A substation knowledge graph construction and optimization method based on multi-view learning

Abstract

The present invention relates to the technical field of knowledge graph construction, and discloses a method for constructing and optimizing a knowledge graph for a substation based on multi-view learning, comprising the following steps: processing multi-source data of the substation through a heterogeneous model, mapping the multi-source heterogeneous data of operation and maintenance texts, monitoring data, procedures and rules, and infrared images of substation equipment to a unified feature space through linear projection, constructing multimodal positive and negative sample pairs of the same equipment, using contrastive learning to shorten the distance between related equipment features, dynamically optimizing the projection matrix of various types of data, jointly detecting entity boundaries of substation equipment in operation and maintenance texts and equipment monitoring data, fusing multimodal equipment features through an attention mechanism, and inferring the relationship types between equipment. The present invention effectively solves the fragmentation and static problems existing in traditional knowledge management systems by deeply integrating multi-source heterogeneous data of substations with expert experience, and significantly improves the accuracy of power equipment status perception and fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph construction, and more specifically, to a substation knowledge graph construction and optimization method based on multi-view learning. Background Art

[0002] With the acceleration of smart grid construction, modern substations have formed a multimodal data ecosystem that includes SCADA real-time monitoring, unstructured operation and maintenance documents, infrared images, etc.

[0003] The traditional knowledge graph construction method based on a single view has defects such as a single feature extraction dimension and insufficient cross-modal semantic association, resulting in a high missing rate of relationship types in the constructed graph, which seriously restricts its application efficiency in scenarios such as fault tracing and intelligent inspection. It is urgent to break through the technical bottleneck of collaborative processing of multi-source heterogeneous data. Summary of the Invention

[0004] The present invention provides a substation knowledge graph construction and optimization method based on multi-view learning, which solves the technical problems in related technologies that multimodal knowledge graph construction cannot be achieved and conflict judgment between the rule base and the knowledge graph cannot be solved.

[0005] The present invention provides a substation knowledge graph construction and optimization method based on multi-view learning, including:

[0006] S100, multi-view feature extraction: processing substation multi-source data through heterogeneous models;

[0007] S200, Substation Multi-Source Data Alignment: This system uses linear projection to map heterogeneous data from substation equipment, including maintenance documents, monitoring data, procedures, and infrared images, into a unified feature space. This system constructs multimodal positive and negative sample pairs for the same equipment, employs contrastive learning to narrow the distance between related equipment features, and dynamically optimizes the projection matrix for each type of data, enabling collaborative representation of substation multi-source data.

[0008] S300, Substation Knowledge Triple Extraction: This system jointly detects substation equipment entity boundaries in operation and maintenance documents and equipment monitoring data, fuses multimodal equipment features through an attention mechanism, infers the relationship types between devices, and uses a dynamic threshold to filter low-confidence triplets to output structured knowledge units that meet substation domain constraints.

[0009] S400, dynamic optimization of substation knowledge graph: clusters similar substation equipment entities to eliminate redundancy, adjusts relationship weights based on the density of associated knowledge between devices, applies time decay to historical operation and maintenance relationships, eliminates conflicts between new and old operation and maintenance knowledge, and optimizes the substation graph topology by minimizing structural entropy;

[0010] S500, Substation Operation and Maintenance Application Generation: This process parses the natural language requirements of operators into substation knowledge graph query patterns, performs multi-path reasoning under the constraints of equipment safety and operating procedures, constructs a candidate operation and maintenance strategy set that includes equipment safety, economy, and reliability assessments, and ultimately generates an executable operation and maintenance instruction sequence that complies with power equipment operating specifications.

[0011] S600, substation operation and maintenance rules and data optimization: By real-time monitoring of deviations between substation equipment operating status and abnormalities triggered by operation and maintenance rules, the system automatically quantifies the degree of rule conflicts, dynamically adjusts equipment operating threshold parameters, and attenuates the weight of conflicting rules. After verification through online AB testing, the optimized operation and maintenance rules are injected into the substation knowledge graph, forming a closed loop of self-evolution of substation operation and maintenance rules.

[0012] Furthermore, S100 specifically includes the following steps:

[0013] S110, Text Modal Feature Extraction: Semantic encoding of unstructured text data;

[0014] S120, time series data feature extraction: performing spatiotemporal feature learning on the equipment monitoring sequences collected by the substation SCADA system;

[0015] S130, expert rule graph feature extraction: construct substation equipment operation and maintenance procedures and expert experience rules into a rule association graph and generate node embedding;

[0016] S140, visual data feature extraction: processes infrared thermal imaging of substation equipment and outputs substation infrared feature vectors;

[0017] Furthermore, in multi-view feature extraction, processing heterogeneous data from multiple sources of substation data includes using BERT to extract semantic features of substation operation and maintenance work orders and fault records, TCN networks to capture the time series patterns of monitoring data of transformers and switchgear, GAT to aggregate the correlation features between substation equipment operation and maintenance procedures and expert experience rules, and 3DResNet to extract visual features of infrared thermal imaging of substation equipment;

[0018] Each modality feature is coded independently to preserve view specificity, forming four types of feature representations including operation and maintenance text, equipment monitoring, procedures and rules, and equipment thermal images.

[0019] Furthermore, S200 specifically includes the following steps:

[0020] S210, unified representation of substation data: mapping multi-source features such as operation and maintenance text, monitoring data, procedures and rules, and infrared images into a unified dimensional space;

[0021] S220, substation equipment sample pair construction: establishing multi-source data sample pairs based on substation equipment numbers;

[0022] S230, Substation Equipment Feature Comparative Learning: Calculates multi-source data contrast loss to enhance the similarity of features of the same equipment;

[0023] S240, Substation Data Projection Optimization: Optimizes the projection matrix of various data types through backpropagation.

[0024] Furthermore, S300 specifically includes the following steps:

[0025] S310, cross-modal entity boundary detection of substation equipment: jointly detect the start / end position of substation equipment entities from multimodal features such as operation and maintenance text, equipment monitoring data, and infrared images;

[0026] S320, multimodal feature fusion of substation equipment: Aggregates cross-modal features such as operation and maintenance text, equipment monitoring data, and infrared images to generate substation equipment entity embeddings;

[0027] S330, power substation equipment relationship reasoning: reasoning about the relationship type between equipment based on the features of power substation equipment entity pairs;

[0028] S340, substation knowledge triple screening: adaptively filtering substation equipment relationship triples with low confidence;

[0029] Its effective substation knowledge triples are as follows:

[0030] ;

[0031] ;

[0032] in, is the mean confidence value of the substation equipment relationship, is the confidence standard deviation, For substation equipment The number of historical occurrences of is the total number of triples in the knowledge graph, represents a valid set of substation knowledge triples, represents the dynamic threshold, Represents the relationship between substation equipment The prediction confidence of , are the characteristic vectors of substation entities i and j, Indicates the relationship type between substation equipment. Represents a triplet of substation equipment relations.

[0033] Furthermore, S400 specifically includes the following steps:

[0034] S410, substation equipment entity similarity clustering: calculating the semantic similarity between substation equipment entity embeddings and merging redundant equipment entities;

[0035] S420, edge weight update guided by the associated knowledge density of substation equipment: dynamically adjust the relationship weight based on the embedding density of the associated subgraph between devices;

[0036] S430, substation operation and maintenance relationship time decay: apply exponential decay to historical operation and maintenance relationships;

[0037] S440, substation operation and maintenance knowledge conflict resolution: resolve conflicts between newly added operation and maintenance triples and existing operation and maintenance graphs;

[0038] S450, Substation Knowledge Graph Topology Entropy Optimization: Minimizing graph structure entropy to improve equipment association query efficiency;

[0039] The calculation formula for structural entropy optimization is as follows:

[0040] ;

[0041] Optimization goal:

[0042] ;

[0043] in is the degree of node i, is the total number of edges, is the number of nodes, is the regularization coefficient, are model parameters, represents the structural entropy of the graph, For the substation knowledge graph, is the L2 norm of the model parameters, represents the minimization operation on the substation knowledge graph G, is the normalized connectivity of node i.

[0044] Furthermore, the calculation formula for edge weight update guided by the associated knowledge density of substation equipment is as follows:

[0045] ;

[0046] ;

[0047] in For the edge The original weight of is the updated weight, is the knowledge density of the neighborhood of node i, represents the knowledge density of the neighborhood of node j, is the neighbor set of node i, represents the entity embedding of node i, represents the entity embedding of node k;

[0048] The calculation formula for the timeliness attenuation of substation operation and maintenance relationship is as follows:

[0049] ;

[0050] ;

[0051] in, is the initial weight, Build time for relationships, is the half-life, is the attenuation rate, represents the relationship weight at time t, is the logarithm with base 2;

[0052] The process of resolving substation operation and maintenance knowledge conflicts is as follows:

[0053] when When replacing old operation and maintenance knowledge, When the system is completed, it will be retained in parallel for review by operation and maintenance experts;

[0054] in is the confidence of the new operation and maintenance triplet, is the confidence of the old operation and maintenance triplet, the confidence difference threshold of operation and maintenance knowledge is 0.2, and the confidence ratio threshold is 2.

[0055] Furthermore, S500 specifically includes the following steps:

[0056] S510, graphical analysis of substation operation and maintenance requirements: converts the natural language requirements of operation and maintenance personnel into a substation knowledge graph query model;

[0057] S520, multi-constraint path reasoning for substation equipment: searching for a reasoning path that satisfies equipment safety constraints and operation timing constraints in the optimized substation knowledge graph;

[0058] S530, Substation Operation and Maintenance Strategy Utility Function Modeling: Quantitatively evaluate the expected utility of each candidate operation and maintenance strategy;

[0059] S540, Substation Operation and Maintenance Strategy Sequence Optimization Generation: Generates the optimal operation and maintenance strategy sequence based on reinforcement learning;

[0060] S550, substation operation and maintenance plan compilation: converts the operation and maintenance strategy into an instruction set that complies with the operating specifications of power equipment.

[0061] Furthermore, the calculation formula for the optimal strategy sequence is as follows:

[0062] ;

[0063] ;

[0064] in is the policy function, is the discount factor at time t, is the state at time t, is the action at time t, For instant rewards, represents the time step, represents the indicator function, represents the expectation operation, represents the optimized substation knowledge graph, It refers to the operation and maintenance strategy that expects the maximum value among all possible operation and maintenance strategies. Refers to the set of all possible operation and maintenance strategies;

[0065] The calculation formula for compiling the operation and maintenance plan is as follows:

[0066] ;

[0067] Satisfy the constraints:

[0068] ;

[0069] in For the final substation operation and maintenance sequence, It is a set of operation and maintenance strategies, including multiple optional disposal solutions. Timestamp of the operation. A topological sorting algorithm considering equipment operation safety constraints.

[0070] Furthermore, when the performance of any substation operation and maintenance rules in actual equipment operation data deviates significantly from the built-in confidence level of the map, an optimization process is automatically triggered. The steps of the optimization process are as follows:

[0071] S610, quantification of the conflict degree of substation operation and maintenance rules: Calculate the deviation between the operation and maintenance rules and the distribution of equipment operation data;

[0072] S620, dynamic optimization of substation equipment operation and maintenance thresholds: adjusts the triggering thresholds of equipment operation and maintenance rules;

[0073] S630, Substation operation and maintenance rule weight decay: Update the influence weight of the operation and maintenance rules in the substation knowledge graph;

[0074] S640, online AB testing and verification of substation operation and maintenance rules: running new and old operation and maintenance rule versions in parallel to evaluate equipment operation performance improvements;

[0075] S650, hot update of substation knowledge graph: injecting optimized operation and maintenance rule parameters.

[0076] The beneficial effects of the present invention are:

[0077] This invention effectively solves the fragmentation and static problems of traditional knowledge management systems by deeply integrating multi-source heterogeneous data from substations with expert experience, and significantly improves the accuracy of power equipment status perception and fault diagnosis.

[0078] At the same time, the present invention also realizes cross-modal semantic association and dynamic knowledge evolution, so that the operation and maintenance strategy generation has multi-dimensional reasoning capabilities, greatly enhancing the decision-making reliability under complex working conditions. The adaptive optimization mechanism ensures the continuous coordination between the knowledge system and actual operation, and provides self-evolutionary cognitive support for the intelligent operation and maintenance of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 This is a flowchart of a substation knowledge graph construction and optimization method based on multi-view learning proposed by the present invention;

[0080] Figure 2 The present invention Figure 1 Flowchart of sub-steps of S100;

[0081] Figure 3 The present invention Figure 1 Flowchart of sub-steps of S200;

[0082] Figure 4 The present invention Figure 1 Flowchart of sub-steps of S300;

[0083] Figure 5 The present invention Figure 1 Flowchart of sub-steps of S400;

[0084] Figure 6 The present invention Figure 1 Flowchart of sub-steps of S500;

[0085] Figure 7 The present invention Figure 1 Flowchart of sub-steps of S600. DETAILED DESCRIPTION

[0086] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0087] like Figure 1-Figure 7 As shown in FIG, a method for constructing and optimizing a substation knowledge graph based on multi-view learning includes the following steps:

[0088] S100, multi-view feature extraction: BERT is used to extract semantic features from substation operation and maintenance work orders and fault records. The TCN network captures the temporal patterns of monitoring data from equipment such as transformers and switches. GAT aggregates the correlation features between substation equipment operation and maintenance procedures and expert experience rules. 3DResNet extracts visual features from infrared thermal imaging of substation equipment.

[0089] Each modality feature is coded independently to preserve view specificity, forming four types of feature representations including operation and maintenance text, equipment monitoring, procedures and rules, and equipment thermal images;

[0090] In one embodiment of the present invention, the following steps are specifically included:

[0091] S110, text modal feature extraction: for unstructured text data (substation equipment fault records , Operation and maintenance inspection report ) for semantic encoding;

[0092] The calculation formula is as follows:

[0093] ;

[0094] ;

[0095] in, ( The length of the substation fault record text is usually 500 to 1000 words. is the size of the substation professional vocabulary, about 20,000 words). is the 12-layer Transformer parameter, is the substation text feature vector, with a dimension of 768, represents the 12-layer substation text query matrix, Represents a 12-layer substation text key matrix, Represents a 12-layer substation text value matrix, Indicates taking the feature vector corresponding to the substation text CLStoken, is the pre-trained BERT model with parameters , used to extract semantic features of substation text;

[0096] S120, Time Series Data Feature Extraction: Equipment monitoring sequences collected by the substation SCADA system (T = sampling time step, usually 1 second, d = substation monitoring dimension, including 15-20 key indicators such as transformer oil temperature, load factor, switch status, etc.) to conduct spatiotemporal feature learning;

[0097] The calculation formula is as follows:

[0098] ;

[0099] ;

[0100] in, is the convolution kernel size of substation time series data, is the substation data expansion rate series (corresponding to the time scale of 1 second, 2 seconds, and 4 seconds), is the substation data sampling step (usually 60 seconds), is the substation timing network parameter, is the substation time series feature matrix, with dimension , Indicates the input substation monitoring data, represents a one-dimensional dilated convolution operation, represents the monitoring time step after downsampling, represents the original monitoring time step, is a temporal convolutional network with parameters , used to extract the time series features of substation monitoring data;

[0101] S130, expert rule graph feature extraction: Substation equipment operation and maintenance procedures and expert experience rules (Including equipment inspection standards, fault diagnosis rules, operating procedures, etc., usually n=500, 1000 rules) constructed as a rule association graph , and generate node embeddings;

[0102] The calculation formula is as follows:

[0103] ;

[0104] ;

[0105] in, is the number of substation operation and maintenance rules, is the substation rule adjacency matrix, is the network parameter of the substation rule diagram, is the attention coefficient between substation rules, is the weight matrix of the h-th regular attention head, is the feature of rule node j, is the ReLU activation function, is the characteristic matrix of the substation rule graph, with dimension , represents the set of related rules of rule i, Represents the substation rule association diagram, is a graph attention network with parameters , used to learn structured representations of regular nodes, It is the feature concatenation operation of 4 attention heads. represents the sum of all neighbors j of node i;

[0106] S140, Visual Data Feature Extraction: Processing infrared thermal imaging of substation equipment (such as transformers, GIS equipment, cable terminals, etc.) , =1024 (infrared image height), =768 (infrared image width), =3 (number of infrared image channels);

[0107] The calculation formula is as follows:

[0108] ;

[0109] in, is the substation infrared image network parameter, is the l-th layer infrared image residual block, is the infrared feature vector of the substation, with a dimension of 512. represents the global average pooling operation, represents the input substation infrared image, represents the concatenation operation of 18 layers of infrared image residual blocks, is the infrared feature vector of the substation, with a dimension of 512. is a 3D residual network with parameters , used to extract deep visual features of substation infrared images;

[0110] S200, Substation Multi-Source Data Alignment: This system uses linear projection to map heterogeneous data from multiple sources, including substation equipment operation and maintenance documents, monitoring data, procedures and rules, and infrared images, into a unified feature space. This system constructs multimodal positive and negative sample pairs for the same equipment, employs contrastive learning to narrow the distance between related equipment features, and dynamically optimizes the projection matrix for each type of data, enabling the coordinated expression of multi-source substation data.

[0111] In one embodiment of the present invention, the following steps are specifically included:

[0112] S210, unified representation of substation data: integrating multi-source features such as operation and maintenance text, monitoring data, procedures and rules, and infrared images Mapping to a unified dimensional space;

[0113] The calculation formula is as follows:

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] in, Unify the dimensions of substation features (256 dimensions, including equipment status, operating parameters, etc.), They are the projection matrix of substation operation and maintenance record text, equipment monitoring data (such as oil temperature, load, etc.), operation and maintenance regulations and equipment infrared images. is the corresponding bias vector, Indicates the characteristics of operation and maintenance text (such as fault description, handling measures, etc.), Indicates equipment monitoring characteristics (such as real-time operating data), Indicates the characteristics of procedures and rules (such as operating specifications, alarm thresholds, etc.), Indicates infrared image features (such as device hotspot distribution), is the time step index after downsampling, is the rule node index;

[0119] S220, construction of substation equipment sample pairs: based on substation equipment number (e.g. "220kV main transformer #1") Establish multi-source data sample pairs;

[0120] The calculation formula is as follows:

[0121] ;

[0122] ;

[0123] Sampling strategy:

[0124] Sampling per batch = 1024 pairs of positive samples from the same substation (e.g. data from the same main transformer at different times);

[0125] Sampling for each device =5 negative samples of different devices (such as transformers of different voltage levels);

[0126] in represents a set of sample pairs of the same substation equipment, represents a set of sample pairs of different substation equipment, Indicates the characteristics of the target substation equipment, Indicates the characteristics of paired substation equipment. is the unique identifier of device a, is the unique identifier of device b;

[0127] S230, Substation Equipment Feature Comparative Learning: Calculates multi-source data contrast loss to enhance the similarity of features of the same equipment;

[0128] The calculation formula is as follows:

[0129] ;

[0130] The device feature similarity measurement is:

[0131] ;

[0132] in, is the temperature parameter (used to adjust the degree of aggregation of the characteristic distribution of substation equipment), is the number of sample pairs of the same substation equipment, is the number of negative samples for each substation, , , are the characteristic vectors of the i-th, j-th and k-th substation equipment respectively, is the cosine similarity between the i-th and j-th eigenvectors, 、 are the L2 norms of the feature vectors of the i-th and j-th substation equipment, is the index of the negative sample, ranging from , contrastive learning loss function, is the contrast loss function value;

[0133] S240, Substation Data Projection Optimization: Optimizes the projection matrix of various data types through backpropagation;

[0134] The calculation formula is as follows:

[0135] ;

[0136] Optimization parameters:

[0137] Initial learning rate: (Set according to the frequency of substation data update);

[0138] Adam optimizer: =0.9, =0.999) (used to balance the weight of historical data and new data of substation equipment);

[0139] in Contains the projection matrix parameters of all substation data, is the learning rate, is the optimizer momentum parameter (used to adjust the stability of substation feature learning), Refers to the contrast loss function About projection matrix parameters The gradient of is used to guide the update direction and amplitude of the projection matrix parameters to minimize the contrast loss between different modal data;

[0140] S300, Substation Knowledge Triple Extraction: This system jointly detects substation equipment entity boundaries in operation and maintenance documents and equipment monitoring data, fuses multimodal equipment features through an attention mechanism, infers the relationship types between devices, and uses a dynamic threshold to filter low-confidence triplets to output structured knowledge units that meet substation domain constraints.

[0141] In one embodiment of the present invention, the following steps are specifically included:

[0142] S310, cross-modal entity boundary detection of substation equipment: jointly detect the start / end position of substation equipment entities from multimodal features such as operation and maintenance text, equipment monitoring data, and infrared images;

[0143] The calculation formula is as follows:

[0144] ;

[0145] ;

[0146] in is the weight matrix for substation equipment boundary detection (used to identify the boundaries of key information such as equipment name and model), represents the feature concatenation operation, is the probability distribution of the device boundary location (such as the start and end locations of "220kV main transformer #1"), is the length of the operation and maintenance text sequence (usually 500, 1000 words), Monitor the characteristic length of the equipment time series (usually 24 hours of sampling data);

[0147] S320, multimodal feature fusion of substation equipment: Aggregates cross-modal features such as operation and maintenance text, equipment monitoring data, and infrared images to generate substation equipment entity embeddings;

[0148] The calculation formula is as follows:

[0149] ;

[0150] ;

[0151] in For Substation Data Modal Adapter:

[0152] ;

[0153] ;

[0154] in, ;

[0155] in, is the attention weight of each modality (e.g. the infrared image has a higher weight at the moment of failure), is the location index of the substation equipment entity i (such as the spatial location of the main transformer and GIS equipment), is the hidden layer dimension (including device status, parameters and other features), is the adapter weight matrix (used to align the feature dimensions of different modal data), Represents the operation and maintenance text modal adapter (processing fault descriptions, operation records, etc.), Represents the equipment monitoring data modal adapter (processing oil temperature, load and other time series data), is the eigenvector of mode m, is the modal index of the summation term, To unify the feature dimension, the value is 256, represents the input feature vector, represents a collection of different data modalities, where Refers to text mode (operation and maintenance records, fault descriptions, etc.), Refers to the time series mode (device monitoring data), refers to the visual modality (infrared image data), is the data modality index, Refers to the corresponding mode The characteristic adaptation function of Refers to the corresponding mode The eigenvector of

[0156] S330, Substation Equipment Relationship Reasoning: Based on Substation Equipment Entity The feature inference type of the relationship between devices;

[0157] The calculation formula is as follows:

[0158] ;

[0159] Feature construction:

[0160] Hadamard product Capture electrical or mechanical coupling characteristics between devices;

[0161] Difference Strengthen the differences in voltage and capacity levels between devices;

[0162] in, The weight matrix for classifying relationships between substation equipment (used to identify 56 typical relationships between equipment), is the number of substation equipment relationship types, represents the Hadamard product operation, , is the feature vector of the substation equipment entity i and j, with dimension , among which the substation equipment entities (such as main transformers, switches, etc.) Indicates that a given device Time Relationship The conditional probability of Indicates the relationship type between substation equipment (such as "connection", "protection", etc.);

[0163] S340, substation knowledge triple screening: adaptively filtering substation equipment relationship triples with low confidence;

[0164] The calculation formula is as follows:

[0165] ;

[0166] ;

[0167] in is the mean confidence level of the substation equipment relationship (usually 0.8, 0.9), is the confidence standard deviation (reflecting the stability of relationship recognition), For substation equipment The number of historical occurrences of is the total number of triples in the knowledge graph (usually on the order of 100,000 or 1 million), represents a valid set of substation knowledge triples, Represents a dynamic threshold (adaptively adjusted according to the frequency of occurrence of the relationship), Represents the relationship between substation equipment Confidence in predictions;

[0168] S400, dynamic optimization of substation knowledge graph: clusters similar substation equipment entities to eliminate redundancy, adjusts relationship weights based on the density of associated knowledge between devices, applies time decay to historical operation and maintenance relationships, eliminates conflicts between new and old operation and maintenance knowledge, and optimizes the substation graph topology by minimizing structural entropy;

[0169] In one embodiment of the present invention, the following steps are specifically included:

[0170] S410, substation equipment entity similarity clustering: calculating the semantic similarity between substation equipment entity embeddings and merging redundant equipment entities;

[0171] The calculation formula is as follows:

[0172] ;

[0173] Merge conditions:

[0174] ;

[0175] in, ;

[0176] in, is the similarity between substation equipment entities i and j (e.g., the similarity between “220kV main transformer #1” and “220kV#1 main transformer”), The commissioning timestamp of the equipment entity (such as the commissioning date of the main transformer). is the device similarity threshold (determined based on the substation equipment naming convention), is the time attenuation coefficient (reflecting the impact of differences in equipment operation time), represents the vector norm, It is the vector transpose operation;

[0177] S420, edge weight update guided by the associated knowledge density of substation equipment: dynamically adjust the relationship weight based on the embedding density of the associated subgraph between devices;

[0178] The calculation formula is as follows:

[0179] ;

[0180] ;

[0181] in, Associate edges with devices The original weight (such as the protection relationship strength between the main transformer and the arrester), is the updated weight, is the knowledge density of the neighborhood of device node i (such as the density of the association rules of devices around the main transformer), represents the knowledge density of the neighborhood of device node j, is the set of adjacent devices of device node i (such as all devices directly connected to the main transformer), is the feature vector of substation equipment entity i (including equipment type, voltage level and other features), The feature vector representing the substation entity k;

[0182] S430, substation operation and maintenance relationship time decay: apply exponential decay to historical operation and maintenance relationships;

[0183] The calculation formula is as follows:

[0184] ;

[0185] ;

[0186] in, Days (corresponding to the standard maintenance cycle of power station equipment), is the initial weight of the operation and maintenance relationship (such as the initial credibility of equipment maintenance records), The time when the operation and maintenance relationship is established (such as the time when the maintenance is completed), is the half-life of operation and maintenance knowledge (corresponding to the equipment maintenance cycle), is the operation and maintenance knowledge decay rate (reflecting the speed at which the timeliness of operation and maintenance knowledge decreases), represents the weight of the operation and maintenance relationship at time t, is the logarithm with base 2;

[0187] S440, Substation Operation and Maintenance Knowledge Conflict Resolution: Solving New Operation and Maintenance Triplet Conflict with existing operations and maintenance maps;

[0188] The calculation formula is as follows:

[0189] Retention decision:

[0190] when When replacing old operation and maintenance knowledge (such as new equipment defect handling methods are obviously better than old methods), When the new and old maintenance standards are not much different, they will be retained in parallel for review by operation and maintenance experts.

[0191] in is the confidence level of the new operation and maintenance triplet (e.g., the reliability of the new maintenance procedure), is the confidence of the old operation and maintenance triplet (e.g., the reliability of the original maintenance method), the operation and maintenance knowledge confidence difference threshold is 0.2 (determined based on substation equipment maintenance experience), and the confidence ratio threshold is 2 (reflecting the difference in effectiveness between the old and new maintenance methods);

[0192] S450, Substation Knowledge Graph Topology Entropy Optimization: Minimizing graph structure entropy to improve equipment association query efficiency;

[0193] The calculation formula is as follows:

[0194] ;

[0195] Optimization goal:

[0196] ;

[0197] in, is the connectivity of substation node i (such as the number of connections between the main transformer and other devices), is the total number of device association edges (the total number of relationships between all devices in the substation), is the number of substation equipment nodes (the total number of equipment in the station), is the regularization coefficient (used to balance the complexity of the graph structure), are model parameters (including parameters for device feature extraction and relationship reasoning), represents the structural entropy of the substation knowledge graph (reflecting the organizational efficiency of the graph), is the substation knowledge graph (including all equipment nodes and relationship edges), is the L2 norm of the model parameters (used to prevent overfitting), represents the minimization operation on graph G, is the normalized connectivity of node i;

[0198] S500, Substation Operation and Maintenance Application Generation: This process parses the natural language requirements of operators into substation knowledge graph query patterns, performs multi-path reasoning under the constraints of equipment safety and operating procedures, constructs a candidate operation and maintenance strategy set that includes equipment safety, economy, and reliability assessments, and ultimately generates an executable operation and maintenance instruction sequence that complies with power equipment operating specifications.

[0199] In one embodiment of the present invention, the following steps are specifically included:

[0200] S510, graphical analysis of substation operation and maintenance requirements: natural language requirements of operation and maintenance personnel (e.g., "Check the cause of abnormal main transformer oil temperature") is converted to substation knowledge graph query mode;

[0201] The calculation formula is as follows:

[0202] ;

[0203] in, For natural language queries by operation and maintenance personnel (such as "check the cause of abnormal oil temperature of 220kV main transformer"), Query the path set for all possible substation equipment (such as main transformer, oil temperature, cooling system path), is a similarity function (used to match operation and maintenance requirements with equipment failure modes), is the feature splicing operation, Represents the substation knowledge graph prompt template (including equipment fault diagnosis, operating procedures, etc. templates), represents the pre-trained substation professional language model, Indicates the substation map mode being queried (such as oil temperature abnormality diagnosis mode), Represents the relationship triple of substation equipment (such as main transformer, monitoring, temperature sensor), represents the equipment fault diagnosis reasoning path, which is a triple sequence. Represents a substation knowledge graph prompt template that converts natural language queries into graph query mode. The input parameters are and ;

[0204] S520, multi-constraint path reasoning for substation equipment: searching for a reasoning path that satisfies equipment safety constraints and operation timing constraints in the optimized substation knowledge graph G′;

[0205] The calculation formula is as follows:

[0206] ;

[0207] in, hours (corresponding to the standard maintenance cycle of power station equipment), For operation and maintenance time window constraints (based on equipment maintenance procedures), is the equipment association edge weight (such as the association strength between the main transformer and the cooler), is the current time, Establish time for equipment relationship (such as last maintenance time), represents the set of valid substation equipment constraint paths, Represents the relationship triple of substation equipment (such as main transformer, monitoring, temperature sensor), represents the equipment fault diagnosis reasoning path, Represents the product of all edge weights on the path;

[0208] S530, Substation Operation and Maintenance Strategy Utility Function Modeling: Quantitatively evaluate the expected utility of each candidate operation and maintenance strategy;

[0209] The calculation formula is as follows:

[0210] ;

[0211] The utility function of each component is defined as:

[0212] ;

[0213] (Reliability of equipment operation and maintenance paths, representing the success rate of maintenance plans);

[0214] ;

[0215] (Operation and maintenance cost efficiency, including power outage losses, labor costs, etc.);

[0216] ;

[0217] (Equipment operation risk, such as the risk of oil temperature exceeding the standard);

[0218] in, The weight coefficient of each operation and maintenance utility is dynamically adjusted based on the importance of the equipment. is the operation and maintenance cost sensitivity parameter, which is related to the equipment maintenance budget. For operation and maintenance strategy The execution cost, including spare parts, labor, etc., is the equipment risk assessment weight matrix, which reflects the degree of harm of different fault types. For operation and maintenance strategy The total utility of is the nth operation and maintenance utility component, It is the timing characteristics of the equipment operation status, such as oil temperature, load and other parameters, is the device association edge weight, ranging from [0,1];

[0219] S540, Substation Operation and Maintenance Strategy Sequence Optimization Generation: Generates the optimal operation and maintenance strategy sequence based on reinforcement learning;

[0220] The calculation formula is as follows:

[0221] ;

[0222] ;

[0223] in, For operation and maintenance strategy functions (such as oil temperature abnormality handling process), is the discount factor at time t (reflecting the uncertainty of equipment state prediction), is the equipment status at time t (such as oil temperature, load and other parameters), is the operation and maintenance action at time t (such as adjusting the cooling fan), Instant rewards for operation and maintenance (such as temperature drop), Indicates the operation and maintenance time step (usually 24 hours), Indicates the relationship type between substation equipment. represents the expectation operation, represents the optimized substation knowledge graph, It refers to the operation and maintenance strategy that expects the maximum value among all possible operation and maintenance strategies. Refers to the set of all possible operation and maintenance strategies;

[0224] S550, substation operation and maintenance plan compilation: converts the operation and maintenance strategy into an instruction set that complies with the power equipment operation specifications;

[0225] The calculation formula is as follows:

[0226] ;

[0227] Satisfy the constraints:

[0228] ;

[0229] in, For the final substation operation and maintenance operation sequence (such as "stop the cooling fan first, then measure the oil temperature"), is a set of operation and maintenance strategies (including multiple optional disposal solutions), Timestamp for operation and maintenance (must comply with equipment operation regulations). A topological sorting algorithm considering the safety constraints of equipment operation is proposed. Indicates the adjacent operation steps in the operation and maintenance operation sequence, which are discrete action labels. Indicates the existence of a quantifier, there are elements that meet the conditions, It represents a universal quantifier, which applies to all elements.

[0230] Based on the substation knowledge graph constructed above, when there is a conflict between the expert operation and maintenance rules and the real-time monitoring data, for example, there is an entry in the expert rule base that says "the load rate of the main transformer is >85% and lasts for 10 minutes, an early warning is required", but the substation SCADA real-time monitoring shows that "the winding temperature rise curve of a main transformer meets the requirements under the load rate of 91%". <0.3℃ / min” safety operation standard.

[0231] This resulted in the following phenomena: the confidence level in substation knowledge reasoning dropped to 53.2% and the trust of operators in the substation monitoring system decreased by 41%;

[0232] After the substation knowledge graph is built, the optimization process begins. The core conditions for triggering the optimization are:

[0233] ;

[0234] in For operation and maintenance rules Verification pass rate in substation real-time data (such as the accuracy of load rate warning rules), For operation and maintenance rules Confidence in the substation knowledge graph (based on historical operation and maintenance experience), , represents the empirical threshold for rule conflict determination, where =0.6 (rule too strict threshold), =1.4 (rule too loose threshold);

[0235] S600, substation operation and maintenance rules and data optimization: By real-time monitoring of deviations between substation equipment operating status and abnormalities triggered by operation and maintenance rules, it automatically quantifies the degree of rule conflicts, dynamically adjusts equipment operating threshold parameters, and attenuates the weight of conflicting rules. After verification through online A / B testing, the optimized operation and maintenance rules are injected into the substation knowledge graph, forming a closed loop of self-evolution of substation operation and maintenance rules.

[0236] When the performance of any substation operation and maintenance rules in actual equipment operation data deviates significantly from the built-in confidence level of the map (more than ±40%), the optimization process is automatically triggered;

[0237] The specific steps of the optimization process are as follows:

[0238] S610, Quantification of the degree of conflict between substation operation and maintenance rules: Calculation of operation and maintenance rules Deviation from the distribution of equipment operation data;

[0239] The calculation formula is as follows:

[0240] ;

[0241] in Thresholds for substation equipment operation and maintenance rules (such as the main transformer oil temperature alarm threshold of 85°C, the SF6 gas pressure alarm threshold of 0.4MPa, the GIS switch operation time threshold of 45ms, etc.) is the average value of actual equipment operating parameters (such as the actual operating oil temperature of the main transformer, the actual SF6 pressure, etc.), is the standard deviation of equipment operating parameters (reflecting equipment operation stability), The number of device alarms that trigger operation and maintenance rules. is the total number of monitoring times of equipment operation, Indicates the degree of deviation between operation and maintenance rules and actual operation status;

[0242] S620, Dynamic Optimization of Substation Equipment Operation and Maintenance Thresholds: Adjusting the trigger thresholds for equipment operation and maintenance rules ;

[0243] ;

[0244] Learning rate (determined based on the rate of change of substation equipment status), variance The fluctuation range of key monitoring quantities from substation equipment (such as main transformer oil temperature ±5°C, SF6 pressure ±0.05MPa, etc.), is the threshold adjustment direction function, is the optimized device alarm threshold, Indicates the original device alarm threshold;

[0245] S630, Substation operation and maintenance rule weight decay: Update the influence weight of the operation and maintenance rules in the substation knowledge graph:

[0246] ;

[0247] in Indicates the attenuation coefficient of the operation and maintenance rules (corresponding to the 90-day equipment maintenance cycle). , initial weight Ratings of the importance of rules from substation operation and maintenance experts (1, 5 points), is the weight of the updated operation and maintenance rules in the knowledge graph;

[0248] S640, online AB testing and verification of substation operation and maintenance rules: running new and old operation and maintenance rule versions in parallel to evaluate equipment operation performance improvements:

[0249] ;

[0250] Improved retention An optimized version of and They respectively represent the performance of the new and old operation and maintenance rules in terms of equipment warning accuracy, fault prediction advance time and other indicators. Indicates performance over relationships;

[0251] S650, Substation Knowledge Graph Hot Update: Injecting optimized operation and maintenance rule parameters:

[0252] ;

[0253] in This is the updated substation knowledge graph. is the original substation knowledge graph, For operation and maintenance rules that need to be replaced (such as outdated device alarm thresholds), For optimized operation and maintenance rules (such as new equipment alarm thresholds), Equipment operation data evidence (such as alarm statistics, fault cases, etc.) to support rule optimization.

[0254] According to the above knowledge graph construction method, the following example is given:

[0255] Multi-view feature extraction (S100):

[0256] Fault record: "2023-08-15 14:23, the B phase winding temperature suddenly rose to 98°C, and the oil chromatography analysis showed that C2H4 exceeded the standard";

[0257] Extract four types of features:

[0258] Text feature (S110): fault work order text;

[0259] Time series characteristics (S120): 24h temperature / oil chromatogram curve data (1s sampling);

[0260] Rule feature (S130): "Oil temperature > 95°C for 10 minutes requires warning" rule;

[0261] Visual features (S140): infrared thermal imaging during the fault period.

[0262] Multi-source data alignment (S200):

[0263] Unified representation (S210): aligning the “B-phase winding” entity with the hotspot coordinates;

[0264] Sample pair construction (S220): associating 14:20-14:25 temperature mutation data;

[0265] Contrastive learning (S230): building cross-modal joint representations;

[0266] Projection optimization (S240): Dynamically adjust the weights of each modality.

[0267] Knowledge triple extraction (S300):

[0268] Entity detection (S310): Identify the "main transformer B phase winding";

[0269] Feature fusion (S320): fusing multimodal information;

[0270] Relational reasoning (S330): extracting the triplet (main transformer B-phase winding, abnormal temperature rise event, oil chromatogram C2H4 exceeding the standard);

[0271] Triplet screening (S340): Confidence level 0.89, associated insulation degradation warning.

[0272] Knowledge graph optimization (S400):

[0273] Entity clustering (S410): merging similar device descriptions;

[0274] Relationship weight update (S420): The weight of "winding temperature-oil color spectrum" is increased to 0.93;

[0275] Time-dependent decay (S430): taking into account the time-dependent nature of historical faults;

[0276] Conflict resolution (S440): dealing with conflicts between new and old knowledge;

[0277] Structural optimization (S450): Added the relationship of “frequent abnormalities → insulation aging”.

[0278] Operation and maintenance application generation (S500):

[0279] Demand analysis (S510): Analyze the newly-occurred "C-phase winding temperature 92°C" event;

[0280] Path reasoning (S520): searching for relevant treatment solutions;

[0281] Utility modeling (S530): Evaluate the effectiveness of each solution;

[0282] Strategy Optimization (S540): Generate disposal recommendations, start online oil chromatography monitoring, retrieve similar failure cases from the past three months, and conduct partial discharge tests within 12 hours.

[0283] Background: Expert rules stipulate that "a warning is required when the load factor exceeds 85%." However, actual monitoring shows that the temperature rise of a certain transformer meets safety standards at a load factor of 91%, resulting in frequent false alarms.

[0284] Optimization process:

[0285] Calculate the conflict score: =1.1;

[0286] Dynamically adjust the threshold to 92°C;

[0287] The rule weight decays from 0.9 to 0.7;

[0288] Online testing and verification reduced the false positive rate from 32% to 7%;

[0289] Update the threshold parameters and weights of the rule in the graph;

[0290] Optimization results:

[0291] The rule triggering accuracy increased from 68% to 89%;

[0292] The trust of operation and maintenance personnel in this rule has risen to 92%;

[0293] Related alarm tickets decreased by 65%;

[0294] The effects are as follows:

[0295]

[0296] This example demonstrates how a knowledge system can continuously adapt to actual working conditions through closed-loop optimization.

[0297] The above describes the embodiments of the present invention, but the present invention is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, all of which are protected by the present invention.

Claims

1. A method for constructing and optimizing a substation knowledge graph based on multi-view learning, characterized in that: include: S100, multi-view feature extraction: processing substation multi-source data through heterogeneous models; S200, Substation Multi-Source Data Alignment: This technology uses linear projection to map heterogeneous data from substation equipment, including operation and maintenance documents, monitoring data, procedures and rules, and infrared images, into a unified feature space. This technology constructs multimodal positive and negative sample pairs for the same equipment, employs contrastive learning to narrow the distance between related equipment features, and dynamically optimizes the projection matrix for each type of data. S300, Substation Knowledge Triple Extraction: This system jointly detects substation equipment entity boundaries in operation and maintenance documents and equipment monitoring data, fuses multimodal equipment features through an attention mechanism, infers the relationship types between devices, and uses a dynamic threshold to filter low-confidence triplets to output structured knowledge units that meet substation domain constraints. S400, dynamic optimization of substation knowledge graph: clusters similar substation equipment entities to eliminate redundancy, adjusts relationship weights based on the density of associated knowledge between devices, applies time decay to historical operation and maintenance relationships, eliminates conflicts between new and old operation and maintenance knowledge, and optimizes the substation graph topology by minimizing structural entropy; S500, Substation Operation and Maintenance Application Generation: This process parses the natural language requirements of operators into substation knowledge graph query patterns, performs multi-path reasoning under the constraints of equipment safety and operating procedures, constructs a candidate operation and maintenance strategy set that includes equipment safety, economy, and reliability assessments, and ultimately generates an executable operation and maintenance instruction sequence that complies with power equipment operating specifications. S600, substation operation and maintenance rules and data optimization: By real-time monitoring of deviations between substation equipment operating status and abnormalities triggered by operation and maintenance rules, the system automatically quantifies the degree of rule conflicts, dynamically adjusts equipment operating threshold parameters, and attenuates the weight of conflicting rules. After verification through online AB testing, the optimized operation and maintenance rules are injected into the substation knowledge graph, forming a closed loop of self-evolution of substation operation and maintenance rules.

2. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 1, characterized in that: S100 specifically includes the following steps: S110, Text Modal Feature Extraction: Semantic encoding of unstructured text data; S120, time series data feature extraction: performing spatiotemporal feature learning on the equipment monitoring sequences collected by the substation SCADA system; S130, expert rule graph feature extraction: construct substation equipment operation and maintenance procedures and expert experience rules into a rule association graph and generate node embedding; S140, visual data feature extraction: Process the infrared thermal imaging of substation equipment and output the substation infrared feature vector.

3. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 2, characterized in that: In multi-view feature extraction, processing heterogeneous data from multiple sources in substations involves using BERT to extract semantic features from substation operation and maintenance work orders and fault records, TCN to capture the temporal patterns of monitoring data from transformers and switchgear, GAT to aggregate the correlation features between substation equipment operation and maintenance procedures and expert experience rules, and 3DResNet to extract visual features from infrared thermal imaging of substation equipment. Each modality feature is coded independently to preserve view specificity, forming four types of feature representations including operation and maintenance text, equipment monitoring, procedures and rules, and equipment thermal images.

4. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 3, characterized in that: S200 specifically includes the following steps: S210, unified representation of substation data: mapping multi-source features of operation and maintenance text, monitoring data, procedures and rules, and infrared images into a unified dimensional space; S220, substation equipment sample pair construction: establishing multi-source data sample pairs based on substation equipment numbers; S230, Substation Equipment Feature Comparative Learning: Calculates multi-source data contrast loss to enhance the similarity of features of the same equipment; S240, Substation Data Projection Optimization: Optimizes the projection matrix of various data types through backpropagation.

5. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 4, characterized in that: S300 specifically includes the following steps: S310, cross-modal entity boundary detection of substation equipment: jointly detecting the start / end position of substation equipment entities from multimodal features of operation and maintenance text, equipment monitoring data, and infrared images; S320, multimodal feature fusion of substation equipment: aggregating operation and maintenance text, equipment monitoring data, and infrared image cross-modal features to generate substation equipment entity embedding; S330, power substation equipment relationship reasoning: reasoning about the relationship type between equipment based on the features of power substation equipment entity pairs; S340, substation knowledge triple screening: adaptively filtering low-confidence substation equipment relationship triples; Its effective substation knowledge triples are as follows: ； ； in, is the mean confidence value of the substation equipment relationship, is the confidence standard deviation, For substation equipment The number of historical occurrences of is the total number of triples in the knowledge graph, represents a valid set of substation knowledge triples, represents the dynamic threshold, Represents the relationship between substation equipment The prediction confidence of , are the characteristic vectors of substation entities i and j, Indicates the relationship type between substation equipment. Represents a triplet of substation equipment relations.

6. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 5, characterized in that: S400 specifically includes the following steps: S410, substation equipment entity similarity clustering: calculating the semantic similarity between substation equipment entity embeddings and merging redundant equipment entities; S420, edge weight update guided by the associated knowledge density of substation equipment: dynamically adjust the relationship weight based on the embedding density of the associated subgraph between devices; S430, substation operation and maintenance relationship time decay: apply exponential decay to historical operation and maintenance relationships; S440, substation operation and maintenance knowledge conflict resolution: resolve conflicts between newly added operation and maintenance triples and existing operation and maintenance graphs; S450, Substation Knowledge Graph Topology Entropy Optimization: Minimizing graph structure entropy to improve equipment association query efficiency; The calculation formula for structural entropy optimization is as follows: ； Optimization goal: ； in, is the degree of node i, is the total number of edges, is the number of nodes, is the regularization coefficient, are model parameters, represents the structural entropy of the graph, For the substation knowledge graph, is the L2 norm of the model parameters, represents the minimization operation on the substation knowledge graph G, is the normalized connectivity of node i.

7. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 6, characterized in that: The calculation formula for edge weight update guided by the associated knowledge density of substation equipment is as follows: ； ； in, For the edge The original weight of is the updated weight, is the knowledge density of the neighborhood of node i, represents the knowledge density of the neighborhood of node j, is the neighbor set of node i, The feature vector representing the substation entity k; The calculation formula for the timeliness attenuation of substation operation and maintenance relationship is as follows: ； ； in, is the initial weight, Build time for relationships, is the half-life, is the attenuation rate, represents the relationship weight at time t, is the logarithm with base 2; The process of resolving substation operation and maintenance knowledge conflicts is as follows: when When replacing old operation and maintenance knowledge, When the system is completed, it will be retained in parallel for review by operation and maintenance experts; in, is the confidence of the new operation and maintenance triplet, is the confidence of the old operation and maintenance triplet, the confidence difference threshold of operation and maintenance knowledge is 0.2, and the confidence ratio threshold is 2.

8. The method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 7 is characterized in that: S500 specifically includes the following steps: S510, graphical analysis of substation operation and maintenance requirements: converts the natural language requirements of operation and maintenance personnel into a substation knowledge graph query model; S520, multi-constraint path reasoning for substation equipment: searching for a reasoning path that satisfies equipment safety constraints and operation timing constraints in the optimized substation knowledge graph; S530, Substation Operation and Maintenance Strategy Utility Function Modeling: Quantitatively evaluate the expected utility of each candidate operation and maintenance strategy; S540, Substation Operation and Maintenance Strategy Sequence Optimization Generation: Generates the optimal operation and maintenance strategy sequence based on reinforcement learning; S550, substation operation and maintenance plan compilation: converts the operation and maintenance strategy into an instruction set that complies with the operating specifications of power equipment.

9. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 8, characterized in that: The calculation formula of the optimal strategy sequence is as follows: ； ； in, is the policy function, is the discount factor at time t, is the state at time t, is the action at time t, For instant rewards, represents the operation and maintenance time step, represents the indicator function, represents the expectation operation, represents the optimized substation knowledge graph, It refers to the operation and maintenance strategy that expects the maximum value among all possible operation and maintenance strategies. Refers to the set of all possible operation and maintenance strategies; The calculation formula for compiling the operation and maintenance plan is as follows: ； Satisfy the constraints: ； in, For the final substation operation and maintenance sequence, It is a set of operation and maintenance strategies, including multiple optional disposal solutions. Timestamp of the operation. A topological sorting algorithm considering the safety constraints of equipment operation is proposed. Indicates the adjacent operation steps in the operation and maintenance operation sequence, which are discrete action labels. Indicates the existence of a quantifier, there are elements that meet the conditions, It represents a universal quantifier, which applies to all elements.

10. A method for constructing and optimizing a substation knowledge graph based on multi-view learning according to claim 9, characterized in that: When the performance of any substation operation and maintenance rule in actual equipment operation data deviates significantly from the built-in confidence level of the map, the optimization process is automatically triggered. The steps of the optimization process are as follows: S610, quantification of the conflict degree of substation operation and maintenance rules: Calculate the deviation between the operation and maintenance rules and the distribution of equipment operation data; S620, dynamic optimization of substation equipment operation and maintenance thresholds: adjusts the triggering thresholds of equipment operation and maintenance rules; S630, Substation operation and maintenance rule weight decay: Update the influence weight of the operation and maintenance rules in the substation knowledge graph; S640, online AB testing and verification of substation operation and maintenance rules: running old and new operation and maintenance rule versions in parallel to evaluate equipment operation performance improvements; S650, hot update of substation knowledge graph: injecting optimized operation and maintenance rule parameters.

Images