Operation order verification method and system based on knowledge graph, medium and equipment

Abstract

The invention discloses an operation order verification method and system based on a knowledge graph, a medium and equipment, and belongs to the field of electric power system automation, and the method comprises the steps: obtaining an operation order of a power grid; performing semantic analysis on the operation ticket to obtain a corresponding operation sequence; according to a preset knowledge graph, performing reasoning matching on the operation sequence to generate a risk step set; wherein the knowledge graph is constructed according to power equipment information in a power system; according to the graph structure and the semantic feature corresponding to each risk step in the risk step set, performing conflict detection to obtain a conflict detection result; and according to a preset generative model, the knowledge graph and the conflict detection result, generating a corresponding optimization suggestion, and completing operation ticket verification. Therefore, by implementing the method and the device, the problem of poor adaptability of a traditional operation order verification method in a dynamic scene of a power grid in the prior art can be solved.

Description

Technical Field

[0001] The present invention belongs to the field of power system automation and relates to an operation ticket verification method, system, medium and equipment based on knowledge graph. Background Art

[0002] In power system operations, compliance verification of operation tickets is a core component of ensuring safe grid operation. With the rapid integration of new power equipment and the dynamic changes in grid topology, operation tickets must adapt to complex scenarios such as device status updates and multi-device collaborative operations in real time. This places higher demands on the verification system's flexibility, real-time nature, and semantic understanding capabilities.

[0003] Existing technologies primarily build static rule bases based on expert experience, matching operation ticket content with predefined templates to verify the compliance of each operation step. However, static rule bases require manual maintenance, making it difficult to respond promptly to dynamic adjustments to grid equipment configurations (such as new equipment or changes to procedures). This results in lagging verification rules and insufficient coverage of multi-device linkage operation scenarios. Traditional methods, especially when faced with non-standardized operation descriptions or complex semantic logic, lack dynamic reasoning capabilities, making them prone to missed detections or misjudgments, posing safety risks and inefficient operations. Summary of the Invention

[0004] The present application provides an operation ticket verification method, system, medium and equipment based on knowledge graph, which can solve the problem of poor adaptability of traditional operation ticket verification methods in the existing technology when facing dynamic power grid scenarios.

[0005] To achieve the above objectives, in a first aspect, the present invention provides an operation ticket verification method based on a knowledge graph, comprising:

[0006] Obtaining an operation ticket for the power grid;

[0007] Perform semantic analysis on the operation ticket to obtain a corresponding operation sequence;

[0008] According to a preset knowledge graph, the operation sequence is reasoned and matched to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system;

[0009] Perform conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result;

[0010] Based on the preset generative model, the knowledge graph and the conflict detection results, corresponding optimization suggestions are generated to complete the operation ticket verification.

[0011] Compared with the existing technology, the embodiments of the present application have the following beneficial effects: by obtaining the power grid operation ticket and generating a structured operation sequence through semantic analysis, the semantic understanding problem of non-standardized operation tickets is solved; by dynamically reasoning and matching the knowledge graph to generate a set of risk steps, it replaces the static rule base and adapts to the dynamic changes in the configuration of power grid equipment; based on the dual conflict detection mechanism of graph structure and semantic features, the verification coverage rate in the multi-device linkage operation scenario is improved; combining the generative model with the knowledge graph to generate optimization suggestions, a closed-loop verification process is realized, and the cost of manual intervention is reduced. The overall solution solves the core problem of poor adaptability to dynamic power grid scenarios caused by the rigidification of rules and lack of reasoning ability of traditional methods through the collaboration of dynamic analysis, knowledge reasoning and intelligent correction, and significantly improves the verification efficiency and safety.

[0012] In some embodiments of the first aspect of the present application, performing semantic parsing on the operation ticket to obtain a corresponding operation sequence includes:

[0013] According to a preset semantic encoder, the operation ticket is segmented and context-encoded to obtain a corresponding first embedding vector sequence;

[0014] Fusing a preset structured hint template with the embedding vector sequence to obtain a second embedding vector sequence guided by the structure;

[0015] Inputting the second embedding vector sequence into a preset bidirectional long short-term memory network to extract context dependencies and obtain a semantic feature vector sequence;

[0016] The semantic feature vector sequence is input into a preset CRF decoder for decoding to obtain an entity label sequence, and the entity label sequence is structured to obtain a corresponding operation sequence.

[0017] Compared with the prior art, the above embodiment has the following beneficial effects: the operation ticket is segmented and context-encoded through the semantic encoder to generate a first embedding vector sequence, thereby solving the semantic discreteness problem of unstructured text; further, by fusing the structured prompt template with the embedding vector sequence, the model is guided to focus on key entities (such as actions and equipment) in the operation ticket, thereby enhancing the semantic capture accuracy of domain terms; then, the context dependency is extracted through the bidirectional long short-term memory network, thereby improving the recognition robustness of nested entities (such as "circuit breaker B connected to bus A"); finally, the semantic feature vectors are sequence-labeled through the CRF decoder, thereby optimizing the global consistency of entity labels, avoiding conflicts between isolated labels, forming a structured operation sequence, and providing a parsable logical unit for subsequent verification.

[0018] In some embodiments of the first aspect of the present application, the reasoning and matching of the operation sequence according to a preset knowledge graph to generate a set of risk steps includes:

[0019] For each operation node in the operation sequence, construct a corresponding executable path set according to the knowledge graph;

[0020] The operation sequence is matched and scored with each executable path in the executable path set, risky operations are identified, and a risk step set is generated.

[0021] Compared with the existing technology, the above embodiment has the following beneficial effects: constructing an executable path set for each operation node, and using the physical layer topology and procedure layer constraints of the knowledge graph to generate a compliant operation chain; quantifying the degree of deviation between operation steps and safety regulations through a scoring function, and realizing dynamic risk grading in multi-device linkage scenarios.

[0022] In some embodiments of the first aspect of the present application, performing conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result includes:

[0023] For each operation step in the risk step set, a corresponding operation graph is constructed and matched with a preset conflict template to obtain a conflict detection result;

[0024] For the operation diagrams that fail to match the conflict template successfully, each operation diagram is encoded according to the preset graph neural network to obtain an embedding vector, and the vector similarity is calculated based on the embedding vector and the preset reference vector to obtain the conflict detection result.

[0025] Compared with the existing technology, the above embodiment has the following beneficial effects: conflict subgraph template matching is used to quickly locate known illegal operations (such as "grounding state + closing action"); for operation graphs that do not match the template, the vector is encoded through the graph neural network and the anomaly score is calculated to quantify its semantic deviation from the benchmark compliant operation, thereby realizing the active discovery of new conflicts.

[0026] In some embodiments of the first aspect of the present application, generating corresponding optimization suggestions based on a preset generative model, the knowledge graph, and the conflict detection results to complete operation ticket verification includes:

[0027] According to the conflict detection result, a plurality of candidate operation sequences are generated through a preset generative model and the knowledge graph;

[0028] Based on the knowledge graph and the operation sequence, the semantic similarity and procedure consistency score of each candidate operation sequence are calculated, and the candidate operation sequence corresponding to the best score is output as an optimization suggestion to complete the operation ticket verification.

[0029] Compared with the existing technology, the above embodiment has the following beneficial effects: generating multiple versions of candidate operation sequences, and screening the optimal suggestions through the scoring function of semantic similarity and procedural consistency, balancing the preservation of operation intentions and the satisfaction of safety constraints, and avoiding the introduction of secondary errors due to manual corrections.

[0030] In some embodiments of the first aspect of the present application, the knowledge graph is constructed based on information about power equipment in the power system, including:

[0031] Acquire power equipment information in the power system; wherein the power equipment information includes each device entity, device attributes, operating status and corresponding operating procedures;

[0032] The device entities, operation actions, and operating states in the power equipment information are used as nodes, and the semantic relationships between the nodes in the power equipment information are used as edges in the graph to construct a knowledge graph; wherein the semantic relationships include: device connection relationships, device operation relationships, status association relationships, and procedure constraint relationships between the nodes, and each edge is also assigned a corresponding edge weight;

[0033] The knowledge graph is updated according to the preset incremental subgraph algorithm. The algorithm is as follows:

[0034] in, represents the structure of the knowledge graph at time t, ΔV t and ΔE t They represent the newly added node set and edge set at time t respectively.

[0035] Compared with the existing technology, the above embodiment has the following beneficial effects: by taking power equipment entities, operating actions and operating status as knowledge graph nodes, and constructing the initial graph with semantic relationships as edges, a unified expression of equipment topology and procedural logic is achieved; further, the graph structure is dynamically updated through the incremental subgraph algorithm, and only newly added nodes and edges are merged to avoid the computing resource consumption of full reconstruction and adapt to real-time changes in power grid equipment configuration; the preset mechanism of edge weights provides a differentiated path priority basis for subsequent dynamic reasoning, supporting the refined scoring of risk steps.

[0036] In some embodiments of the first aspect of the present application, each of the edges is further provided with a corresponding edge weight, including:

[0037] According to the preset graph attention network, the weight of each edge is calculated. The algorithm is as follows:

[0038] Among them, α ij represents the attention weight between node i and adjacent node j, and are the eigenvectors corresponding to nodes i, j, and k respectively, W is the linear transformation matrix, is the attention vector, T represents the transposition operation, is the set of adjacent nodes of node i, and || represents the vector concatenation operation.

[0039] Compared with the existing technology, the above embodiment has the following beneficial effects: edge weights are calculated through the graph attention network (GAT), and the node feature vectors are interactively weighted using the attention vector and the linear transformation matrix, thereby dynamically strengthening the association relationship between key equipment (such as the operational dependency between the circuit breaker and the busbar); the dynamic screening mechanism of the adjacent node set is combined with the real-time status to filter invalid connection edges (such as disconnected disconnectors), thereby reducing redundant calculations and improving the real-time response efficiency and path selection accuracy of knowledge graph reasoning.

[0040] In a second aspect, the present invention also provides an operation ticket verification system based on a knowledge graph, comprising: a data acquisition module, a semantic parsing module, a reasoning and matching module, a conflict detection module, and an optimization output module;

[0041] Wherein, the data acquisition module is used to obtain the operation ticket of the power grid;

[0042] The semantic parsing module is used to perform semantic parsing on the operation ticket to obtain a corresponding operation sequence;

[0043] The reasoning and matching module is used to perform reasoning and matching on the operation sequence according to a preset knowledge graph to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system;

[0044] The conflict detection module is used to perform conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result;

[0045] The optimization output module is used to generate corresponding optimization suggestions based on the preset generative model, the knowledge graph and the conflict detection results, and complete the operation ticket verification.

[0046] Compared with the existing technology, the above embodiments of the present application have the following beneficial effects: by obtaining the power grid operation ticket and generating a structured operation sequence through semantic analysis, the semantic understanding problem of non-standardized operation tickets is solved; by dynamically reasoning and matching the knowledge graph to generate a set of risk steps, it replaces the static rule base and adapts to the dynamic changes in the configuration of power grid equipment; based on the dual conflict detection mechanism of graph structure and semantic features, the verification coverage rate in the multi-device linkage operation scenario is improved; combining the generative model with the knowledge graph to generate optimization suggestions, a closed-loop verification process is realized, and the cost of manual intervention is reduced. The overall solution solves the core problem of poor adaptability of traditional methods to dynamic power grid scenarios due to the rigidification of rules and lack of reasoning ability through the collaboration of dynamic analysis, knowledge reasoning and intelligent correction, and significantly improves the verification efficiency and safety.

[0047] In the third aspect, the present invention also provides an operation ticket verification device based on a knowledge graph, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the computer program is loaded into the processor, it implements the steps of any one of the operation ticket verification methods based on a knowledge graph of the present invention.

[0048] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the operation ticket verification methods based on the knowledge graph of the present invention are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 : A flow chart of an operation ticket verification method based on knowledge graph provided in some embodiments of the present invention.

[0050] Figure 2 : A structural diagram of an operation ticket verification system based on a knowledge graph provided in some embodiments of the present invention.

[0051] Figure 3 : A structural diagram of an operation ticket verification device based on a knowledge graph provided in some embodiments of the present invention.

[0052] Figure 4 : A schematic diagram of an operation ticket analysis provided in some embodiments of the present invention. DETAILED DESCRIPTION

[0053] 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] Example 1:

[0055] Please refer to Figure 1 To solve the problem in the prior art that traditional operation ticket verification methods have poor adaptability to dynamic power grid scenarios, an embodiment of the present invention provides an operation ticket verification method based on a knowledge graph, including steps S1 to S5:

[0056] Step S1: Obtain an operation ticket for the power grid.

[0057] Step S2: Perform semantic analysis on the operation ticket to obtain a corresponding operation sequence.

[0058] Furthermore, step S2 can be implemented by the following preferred implementation, including steps S21-S24, as follows:

[0059] S21: performing word segmentation and context encoding on the operation ticket according to a preset semantic encoder to obtain a corresponding first embedding vector sequence;

[0060] S22: fusing a preset structured hint template with the embedding vector sequence to obtain a second embedding vector sequence guided by the structure;

[0061] S23: Inputting the second embedding vector sequence into a preset bidirectional long short-term memory network to extract context dependencies and obtain a semantic feature vector sequence;

[0062] S24: Input the semantic feature vector sequence into a preset CRF decoder for decoding to obtain an entity label sequence, and perform structured processing on the entity label sequence to obtain a corresponding operation sequence.

[0063] In this preferred embodiment, the operation ticket is segmented and context-encoded through a semantic encoder to generate a first embedding vector sequence, thereby solving the semantic discreteness problem of unstructured text. Furthermore, the structured prompt template is fused with the embedding vector sequence to guide the model to focus on key entities (such as actions and equipment) in the operation ticket, thereby enhancing the semantic capture accuracy of domain terms. The context dependencies are then extracted through a bidirectional long short-term memory network to improve the recognition robustness of nested entities (such as "circuit breaker B connected to bus A"). Finally, the semantic feature vectors are sequence-labeled through a CRF decoder to optimize the global consistency of entity labels, avoid conflicts in isolated labels, form a structured operation sequence, and provide a parsable logical unit for subsequent verification.

[0064] In specific implementation, the purpose of parsing the operation ticket is to convert complex natural language text into a logical representation that can be understood and processed by machines. Figure 4The following figure shows a schematic diagram of an operation ticket parsing scheme. This scheme relies on a generative large model to parse the operation ticket content step by step, identifying the equipment name, operation type, and execution conditions involved in the operation ticket. For example, for descriptions such as "open the 110kV line A circuit breaker and close the line B disconnector", key elements are extracted and a structured (logical) operation sequence is generated to ensure that the parsing results fully cover all key operation links. During the parsing process, a knowledge graph can also be introduced as an auxiliary reference tool to correct and supplement the parsing results generated by the large model to ensure that the parsing results are consistent with the actual power grid equipment status and operating procedures.

[0065] To improve the accuracy and industry adaptability of semantic parsing, a semantic parsing framework based on the joint structure of RoBERTa+Prompt-Tuning+BiLSTM is designed in the large model of this solution, including the following:

[0066] ①RoBERTa is used as a basic semantic encoder to model the contextual relationships and industry vocabulary semantics in the input text. For example, the natural language text of the original operation ticket is input (such as "open the 110kV line A circuit breaker and close the line B disconnector"); then RoBERTa is used to perform word segmentation and context encoding on this text to generate an embedding vector sequence E = [e1, e2, ..., e n ] (i.e., the first embedding vector sequence of step S21), each vector dimension is D, which is used to represent the word context.

[0067] ② Design a structured prompt-tuning mechanism. By designing prompt templates (PromptTemplates) for power grid tasks, the "action-device-condition" triples in the operation ticket are embedded into the input vector, guiding the model to focus on key elements. For example, a structured prompt template (such as "action-device-condition") is designed, and then the triple template is embedded into the original text or the vector sequence output by RoBERTa (i.e., E above), forming a structure-guided semantic vector sequence E' (i.e., the second embedded vector sequence in step S22).

[0068] ③ In the output layer, the BiLSTM (bidirectional long short-term memory) model is used to annotate the semantic vector sequence E' with entity labels to capture action verbs, device names, logical sequence words, etc. For example, E' is directly input into the BiLSTM network to extract the context dependency information and obtain the enhanced semantic feature vector sequence H = [h1,h2,...,h n] (i.e., the semantic feature vector sequence in step S23). Furthermore, when labeling entities, a span-based extraction strategy can be introduced to enhance the model’s ability to recognize multi-layered nested and composite structural entities (e.g., “isolating switch on busbar 1”), thereby addressing the complex structure and strong ambiguity of natural language expressions in the power industry.

[0069] In order to further improve the consistency and accuracy of the label prediction path, the present invention can also introduce conditional random fields (CRF) in the output layer to perform joint sequence modeling. The loss function of the conditional random field in the training phase is defined as follows:

[0070] in, represents the overall loss function, T t is the length of the input sequence, x t is the input vector of the t-th word (generated by RoBERTa+Prompt encoder), y t is the true label of position t in the label sequence, ψ t (y t ,x t ) represents the emission score at position t, φ(y t-1 ,y t ) is the transfer score between labels, Represents the set of all possible label paths.

[0071] ④CRF sequence label decoding: Apply the CRF decoder to the above enhanced semantic feature vector sequence H and output the optimal label path, such as B-Action, B-Device, B-Condition, etc., to form the annotated entity label sequence S:

[0072] For example: S = {("open", action), ("110kV line A circuit breaker", equipment), ("close", action), ("line B disconnector", equipment)}.

[0073] ⑤ After decoding, structural processing is required. Based on the above label sequence S, the entity labels are combined into standardized triples, for example:

[0074] {(action: open, device: circuit breaker A, condition: no load)},

[0075] {(action: close, device: disconnector B, condition: linkage)}; finally forming the operation sequence L in step S24 = [T1, T2, ..., T n ].

[0076] Compared to existing technologies, the multi-model fusion architecture proposed in this paper has stronger semantic integrity preservation and operation chain logic recognition capabilities. In a test set of internal real-world operation tickets, the named entity recognition F1 score reached 91.3%, significantly outperforming the general BERT-NER solution (approximately 84.7%).

[0077] Step S3: Based on a preset knowledge graph, the operation sequence is inferred and matched to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system.

[0078] Furthermore, step S3 can be implemented by the following preferred implementation, including steps S31-S32, as follows:

[0079] S31: For each operation node in the operation sequence, construct a corresponding executable path set according to the knowledge graph;

[0080] S32: Match and score the operation sequence with each executable path in the executable path set, identify risky operations, and generate a risk step set.

[0081] In this preferred embodiment, an executable path set is constructed for each operation node, and a compliant operation chain is generated using the physical layer topology and procedural layer constraints of the knowledge graph; the degree of deviation between the operation steps and the safety regulations is quantified through a scoring function to achieve dynamic risk grading in multi-device linkage scenarios.

[0082] Furthermore, the construction of the knowledge graph can be achieved through the following preferred implementation, including steps S33-S35, as follows:

[0083] S33: Acquire power equipment information in the power system; wherein the power equipment information includes each device entity, device attributes, operating status and corresponding operating procedures;

[0084] S34: Using the device entities, operation actions, and operating states in the power device information as nodes, and the semantic relationships between the nodes in the power device information as edges of a graph, to construct a knowledge graph; wherein the semantic relationships include: device connection relationships, device operation relationships, state association relationships, and procedure constraint relationships between the nodes, and each edge is also assigned a corresponding edge weight;

[0085] S35: Update the knowledge graph according to a preset incremental subgraph algorithm. The algorithm is as follows:

[0086] in, represents the structure of the knowledge graph at time t, ΔV t and ΔEt They represent the newly added node set and edge set at time t respectively.

[0087] In this preferred embodiment, by taking power equipment entities, operating actions and operating status as knowledge graph nodes and constructing an initial graph with semantic relationships as edges, a unified expression of equipment topology and procedural logic is achieved; further, the graph structure is dynamically updated through an incremental subgraph algorithm, and only newly added nodes and edges are merged to avoid the computing resource consumption of full reconstruction and adapt to real-time changes in power grid equipment configuration; the preset mechanism of edge weights provides a differentiated path priority basis for subsequent dynamic reasoning, supporting the refined scoring of risk steps.

[0088] Furthermore, in the knowledge graph, the edge weight corresponding to each edge can be calculated by the following preferred implementation method, as follows:

[0089] According to the preset graph attention network, the weight of each edge is calculated. The algorithm is as follows:

[0090] Among them, α ij represents the attention weight between node i and adjacent node j, and are the eigenvectors corresponding to nodes i, j, and k respectively, W is the linear transformation matrix, is the attention vector, T represents the transposition operation, is the set of adjacent nodes of node i, and || represents the vector concatenation operation.

[0091] In this preferred embodiment, edge weights are calculated through a graph attention network (GAT), and the node feature vectors are interactively weighted using attention vectors and linear transformation matrices to dynamically strengthen the association between key equipment (such as the operational dependency between circuit breakers and busbars). The dynamic screening mechanism of the adjacent node set is combined with real-time status to filter invalid connection edges (such as disconnected disconnectors), reducing redundant calculations and improving the real-time response efficiency and path selection accuracy of knowledge graph reasoning.

[0092] When constructing a knowledge graph, we first extract basic equipment information from the power grid dispatching system and equipment inventory systems (such as SCADA and EMS), including equipment name, type, wiring relationship, operating status, etc., as well as operating procedure text, including procedure number, operation sequence, and logical conditions. Then, for each piece of equipment data, we use a rule engine or pre-trained text extraction model to generate semantic triples in RDF format, for example:

[0093] (Circuit breaker A, belongs to, line 110kV),

[0094] (isolating switch B, status, closed),

[0095] (Step 1, Operation, Circuit Breaker A).

[0096] In an RDF triple, the three elements are called subject, predicate, and object. When constructing a graph, the subject and object can be directly used as graph nodes and defined in the following way, for example:

[0097] Equipment nodes: such as "circuit breaker", "switch", "transformer", etc.;

[0098] Action nodes: such as "close" and "open";

[0099] Status node: such as "closed", "open";

[0100] Among these nodes, each node is accompanied by corresponding attribute labels, including: device type, operating status, topology level and site to which it belongs.

[0101] When constructing edges between nodes, the predicate in each triple can be used as the edge type in the graph. For example, for multiple semantic connections between similar nodes (such as "connection", "control" and "operation"), different edge categories are defined respectively. In addition, each edge also records information such as upstream and downstream relationships, procedure sources and priorities.

[0102] Furthermore, after constructing the knowledge graph, we can also introduce the graph attention network (GAT) to calculate the edge weights between nodes to express the degree of influence of the edge on the reasoning result. The calculation method is as above α ij The expression can also be expressed equivalently as: The calculated weight values ​​will be used in subsequent reasoning path selection and conflict detection tasks.

[0103] After the graph is built, if an update is needed, a combination of incremental subgraph learning and a change monitoring mechanism can be used. In specific implementation, by monitoring equipment change events in the power grid system (such as newly connected equipment, maintenance outage status, etc.), the update of the local subgraph of the graph can be automatically triggered to avoid system pauses and redundancy caused by full reconstruction. The specific method is: using the structured event stream (Structured Event Stream) to parse the equipment change type; then automatically updating the graph incrementally based on the change type. The update method is as follows:

[0104] After the graph is acquired, the operation sequence can be inferred and matched. The operation sequence can be expressed as: L = {T1, T2, ..., T n Then, according to step S31, an executable path set is constructed, which can be expressed as: P = {p1, p2, ..., pm}; Next, the operation sequence and the executable path set are inferred and matched as follows:

[0105] Score(T i )=λ1·Sim struct (T i ,p j )+λ2·Sim logic (T i ,p j ); λ1+λ2=1; where,

[0106] Sim struct Indicates the device structure matching degree, which can be expressed by the shortest path similarity between two steps or subgraph embedding similarity. logic It indicates the matching degree of the operation sequence, state constraints and procedure logic between steps, which can be compared based on the rule tree. λ1 and λ2 represent weighting coefficients.

[0107] When matching scores, a threshold θ can be set. When Score(T i )<θ, the step is marked as a high-risk operation and recorded in the risk step set: R={T i ∣Score(T i )<θ}.

[0108] Step S4: performing conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result.

[0109] Furthermore, step S4 can be implemented by the following preferred implementation, including steps S41-S42, as follows:

[0110] S41: constructing a corresponding operation diagram for each operation step in the risk step set, and matching it with a preset conflict template to obtain a conflict detection result;

[0111] S42: For the operation diagrams that are not successfully matched with the conflict template, each operation diagram is encoded according to a preset graph neural network to obtain an embedding vector, and the vector similarity is calculated based on the embedding vector and a preset reference vector to obtain a conflict detection result.

[0112] In this preferred embodiment, conflict subgraph template matching is used to quickly locate known illegal operations (such as "grounding state + closing action"). For operation graphs that do not match the template, a graph neural network is used to encode vectors and calculate anomaly scores to quantify their semantic deviation from the baseline compliant operation, enabling the proactive discovery of new conflicts.

[0113] When implementing conflict template matching, each high-risk step is first constructed into an operation graph G op (T i ), the conflict template can be expressed as: Then perform template matching, which is expressed as: Mark as known conflict types; for those that are not successfully matched, use a graph neural network (such as GCN) to encode the operation steps into a vector and calculate the conflict detection score:

[0114] in, is the embedding mean of the normal sample (i.e., the reference vector), Indicates that step T is processed by a graph neural network (such as GCN) i The corresponding operation graph is encoded and then the embedding vector is obtained. If the score exceeds the threshold δ, it is determined to be a potential new conflict.

[0115] Step S5: Generate corresponding optimization suggestions based on the preset generative model, the knowledge graph and the conflict detection results, and complete the operation ticket verification.

[0116] Furthermore, step S5 can be implemented by the following preferred implementation, including steps S51-S52, as follows:

[0117] S51: Based on the conflict detection result, a plurality of candidate operation sequences are generated using a preset generative model and the knowledge graph;

[0118] S52: Based on the knowledge graph and the operation sequence, the semantic similarity and procedure consistency score of each candidate operation sequence are calculated, and the candidate operation sequence corresponding to the best score is output as an optimization suggestion to complete the operation ticket verification.

[0119] In this preferred embodiment, multiple versions of candidate operation sequences are generated, and the optimal suggestions are screened through a scoring function of semantic similarity and procedural consistency, balancing the preservation of operation intent and the satisfaction of safety constraints to avoid secondary errors introduced by manual corrections.

[0120] In the specific implementation, based on the conflict detection results, the large model and knowledge graph are used to guide the generation of multiple semantically reasonable but structurally different candidate operation sequences, which are recorded as sets Among them, each sequence S i Represents a complete revised version of an operation ticket (i.e., a candidate operation sequence).

[0121] Then calculate the semantic and procedural consistency score, and introduce the scoring function as follows:

[0122] Score(S i )=α·Sim sem(S i ,S orig )+β·Conf(S i ); where S i represents a candidate operation sequence; S orig Indicates the operation sequence corresponding to the original operation ticket; Sim sem (S i ,S orig ) indicates S i and S orig The semantic similarity score is used to measure the degree of retention of the candidate operation sequence and the original operation ticket in terms of intent. It can be calculated based on the BERT vector cosine similarity. i ) represents the procedure consistency score, which indicates whether the candidate operation sequence meets the logical constraints and operation procedures of the power grid equipment (which can be obtained from the knowledge graph). α and β are weighted coefficients, and α+β=1.

[0123] Finally, the candidate operation sequence with the highest score among all candidates is selected as the optimization suggestion, which is expressed as:

[0124] Among them, S * represents the candidate operation sequence with the highest score, that is, the final recommended optimization suggestion. * It can be accompanied by a highlighted prompt of the modified location and an explanation of the reason for the adjustment, and support manual secondary confirmation and adjustment.

[0125] In summary, compared with the existing technology, the above embodiments of the present application have the following beneficial effects: by obtaining the power grid operation ticket and generating a structured operation sequence through its semantic analysis, the semantic understanding problem of non-standardized operation tickets is solved; by dynamically reasoning and matching the knowledge graph to generate a set of risk steps, it replaces the static rule base and adapts to the dynamic changes in the configuration of power grid equipment; based on the dual conflict detection mechanism of graph structure and semantic features, the verification coverage rate in the multi-device linkage operation scenario is improved; combining the generative model with the knowledge graph to generate optimization suggestions, a closed-loop verification process is realized, and the cost of manual intervention is reduced. The overall solution solves the core problem of poor adaptability of traditional methods to dynamic power grid scenarios due to the solidification of rules and lack of reasoning ability through the collaboration of dynamic analysis, knowledge reasoning and intelligent correction, and significantly improves the verification efficiency and safety.

[0126] Example 2:

[0127] Please refer to Figure 2 Based on the same inventive concept, an operation ticket verification system based on a knowledge graph disclosed in an embodiment of the present invention includes: a data acquisition module M1, a semantic parsing module M2, a reasoning and matching module M3, a conflict detection module M4, and an optimization output module M5;

[0128] The data acquisition module M1 is used to obtain the operation ticket of the power grid.

[0129] The semantic parsing module M2 is used to perform semantic parsing on the operation ticket to obtain a corresponding operation sequence.

[0130] Furthermore, the semantic parsing module M2 includes: a word segmentation encoding unit, a template fusion unit, a feature extraction unit and a decoding output unit;

[0131] The word segmentation encoding unit is used to perform word segmentation and context encoding on the operation ticket according to a preset semantic encoder to obtain a corresponding first embedding vector sequence;

[0132] The template fusion unit is used to fuse the preset structured prompt template with the embedding vector sequence to obtain a second embedding vector sequence after structure guidance;

[0133] The feature extraction unit is configured to input the second embedding vector sequence into a preset bidirectional long short-term memory network to extract context dependencies and obtain a semantic feature vector sequence;

[0134] The decoding output unit is used to input the semantic feature vector sequence into a preset CRF decoder to perform decoding processing to obtain an entity label sequence, and perform structured processing on the entity label sequence to obtain a corresponding operation sequence.

[0135] In this preferred embodiment, the operation ticket is segmented and context-encoded through a semantic encoder to generate a first embedding vector sequence, thereby solving the semantic discreteness problem of unstructured text. Furthermore, the structured prompt template is fused with the embedding vector sequence to guide the model to focus on key entities (such as actions and equipment) in the operation ticket, thereby enhancing the semantic capture accuracy of domain terms. The context dependencies are then extracted through a bidirectional long short-term memory network to improve the recognition robustness of nested entities (such as "circuit breaker B connected to bus A"). Finally, the semantic feature vectors are sequence-labeled through a CRF decoder to optimize the global consistency of entity labels, avoid conflicts in isolated labels, form a structured operation sequence, and provide a parsable logical unit for subsequent verification.

[0136] The reasoning and matching module M3 is used to perform reasoning and matching on the operation sequence according to a preset knowledge graph to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system.

[0137] Furthermore, the reasoning and matching module M3 includes: an executable path construction unit and a risk matching unit;

[0138] The executable path construction unit is configured to construct a corresponding executable path set for each operation node in the operation sequence according to the knowledge graph;

[0139] The risk matching unit is used to match and score the operation sequence with each executable path in the executable path set, identify risky operations, and generate a risk step set.

[0140] In this preferred embodiment, an executable path set is constructed for each operation node, and a compliant operation chain is generated using the physical layer topology and procedural layer constraints of the knowledge graph; the degree of deviation between the operation steps and the safety regulations is quantified through a scoring function to achieve dynamic risk grading in multi-device linkage scenarios.

[0141] Furthermore, the reasoning and matching module M3 further includes: a device information acquisition unit, a graph construction unit, and a graph update unit;

[0142] The device information acquisition unit is configured to acquire information about power devices in the power system; the power device information includes each device entity, device attributes, operating status, and corresponding operating procedures;

[0143] The graph construction unit is configured to construct a knowledge graph by using the device entities, operation actions, and operating states in the power device information as nodes and the semantic relationships between the nodes in the power device information as edges of the graph; wherein the semantic relationships include: device connection relationships, device operation relationships, state association relationships, and procedure constraint relationships between the nodes, and each edge is also assigned a corresponding edge weight;

[0144] The graph updating unit is used to update the knowledge graph according to a preset incremental subgraph algorithm. The algorithm is as follows:

[0145] in, represents the structure of the knowledge graph at time t, ΔV t and ΔE t They represent the newly added node set and edge set at time t respectively.

[0146] In this preferred embodiment, by taking power equipment entities, operating actions and operating status as knowledge graph nodes and constructing an initial graph with semantic relationships as edges, a unified expression of equipment topology and procedural logic is achieved; further, the graph structure is dynamically updated through an incremental subgraph algorithm, and only newly added nodes and edges are merged to avoid the computing resource consumption of full reconstruction and adapt to real-time changes in power grid equipment configuration; the preset mechanism of edge weights provides a differentiated path priority basis for subsequent dynamic reasoning, supporting the refined scoring of risk steps.

[0147] Furthermore, the graph construction unit further includes: a weight calculation subunit;

[0148] The weight calculation subunit is used to calculate the weight of each edge according to the preset graph attention network. The algorithm is as follows:

[0149] Among them, α ij represents the attention weight between node i and adjacent node j, and are the eigenvectors corresponding to nodes i, j, and k respectively, W is the linear transformation matrix, is the attention vector, T represents the transposition operation, is the set of adjacent nodes of node i, and || represents the vector concatenation operation.

[0150] In this preferred embodiment, edge weights are calculated through a graph attention network (GAT), and the node feature vectors are interactively weighted using attention vectors and linear transformation matrices to dynamically strengthen the association between key equipment (such as the operational dependency between circuit breakers and busbars). The dynamic screening mechanism of the adjacent node set is combined with real-time status to filter invalid connection edges (such as disconnected disconnectors), reducing redundant calculations and improving the real-time response efficiency and path selection accuracy of knowledge graph reasoning.

[0151] The conflict detection module M4 is used to perform conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result.

[0152] Furthermore, the conflict detection module M4 includes: a first conflict detection unit and a second conflict detection unit;

[0153] The first conflict detection unit is configured to construct a corresponding operation diagram for each operation step in the risk step set, and match the operation diagram with a preset conflict template to obtain a conflict detection result;

[0154] The second conflict detection unit is used to encode each operation diagram that has not been successfully matched with the conflict template according to a preset graph neural network to obtain an embedding vector, and calculate the vector similarity between the embedding vector and a preset reference vector to obtain a conflict detection result.

[0155] In this preferred embodiment, conflict subgraph template matching is used to quickly locate known illegal operations (such as "grounding state + closing action"); for operation graphs that do not match the template, the vector is encoded through the graph neural network and the anomaly score is calculated to quantify its semantic deviation from the benchmark compliant operation, thereby realizing the active discovery of new conflicts.

[0156] The optimization output module M5 is used to generate corresponding optimization suggestions based on the preset generative model, the knowledge graph and the conflict detection results, and complete the operation ticket verification.

[0157] Furthermore, the optimization output module M5 includes: a candidate generation unit and a screening unit;

[0158] The candidate generation unit is configured to generate a plurality of candidate operation sequences according to the conflict detection result through a preset generative model and the knowledge graph;

[0159] The screening unit is used to calculate the semantic similarity and procedure consistency score of each candidate operation sequence based on the knowledge graph and the operation sequence, and output the candidate operation sequence corresponding to the best score as an optimization suggestion to complete the operation ticket verification.

[0160] In this preferred embodiment, multiple versions of candidate operation sequences are generated, and the optimal suggestions are screened through a scoring function of semantic similarity and procedural consistency, balancing the preservation of operation intent and the satisfaction of safety constraints to avoid secondary errors introduced by manual corrections.

[0161] In summary, compared with the existing technology, the embodiments of the present application have the following beneficial effects: by obtaining the power grid operation ticket and generating a structured operation sequence through semantic analysis, the semantic understanding problem of non-standardized operation tickets is solved; by dynamically reasoning and matching the knowledge graph to generate a set of risk steps, it replaces the static rule base and adapts to the dynamic changes in the configuration of power grid equipment; based on the dual conflict detection mechanism of graph structure and semantic features, the verification coverage rate in the multi-device linkage operation scenario is improved; combining the generative model with the knowledge graph to generate optimization suggestions, a closed-loop verification process is realized, and the cost of manual intervention is reduced. The overall solution solves the core problem of poor adaptability to dynamic power grid scenarios caused by the rigidification of rules and lack of reasoning ability of traditional methods through the collaboration of dynamic analysis, knowledge reasoning and intelligent correction, and significantly improves the verification efficiency and safety.

[0162] Example 3:

[0163] Figure 3 The structure diagram of the operation ticket verification device based on the knowledge graph of this application is presented. Figure 3 As shown, the knowledge graph-based operation ticket verification device may include: a processor N1, a memory N2, a data interface N3 and a communication bus N4.

[0164] Among them: the processor N1, the memory N2, and the data interface N3 communicate with each other through the communication bus N4; the data interface N3 is used for data communication with other devices such as input devices or output devices; the processor N1 is used to execute the program N5, which can specifically execute the relevant steps in any of the above-mentioned knowledge graph-based operation ticket verification method embodiments.

[0165] Specifically, the program N5 may include program code, which includes computer-executable instructions.

[0166] Processor N1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the knowledge graph-based operation ticket verification device may be processors of the same type, such as one or more CPUs, or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0167] The memory N2 is used to store the program N5. The memory N2 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0168] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. In addition, the embodiments of the present application are not directed to any particular programming language.

[0169] Example 4:

[0170] An embodiment of the present invention also provides a computer-readable storage medium, which stores at least one executable instruction. When the executable instruction is run on an operation ticket verification device / system based on a knowledge graph, the operation ticket verification device / system based on the knowledge graph executes an operation ticket verification method based on a knowledge graph in any of the above method embodiments.

[0171] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. Similarly, in order to streamline the application and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the application, the various features of the embodiments of the application are sometimes grouped together into a single embodiment, figure, or description thereof. Wherein, the claims that follow the specific embodiment are hereby clearly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the application.

[0172] Those skilled in the art will appreciate that the modules in the devices of the embodiments can be adaptively changed and installed in one or more devices different from the embodiments. The modules, units, or components in the embodiments can be combined into one module, unit, or component, and furthermore, they can be divided into multiple submodules, subunits, or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive.

Claims

1. A method for verifying an operation ticket based on a knowledge graph, characterized in that: include: Obtaining an operation ticket for the power grid; Perform semantic analysis on the operation ticket to obtain a corresponding operation sequence; According to a preset knowledge graph, the operation sequence is reasoned and matched to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system; Perform conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result; Based on the preset generative model, the knowledge graph and the conflict detection results, corresponding optimization suggestions are generated to complete the operation ticket verification.

2. The method for verifying an operation ticket based on a knowledge graph according to claim 1, characterized in that: The semantic parsing of the operation ticket to obtain a corresponding operation sequence includes: According to a preset semantic encoder, the operation ticket is segmented and context-encoded to obtain a corresponding first embedding vector sequence; Fusing a preset structured hint template with the embedding vector sequence to obtain a second embedding vector sequence guided by the structure; Inputting the second embedding vector sequence into a preset bidirectional long short-term memory network to extract context dependencies and obtain a semantic feature vector sequence; The semantic feature vector sequence is input into a preset CRF decoder for decoding to obtain an entity label sequence, and the entity label sequence is structured to obtain a corresponding operation sequence.

3. The method for verifying an operation ticket based on a knowledge graph according to claim 2, characterized in that: The process of performing reasoning and matching on the operation sequence based on the preset knowledge graph to generate a set of risk steps includes: For each operation node in the operation sequence, construct a corresponding executable path set according to the knowledge graph; The operation sequence is matched and scored with each executable path in the executable path set, risky operations are identified, and a risk step set is generated.

4. The method for verifying an operation ticket based on a knowledge graph according to claim 3, wherein: The conflict detection is performed based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result, including: For each operation step in the risk step set, a corresponding operation graph is constructed and matched with a preset conflict template to obtain a conflict detection result; For the operation diagrams that fail to match the conflict template successfully, each operation diagram is encoded according to the preset graph neural network to obtain an embedding vector, and the vector similarity is calculated based on the embedding vector and the preset reference vector to obtain the conflict detection result.

5. The method for verifying an operation ticket based on a knowledge graph according to claim 4, characterized in that: The process of generating corresponding optimization suggestions based on the preset generative model, the knowledge graph, and the conflict detection results to complete the operation ticket verification includes: According to the conflict detection result, a plurality of candidate operation sequences are generated through a preset generative model and the knowledge graph; Based on the knowledge graph and the operation sequence, the semantic similarity and procedure consistency score of each candidate operation sequence are calculated, and the candidate operation sequence corresponding to the best score is output as an optimization suggestion to complete the operation ticket verification.

6. The operation ticket verification method based on knowledge graph according to any one of claims 1 to 5, characterized in that: The knowledge graph is constructed based on the information of power equipment in the power system, including: Acquire power equipment information in the power system; wherein the power equipment information includes each device entity, device attributes, operating status and corresponding operating procedures; The device entities, operation actions, and operating states in the power equipment information are used as nodes, and the semantic relationships between the nodes in the power equipment information are used as edges in the graph to construct a knowledge graph; wherein the semantic relationships include: device connection relationships, device operation relationships, status association relationships, and procedure constraint relationships between the nodes, and each edge is also assigned a corresponding edge weight; The knowledge graph is updated according to the preset incremental subgraph algorithm. The algorithm is as follows: in, represents the structure of the knowledge graph at time t, ΔV t and ΔE t They represent the newly added node set and edge set at time t respectively.

7. The method for verifying an operation ticket based on a knowledge graph according to claim 6, characterized in that: Each of the edges is also equipped with a corresponding edge weight, including: According to the preset graph attention network, the weight of each edge is calculated. The algorithm is as follows: Among them, α ij represents the attention weight between node i and adjacent node j, and are the eigenvectors corresponding to nodes i, j, and k respectively, W is the linear transformation matrix, is the attention vector, T represents the transposition operation, is the set of adjacent nodes of node i, and || represents the vector concatenation operation.

8. An operation ticket verification system based on knowledge graph, characterized in that: include: Data acquisition module, semantic parsing module, reasoning and matching module, conflict detection module and optimization output module; Wherein, the data acquisition module is used to obtain the operation ticket of the power grid; The semantic parsing module is used to perform semantic parsing on the operation ticket to obtain a corresponding operation sequence; The reasoning and matching module is used to perform reasoning and matching on the operation sequence according to a preset knowledge graph to generate a set of risk steps; wherein the knowledge graph is constructed based on the power equipment information in the power system; The conflict detection module is used to perform conflict detection based on the graph structure and semantic features corresponding to each risk step in the risk step set to obtain a conflict detection result; The optimization output module is used to generate corresponding optimization suggestions based on the preset generative model, the knowledge graph and the conflict detection results, and complete the operation ticket verification.

9. An operation ticket verification device based on a knowledge graph, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into the processor, the steps of the operation ticket verification method based on the knowledge graph according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the steps of the operation ticket verification method based on the knowledge graph according to any one of claims 1 to 7 are implemented.

Images