a direct impact on the occurrence of the anomaly.

[0134] Continuing with the above example, the first reasoning result is "belongs to an entity" (transformer A is in the target propagation path), the second reasoning result is "existence of an associated fault mode" (transformer A and transformer B have an effective coupling mode with a shared power source), and the third reasoning result is "no non-standard association." Therefore, the final abnormality determination result is "upper-layer fault propagation and same-layer effective fault coupling jointly cause the abnormality." This means that the temperature abnormality of transformer A is caused by the upper-layer fault propagation of the bus voltage abnormality and the cooling fan fault, as well as the effective fault coupling of the shared power source on the same layer.

[0135] In step 40453, if the first reasoning result is that the entity belongs, the second reasoning result is that there is no associated fault mode, and the third reasoning result is that there is a non-standard association, then the final abnormality judgment result is determined to be that the abnormality is caused by the coupling of upper-layer fault propagation and same-layer invalid fault.

[0136] Furthermore, if the first inference result indicates that the entity corresponding to the suspected anomaly data belongs to an entity in the target propagation path, the second inference result indicates that there is no associated fault mode, and the third inference result indicates that there is a non-standard association, then the anomaly is determined to be caused by both upper-layer fault propagation and invalid fault coupling at the same layer. In this case, upper-layer fault propagation is one of the main causes of the anomaly, and the presence of non-standard associations (i.e., invalid fault coupling at the same layer) not recorded in the knowledge graph also has a certain impact on the anomaly.

[0137] Continuing with the above example, assume that the upper-level node bus voltage is abnormal, the target propagation path is bus → transformer A, and the first reasoning result is "belongs to entity." There is no valid coupling pattern recorded in the knowledge graph between the associated nodes transformer B and transformer A on the same level (the second reasoning result is "no associated fault pattern"), but there is a new non-standard association not recorded in the knowledge graph (such as electromagnetic interference caused by nearby construction, the third reasoning result is "non-standard association exists"). The final anomaly determination result is "the anomaly is caused by the upper-level fault propagation and the invalid fault coupling on the same level." This means that the temperature anomaly of transformer A is caused by the upper-level fault propagation of the bus voltage anomaly and the invalid fault coupling on the same level of electromagnetic interference.

[0138] In step 40454, if the first reasoning result is that the entity belongs to the entity, the second reasoning result is that there is an associated fault mode, and the third reasoning result is that there is a non-standard association, then the final abnormality judgment result is determined to be that the abnormality is caused by the upper layer fault propagation, the same layer effective fault coupling and the same layer invalid fault coupling.

[0139] Furthermore, if the first inference result indicates that the entity corresponding to the suspected abnormal data belongs to an entity in the target propagation path, the second inference result indicates the presence of an associated fault mode, and the third inference result indicates the presence of a non-standard association, the anomaly is determined to be caused by a combination of upper-layer fault propagation, valid fault coupling at the same layer, and invalid fault coupling at the same layer. This situation is more complex, as the anomaly involves multiple levels and multiple types of fault factors.

[0140] Continuing with the above example, the upper-level node bus voltage is abnormal and the cooling fan is faulty. The target propagation path is bus → transformer A and cooling fan → transformer A. The first reasoning result is "belongs to an entity." The same-level associated nodes, transformer B and transformer A, have an effective coupling mode with a shared power supply (the second reasoning result is "there is an associated fault mode"), and there is also a non-standard association not recorded in the knowledge graph (such as a sudden increase in ambient temperature, resulting in the third reasoning result being "there is a non-standard association"). The final anomaly determination result is "the anomaly is caused by the combined effects of upper-level fault propagation, effective fault coupling at the same level, and invalid fault coupling at the same level." This means that the temperature anomaly of transformer A is caused by the upper-level fault propagation of the bus voltage anomaly and the cooling fan fault, the effective fault coupling of the shared power supply at the same level, and the invalid fault coupling of the increased ambient temperature.

[0141] In step 40455, if the first reasoning result is that it does not belong to an entity, the second reasoning result is that there is an associated fault mode, and the third reasoning result is that there is a non-standard association, then the final abnormality judgment result is determined to be that the abnormality is caused by the same-layer valid fault coupling and the same-layer invalid fault coupling.

[0142] Furthermore, if the first inference result indicates that the entity corresponding to the suspected abnormal data does not belong to the entity in the target propagation path, the second inference result indicates the presence of an associated fault mode, and the third inference result indicates the presence of a non-standard association, the anomaly is determined to be caused by both valid and invalid fault coupling at the same layer. In this case, the anomaly is not related to the propagation of the upper layer fault but is the result of the coupling of different types of faults at the same layer.

[0143] Continuing with the above example, if the upper-level node data is normal and the target propagation path does not include transformer A, the first reasoning result is "not an entity." Transformer B, a node associated with the same level, has an effective coupling mode with transformer A, sharing a common power supply (the second reasoning result is "an associated fault mode exists"), and also has a non-standard association due to a sudden increase in ambient temperature (the third reasoning result is "a non-standard association exists"). The final anomaly determination is "both effective and invalid fault coupling at the same level cause the anomaly," meaning the temperature anomaly of transformer A is caused by both effective fault coupling at the same level, sharing a common power supply, and invalid fault coupling due to a rise in ambient temperature.

[0144] Step 40456: If the first reasoning result is that it does not belong to an entity, the second reasoning result is that there is no associated fault mode, and the third reasoning result is that there is no non-standard association, then the final abnormality determination result is determined to be a misjudgment abnormality.

[0145] Furthermore, if the first inference result indicates that the entity corresponding to the suspected anomaly data does not belong to the entity in the target propagation path, the second inference result indicates that there is no associated fault mode, and the third inference result indicates that there is no non-standard association, the anomaly is determined to be a false positive. This means that the existing knowledge graph and data feature analysis cannot find a reasonable cause for the anomaly, and the preliminary anomaly characteristics may be caused by non-fault factors (such as data collection errors, temporary interference, etc.).

[0146] Continuing with the above example, if the upper-level node data is normal and the target propagation path does not include transformer A, the first reasoning result is "not an entity." There is no valid coupling pattern between transformer B and transformer A (the second reasoning result is "no associated fault pattern"), nor is there a non-standard association (the third reasoning result is "no non-standard association"). The final abnormality determination result is "falsely diagnosed abnormality," indicating that the initial abnormal characteristics of transformer A's temperature data are likely due to a temporary sensor failure or data transmission error, not a true equipment failure.

[0147] The embodiment of the present invention performs secondary anomaly determination in coordination with the inference results of suspected abnormal data on upper-layer fault propagation, same-layer effective fault coupling, and same-layer invalid fault coupling. This can more accurately determine the true cause of the anomaly, clearly distinguish between different situations such as upper-layer fault propagation, same-layer effective fault coupling, same-layer invalid fault coupling, and misjudgment of anomalies, and improve the accuracy and explainability of fault location.

[0148] Furthermore, the power data processing system based on the knowledge graph provided by the present invention is described below. The power data processing system based on the knowledge graph described below and the power data processing method based on the knowledge graph described above can be referenced to each other.

[0149] Optional, see Figure 2 , Figure 2 This is a structural diagram of the power data processing system based on the knowledge graph provided by the present invention. The power data processing system based on the knowledge graph includes:

[0150] The data processing module 210 is used to establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records the data source information and the corresponding data;

[0151] The first anomaly determination module 220 is configured to perform hierarchical extraction from the root node of the data source binary tree in hierarchical order to obtain the first data feature of each data source, and perform preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected anomaly data;

[0152] The abnormal data backtracking module 230 is used to backtrack from the node corresponding to the suspected abnormal data in the data source binary tree based on the data logical relationship, and determine the upper layer node and the same layer related nodes of the suspected abnormal data;

[0153] The second abnormality determination module 240 is used to perform secondary abnormality determination based on the second data feature of the upper-level node and the third data feature of the associated node at the same level, and the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph to determine the final abnormality determination result.

[0154] The embodiment of the present invention uses a knowledge graph in the power field that contains rich equipment operation rules and fault characteristics. It can flexibly adjust the basis for abnormality judgment according to different operation scenarios, fault modes and fault propagation paths, thereby improving adaptability to complex scenarios. At the same time, by establishing a binary tree of data sources to sort out the hierarchical relationship of data, in the process of feature extraction and backtracking analysis, it deeply mines the correlation characteristics between data from the same source and different sources, and can more accurately analyze the complex connections between data. In the secondary judgment link, combined with the fault logic of the knowledge graph and the backtracking results of the data binary tree, the data is analyzed from multiple dimensions and multiple correlation perspectives, effectively avoiding the misjudgment of normal data and the omission of abnormal data, and improving the accuracy and reliability of power data anomaly detection.

[0155] See also Figure 3 , Figure 3 This is a diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, the following steps are implemented:

[0156] A data source binary tree is established based on the data source of the collected power data, and the power data is associated and mapped with entities in the pre-built power knowledge graph; each node in the data source binary tree records the data source information and the corresponding data;

[0157] Perform hierarchical extraction from the root node of the data source binary tree in hierarchical order to obtain the first data feature under each data source. Then, perform preliminary anomaly determination based on the first data feature and the entity attributes of the entity corresponding to each data source in the power knowledge graph to determine suspected abnormal data.

[0158] Based on the data logic relationship, trace back from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper-level node and the related nodes at the same level of the suspected abnormal data;

[0159] Based on the second data feature of the upper-level node and the third data feature of the associated node at the same level, a secondary anomaly judgment is performed on the association relationship between the fault mode and the fault propagation path between entities in the power knowledge graph to determine the final anomaly judgment result.

[0160] See also Figure 4 , Figure 4 Detailed description of an embodiment of a computer-readable storage medium provided by an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, the following steps are implemented:

[0161] A data source binary tree is established based on the data source of the collected power data, and the power data is associated and mapped with entities in the pre-built power knowledge graph; each node in the data source binary tree records the data source information and the corresponding data;

[0162] Perform hierarchical extraction from the root node of the binary tree of data sources in hierarchical order to obtain the first data feature of each data source. Then, perform preliminary anomaly determination based on the first data feature and the corresponding entity attributes of each data source in the power knowledge graph to identify suspected abnormal data.

[0163] Based on the data logic relationship, trace back from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper-level node and the related nodes at the same level of the suspected abnormal data;

[0164] Based on the second data feature of the upper-level node and the third data feature of the associated node at the same level, a secondary anomaly judgment is performed on the association relationship between the fault mode and the fault propagation path between entities in the power knowledge graph to determine the final anomaly judgment result.

[0165] On the other hand, the present invention further provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the power data processing method based on the knowledge graph provided by the above methods, which includes:

[0166] A data source binary tree is established based on the data source of the collected power data, and the power data is associated and mapped with entities in the pre-built power knowledge graph; each node in the data source binary tree records the data source information and the corresponding data;

[0167] Perform hierarchical extraction from the root node of the binary tree of data sources in hierarchical order to obtain the first data feature of each data source. Then, perform preliminary anomaly determination based on the first data feature and the corresponding entity attributes of each data source in the power knowledge graph to identify suspected abnormal data.

[0168] Based on the data logic relationship, trace back from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper-level node and the related nodes at the same level of the suspected abnormal data;

[0169] Based on the second data feature of the upper-level node and the third data feature of the associated node at the same level, a secondary anomaly judgment is performed on the association relationship between the fault mode and the fault propagation path between entities in the power knowledge graph to determine the final anomaly judgment result.

[0170] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0171] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A power data processing method based on knowledge graph, characterized in that: include: Establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records data source information and corresponding data; Performing hierarchical extraction from the root node of the binary tree of the data sources in hierarchical order to obtain a first data feature under each data source, and performing preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data; Based on the data logical relationship, backtrack from the node corresponding to the suspected abnormal data in the data source binary tree to determine the upper layer node and the same layer associated node of the suspected abnormal data; Based on the second data feature of the upper-layer node and the third data feature of the associated node at the same layer, and the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph, a secondary abnormality determination is performed to determine a final abnormality determination result; The steps of determining the final abnormality determination result include: The fault modes and propagation paths between entities in the power knowledge graph are used to construct an entity fault mode propagation directed graph; each node in the entity fault mode propagation directed graph represents a power equipment entity, the edge represents the fault propagation direction, and the edge attributes record the triggering conditions and impact types of the fault propagation; The second data feature is converted into a fault feature vector corresponding to the entity attribute of the power knowledge graph, and a fault feature matrix is ​​constructed based on the third data feature; the vector elements in the fault feature vector represent parameter abnormality identifiers of the equipment operation status; the rows in the fault feature matrix represent different same-layer associated nodes, the columns represent parameter abnormality identifiers of the corresponding entity attributes, and each matrix element represents the fault correlation feature between the same-layer associated nodes; Reasoning in the entity fault mode propagation directed graph based on the fault feature vector determines the target fault mode triggered by the second data feature, and analyzing the fault coupling relationship between nodes associated with the same layer based on the fault feature matrix to determine a coupling mode in which multiple nodes are simultaneously abnormal; Performing a secondary abnormality determination based on the target fault mode and the coupling mode, and determining the final abnormality determination result; The performing secondary abnormality determination based on the target fault mode and the coupling mode to determine the final abnormality determination result includes: Based on the consistency comparison between the coupling mode type of each coupling mode and the failure mode type of the failure mode between entities in the power knowledge graph, a valid coupling mode and an invalid coupling mode are determined; Taking the target fault pattern as a starting point, a path search is performed in the entity fault pattern propagation directed graph to obtain a target propagation path from an entity corresponding to an upper-level node to an entity corresponding to suspected abnormal data; the path satisfies the logical order of fault propagation; Reasoning based on whether the suspected abnormal data belongs to an entity in the target propagation path, obtaining a first reasoning result; Reasoning based on whether the suspected abnormal data has an associated fault mode with the node in the valid coupling mode to obtain a second reasoning result, and reasoning based on whether the suspected abnormal data has a non-standard association with the node in the invalid coupling mode to obtain a third reasoning result; A secondary abnormality determination is performed based on the first inference result, the second inference result, and the third inference result of the suspected abnormal data to determine the final abnormality determination result.

2. The power data processing method based on knowledge graph according to claim 1 is characterized in that: The performing secondary abnormality determination based on the first reasoning result, the second reasoning result, and the third reasoning result of the suspected abnormal data to determine the final abnormality determination result includes: If the first reasoning result is that the entity belongs to the entity, or the second reasoning result is that there is an associated fault mode, or the third reasoning result is that there is a non-standard association, then the final abnormality determination result is determined to be that the abnormality is caused by upper-layer fault propagation alone, or by same-layer valid fault coupling alone, or by same-layer invalid fault coupling alone. If the first reasoning result is that the entity belongs, the second reasoning result is that there is an associated fault mode, and the third reasoning result is that there is no non-standard association, then it is determined that the final abnormality determination result is that the abnormality is caused by the coupling of upper-layer fault propagation and same-layer effective faults; If the first reasoning result is that the entity belongs, and the second reasoning result is that there is no associated fault mode, and the third reasoning result is that there is a non-standard association, then it is determined that the final abnormality determination result is that the abnormality is caused by the coupling of upper-layer fault propagation and same-layer invalid fault; If the first reasoning result indicates that the entity belongs to the entity, the second reasoning result indicates that an associated fault mode exists, and the third reasoning result indicates that a non-standard association exists, then determining that the final abnormality determination result is that the abnormality is caused by upper-layer fault propagation, same-layer valid fault coupling, and same-layer invalid fault coupling. If the first reasoning result is that the entity does not exist, the second reasoning result is that an associated fault mode exists, and the third reasoning result is that a non-standard association exists, then it is determined that the final abnormality determination result is that the abnormality is caused by both the same-layer valid fault coupling and the same-layer invalid fault coupling; If the first reasoning result is that it does not belong to an entity, and the second reasoning result is that there is no associated fault mode, and the third reasoning result is that there is no non-standard association, then it is determined that the final abnormality determination result is a misjudgment abnormality.

3. The power data processing method based on knowledge graph according to claim 1 or 2, characterized in that: The performing preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data includes: Based on the normal operating state characteristics and fault case characteristic patterns of the entities corresponding to each data source in the power knowledge graph, combined with the first data characteristics of each data source, a three-dimensional association network is constructed; each network node in the three-dimensional association network represents various characteristics, and if there is a direct or indirect semantic association path between the first data feature and the normal operating state characteristics or the fault case characteristic pattern in the power knowledge graph, an edge connection is established between the corresponding network nodes; Determining a feature deviation based on the three-dimensional association network, based on a semantic path length of the first data feature and a target normal operating state feature to which the feature is connected in the power knowledge graph, and a feature value of the first data feature and a feature value of the target normal operating state feature; Determining a feature matching degree based on the three-dimensional association network, based on the semantic similarity between the first data feature and the target fault case feature pattern to which it is connected, and the feature value of the first data feature and the feature value of the target fault case feature pattern; A preliminary abnormality determination is performed based on the feature deviation and feature matching of each first data feature to determine the suspected abnormal data.

4. The power data processing method based on knowledge graph according to claim 3 is characterized in that: The performing preliminary abnormality determination based on the feature deviation and feature matching of each first data feature to determine the suspected abnormal data includes: Determining a first data feature whose feature deviation is greater than a deviation threshold and whose feature matching is greater than a matching threshold as a preliminary abnormal feature, and constructing an abnormal feature association subgraph based on the association relationship between the preliminary abnormal features in the three-dimensional association network; Determining the node degree of each preliminary abnormal feature based on the abnormal feature association subgraph, and determining the abnormal path dependency based on the first path number of paths between any two nodes in the abnormal feature association subgraph and the second path number of paths passing through the node corresponding to the preliminary abnormal feature; Determining the number of neighbor nodes and the number of node edges between neighbor nodes of each preliminary abnormal feature based on the abnormal feature association subgraph, and determining a clustering coefficient based on the number of neighbor nodes and the number of node edges; A preliminary abnormality judgment is performed based on the node degree, abnormal path dependency, clustering coefficient, feature deviation and feature matching of each preliminary abnormal feature to determine the suspected abnormal data.

5. The power data processing method based on knowledge graph according to claim 4 is characterized in that: The preliminary abnormality determination is performed based on the node degree, abnormal path dependency, clustering coefficient, feature deviation and feature matching of each preliminary abnormal feature to determine the suspected abnormal data, including: Determine the data corresponding to the preliminary abnormal features with a node degree greater than or equal to a preset threshold as the suspected abnormal data; or If the node degree of each preliminary abnormal feature is less than a preset threshold, then the abnormal betweenness value is determined based on the abnormal path dependency and clustering coefficient of each preliminary abnormal feature; or, Determine the data corresponding to the preliminary abnormal features whose abnormal betweenness value is greater than or equal to the preset abnormal threshold as the suspected abnormal data; or If the anomaly betweenness value of each preliminary anomaly feature is less than the preset anomaly threshold, the anomaly propagation intensity is determined based on the feature deviation and feature matching of each preliminary anomaly feature; The data corresponding to the preliminary abnormal feature with an abnormal propagation intensity greater than or equal to a preset propagation intensity threshold is determined as the suspected abnormal data.

6. A power data processing system based on knowledge graph, characterized in that: Applicable to the power data processing method based on knowledge graph as described in any one of claims 1 to 5; The power data processing system based on knowledge graph includes: A data processing module is used to establish a data source binary tree based on the data source of the collected power data, and associate and map the power data with entities in the pre-built power knowledge graph; each node in the data source binary tree records data source information and corresponding data; a first anomaly determination module, configured to perform hierarchical extraction from the root node of the binary tree of the data sources in hierarchical order to obtain a first data feature under each data source, and perform preliminary anomaly determination based on the first data feature and the entity attributes corresponding to each data source in the power knowledge graph to determine suspected abnormal data; An abnormal data backtracking module is used to backtrack from the node corresponding to the suspected abnormal data in the data source binary tree based on the data logical relationship, and determine the upper layer node and the same layer associated node of the suspected abnormal data; A second anomaly determination module is configured to perform a secondary anomaly determination based on the second data feature of the upper-layer node and the third data feature of the associated node at the same layer, and the association relationship between the fault mode and the fault propagation path between the entities in the power knowledge graph, and determine a final anomaly determination result; The steps of determining the final abnormality determination result include: The fault modes and propagation paths between entities in the power knowledge graph are used to construct an entity fault mode propagation directed graph; each node in the entity fault mode propagation directed graph represents a power equipment entity, the edge represents the fault propagation direction, and the edge attributes record the triggering conditions and impact types of the fault propagation; The second data feature is converted into a fault feature vector corresponding to the entity attribute of the power knowledge graph, and a fault feature matrix is ​​constructed based on the third data feature; the vector elements in the fault feature vector represent parameter abnormality identifiers of the equipment operation status; the rows in the fault feature matrix represent different same-layer associated nodes, the columns represent parameter abnormality identifiers of the corresponding entity attributes, and each matrix element represents the fault correlation feature between the same-layer associated nodes; Reasoning in the entity fault mode propagation directed graph based on the fault feature vector determines the target fault mode triggered by the second data feature, and analyzing the fault coupling relationship between nodes associated with the same layer based on the fault feature matrix to determine a coupling mode in which multiple nodes are simultaneously abnormal; Performing a secondary abnormality determination based on the target fault mode and the coupling mode, and determining the final abnormality determination result; The performing secondary abnormality determination based on the target fault mode and the coupling mode to determine the final abnormality determination result includes: performing consistency comparison between the coupling mode type of each coupling mode and the failure mode types of the failure modes between entities in the power knowledge graph to determine a valid coupling mode and an invalid coupling mode; Taking the target fault pattern as a starting point, a path search is performed in the entity fault pattern propagation directed graph to obtain a target propagation path from an entity corresponding to an upper-level node to an entity corresponding to suspected abnormal data; the path satisfies the logical order of fault propagation; Reasoning based on whether the suspected abnormal data belongs to an entity in the target propagation path, obtaining a first reasoning result; Reasoning based on whether the suspected abnormal data has an associated fault mode with the node in the valid coupling mode to obtain a second reasoning result, and reasoning based on whether the suspected abnormal data has a non-standard association with the node in the invalid coupling mode to obtain a third reasoning result; A secondary abnormality determination is performed based on the first inference result, the second inference result, and the third inference result of the suspected abnormal data to determine the final abnormality determination result.

7. An electronic device comprising: Memory for storing computer software programs; A processor for reading and executing the computer software program, characterized in that when the processor executes the computer software program, it implements the power data processing method based on the knowledge graph as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium storing a computer software program, characterized in that: When the computer software program is executed by a processor, it implements the power data processing method based on the knowledge graph as described in any one of claims 1 to 5.

Images