Construction method of power distribution operation three-dimensional virtual model library based on domain knowledge graph

Abstract

The application relates to the technical field of power distribution operation simulation, and discloses a power distribution operation three-dimensional virtual model library construction method based on a domain knowledge graph. The application realizes real-time interaction feedback by constructing a power distribution domain knowledge graph and performing semantic enhancement, performing semantic annotation and feature extraction on a three-dimensional model, constructing an intelligent model library and maintaining consistency, intelligently generating an operation scene and optimizing compliance, and adopting a semantic topology energy quantization algorithm to establish entity and model semantic mapping and using a process constraint optimization function to evaluate scene quality; the application solves the problems of missing semantic correlation of a traditional model library, low efficiency of scene generation and poor compliance, improves model retrieval accuracy, scene generation efficiency and virtual training effectiveness, and is suitable for power distribution operation virtual simulation and personnel training.

Description

Technical Field

[0001] This invention relates to the field of power distribution operation simulation technology, and in particular to a method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs. Background Technology

[0002] Currently, the construction of 3D virtual model libraries for power distribution operations largely relies on traditional 3D modeling and basic data management technologies. The industry commonly uses tools such as CAD and Blender to complete the 3D geometric modeling of power distribution equipment and operational scenarios. Model storage primarily utilizes relational databases or file servers, categorizing and archiving models according to dimensions such as equipment type, voltage level, and operational scenario. To improve retrieval efficiency, some solutions add basic attribute tags to the models, supporting fuzzy queries based on keywords. While a few studies have introduced knowledge graph technology, it is only used for the structured input of equipment attributes, failing to achieve deep integration of domain knowledge with the geometric features and operational constraints of 3D models. Scene generation still mainly relies on manual drag-and-drop model assembly or preset template matching, lacking the ability to dynamically adapt to the logic of power distribution operations.

[0003] Existing technologies have significant shortcomings, making it difficult to meet the needs of virtual training and practical application in power distribution operations: First, the semantic association of the model library is lacking. Traditional classification and attribute retrieval cannot establish a deep mapping between domain knowledge and 3D models, resulting in low model retrieval accuracy and poor cross-scenario reusability. Second, scenario generation and compliance verification are disconnected. Manual assembly or template matching is not only inefficient but also cannot automatically verify whether the generated scenarios comply with power distribution safety regulations, which can easily lead to safety hazards due to human negligence. Third, there is a lack of a knowledge and model collaborative update mechanism. When power distribution operation procedures and equipment technical standards are updated, it is necessary to manually adjust the semantic tags and associations of the models one by one, which is not only time-consuming and labor-intensive but also prone to inconsistencies between knowledge and models. Summary of the Invention

[0004] The technical problem to be solved by this invention is that existing technologies suffer from low efficiency in building 3D model libraries, lack of semantic information, unintelligent scene generation, and difficulty in ensuring compliance. To address this, we propose a method for building a 3D virtual model library for power distribution operations based on domain knowledge graphs.

[0005] To achieve the above objectives, this application adopts the following technical solution: a method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs, comprising the following steps:

[0006] Step 1: Domain Knowledge Graph Construction and Semantic Enhancement: Extract key entities, attributes, and relationships from power distribution operation procedures, equipment standards, and historical work orders to construct a domain knowledge graph; learn distributed vector representations of entities and relationships through graph embedding technology to capture deep semantic associations; establish dynamic graph updates, and achieve incremental learning and semantic expansion based on newly added specifications and data to ensure timeliness and completeness;

[0007] Step 2: Semantic annotation and feature extraction of 3D model: Analyze the geometric structure and topological features of the 3D model of power distribution operation to generate a digital description; perform cross-modal alignment between the semantic vectors of knowledge graph entities and the geometric feature vectors of the model to establish a semantic mapping relationship; based on this relationship, automatically annotate the functional roles, safety attributes and operational constraints of the model to form semantic 3D model primitives;

[0008] Step 3: Building and Maintaining the Intelligent Model Library: Integrate the semantically annotated 3D models to build an intelligent model library with semantic retrieval and reasoning capabilities; use a graph database to store and manage the semantic relationships between models, supporting querying and reasoning; establish a linkage between the model library and the knowledge graph to achieve synchronous updates and consistency verification of semantic attributes;

[0009] Step 4: Intelligent generation and optimization of work scenarios: Analyze work orders or training task requirements, extract work elements and safety constraints; generate work processes and layout schemes that comply with safety regulations based on knowledge graph reasoning, retrieve matching 3D models and assemble the scenario; evaluate and optimize the preliminary scenario to ensure that the logic, safety and visual effects meet the requirements of power distribution work.

[0010] Step 5: Interaction and Feedback: Track user operation behavior sequences in the virtual training environment, conduct real-time compliance checks and risk assessments based on security rules and operating procedures in the knowledge graph; push alerts and operation guidance in real time when risky or illegal operations are detected; record and analyze training data to optimize user profiles and personalized training strategies.

[0011] Preferably, the domain knowledge graph construction in step one specifically includes:

[0012] Identify and extract power distribution equipment entities, operation action entities, safety specification entities, and environmental entities from unstructured power industry text;

[0013] Analyze and define the types of relationships between entities, including functional associations, security dependencies, temporal logical relationships, and spatial relationships;

[0014] Construct a graph structure with entities as nodes and relations as edges, and use graph neural networks for semantic embedding learning to generate low-dimensional dense vectors of entities and relations;

[0015] Establish version management based on timestamps and data sources to enable traceable incremental updates of the knowledge graph.

[0016] Preferably, the semantic annotation of the 3D model in step two specifically includes:

[0017] Analyze the geometric mesh, texture mapping, and skeletal animation data of the 3D model to extract visual and structural features;

[0018] Calculate the similarity between entity semantic vectors and 3D model feature vectors in a shared semantic space within a knowledge graph;

[0019] Based on the similarity measurement results, functional semantic tags, security level tags, and behavioral logic tags are automatically assigned to the 3D model;

[0020] The confidence level of the annotation results is assessed, and a manual review process is initiated for annotations with low confidence.

[0021] Preferably, the construction of the intelligent model library in step three specifically includes:

[0022] Establish a hierarchical index structure based on semantic similarity to achieve rapid retrieval and localization of 3D models;

[0023] Design the access interface for the model library to support fuzzy queries based on natural language descriptions and precise queries based on semantic graphs;

[0024] Monitor knowledge graph change events and formulate strategies for synchronizing and updating semantic attributes of the model library;

[0025] Regularly perform integrity scans and consistency checks on the model library to identify and fix semantic breaks or logical conflicts.

[0026] Preferably, the intelligent generation of the job scenario in step four specifically includes:

[0027] Natural language processing is performed on the input work order to identify the operation object, work steps, safety measures and environmental conditions;

[0028] Infer the work path and equipment layout constraints that meet the work order requirements and comply with safety regulations from the knowledge graph;

[0029] Based on the reasoning results, the required 3D model is retrieved from the model library, and its initial position and orientation in the virtual space are determined.

[0030] The assembled scene is rendered, optimized, and its logic verified to output the final 3D virtual work scene.

[0031] Preferably, the real-time interaction and dynamic feedback in step five specifically include:

[0032] Capture user commands and movement trajectories on device models within a virtual environment;

[0033] Real-time matching and comparison of user operation sequences with predefined compliance operation processes in the knowledge graph;

[0034] When the comparison result exceeds the safety threshold, a graded early warning message and specific corrective action recommendations are generated and pushed out.

[0035] By aggregating and analyzing users' historical operation data, a personalized capability model is constructed, and the difficulty and focus of training content are dynamically adjusted accordingly.

[0036] Preferred options also include:

[0037] The entire process of intelligent construction methods is monitored for performance, recording knowledge graph query response time, model retrieval accuracy, scene generation time, and user operation recognition accuracy.

[0038] Based on monitoring data, a system performance baseline is established, and the allocation of computing resources and algorithm parameters for each processing stage are dynamically adjusted.

[0039] Establish an anomaly handling mechanism to diagnose, record, and recover from abnormal states that occur during system operation.

[0040] Preferably, the semantic mapping relationship establishment in step two adopts a model association quantification algorithm based on semantic topological energy. This algorithm calculates the entity association in the knowledge graph using the following formula. With 3D model Semantic association strength between :

[0041] ;

[0042] in: This represents the core entity nodes in the knowledge graph of the power distribution operation field, including but not limited to power distribution equipment entities, operation action entities, safety standard entities, and environmental entities; It is a 3D model in the power distribution operation 3D model library, covering power distribution equipment models, operation scene environment models and tool models, and has geometric mesh, texture mapping and skeletal animation data features; For entities Low-dimensional dense embedding vectors in a shared semantic space are used to represent the deep semantic information of entities; For 3D models Low-dimensional dense embedding vectors in the shared semantic space are used to characterize the geometric structure and functional semantic fusion features of the model; For entity-based With model feature dimension The corresponding adaptive weight coefficients constitute A weight matrix with weights dynamically adjusted based on the semantic importance of the corresponding feature dimensions; For entities in a knowledge graph To model The shortest distance of the associated path is a non-negative integer, and the path is constructed based on the semantic relationship between the entity and the model; and The attenuation coefficients for semantic distance and topological distance are respectively determined through experimental optimization based on the characteristics of the power distribution operation field; For the improved Sigmoid function, ,in The Gaussian error function takes as input the combined result of the preceding operations and outputs in the range [0,1]. Its function is to normalize the semantic association strength to the standard scoring interval, which facilitates the horizontal comparison of the association strength of different entity model pairs.

[0043] Preferably, the scene optimization in step four employs a process-constraint-based scene generation optimization function, which evaluates and optimizes the generated virtual scene using the following formula. Overall quality :

[0044] ;

[0045] in: A three-dimensional virtual scene of power distribution operations to be evaluated and optimized, including complete elements such as equipment models, spatial layout, and operation process logic in the scene; For the first The vector field represents the degree of constraint satisfaction for each process constraint. The constraint types include operation timing constraints, equipment space layout constraints, and functional association constraints. Each component of the vector field represents the degree of constraint satisfaction at the corresponding position or step. For constraint satisfaction function based on process logic, The input is the gradient divergence value; For the scene The Middle The state vectors of key task nodes, with dimensions determined according to node type, represent the actual running or operational state of the nodes in the scenario; For the first The target status of each key operation node is determined based on the power distribution operation procedures and work order requirements, and represents the compliance status that the node should achieve. This is the state deviation penalty function. ,in It is a hyperbolic secant function. Sensitivity parameters are set according to the operational safety level; For the first The penalty coefficient for each type of safety violation is set according to the severity of the violation. For the scene The indicator function for violations of security rules is based on security constraints and compliance check rules.

[0046] Preferred options also include:

[0047] The optimized 3D virtual scene is output to the display device through a graphical interface, and scene editing tools are provided to users to support manual fine-tuning and confirmation of the automatically generated scene. This ensures that the final output virtual training scene strictly complies with power distribution operation specifications while flexibly adapting to the user's specific training needs and operating habits.

[0048] The technical effects and advantages of this invention are as follows:

[0049] This invention integrates knowledge from the power distribution field with a 3D model and establishes a precise semantic mapping between entities and the model using a semantic topological energy quantization algorithm. This solves the problems of low retrieval accuracy and poor cross-scenario reusability caused by the lack of semantic association in traditional model libraries. Furthermore, through knowledge-driven intelligent generation of work scenarios and process constraint optimization functions, it achieves integrated automatic scenario assembly and compliance verification, significantly improving scenario generation efficiency and avoiding the safety hazards of manual assembly. This significantly enhances the effectiveness and safety of virtual training for power distribution operations, providing efficient, intelligent, and compliant technical support for virtual simulation of power distribution operations, personnel training, and safety management. Attached Figure Description

[0050] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0051] Figure 1 This is a schematic diagram of the process of the present invention;

[0052] Figure 2 This is a schematic diagram of the topology of the present invention. Detailed Implementation

[0053] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0054] Reference Figures 1-2This invention provides a method for constructing a three-dimensional virtual model library for power distribution operations based on a domain knowledge graph. It realizes intelligent mapping from knowledge to models, semantic management of the model library, intelligent generation and optimization of operation scenarios, and real-time feedback of user operations, thereby providing an efficient, safe, and intelligent virtual training environment for power distribution operations.

[0055] The overall process of this invention includes the following core steps: domain knowledge graph construction and semantic enhancement, semantic annotation and feature extraction of 3D models, intelligent model library construction and maintenance, intelligent generation and optimization of job scenarios, and real-time interaction and dynamic feedback. Furthermore, this invention also includes performance monitoring of the entire process and the output and fine-tuning of the final scenario.

[0056] Example 1: Domain Knowledge Graph Construction and Semantic Enhancement, and Semantic Annotation of 3D Models:

[0057] This embodiment describes in detail steps one and two of the method of the present invention.

[0058] Step 1: Domain Knowledge Graph Construction and Semantic Enhancement

[0059] First, natural language processing technology is used to identify and extract data from massive amounts of unstructured text data, such as power distribution operation procedures, equipment technical standards, and historical work orders.

[0060] Physical electrical equipment, such as circuit breakers, disconnect switches, transformers, etc.

[0061] The physical actions involved in the operation, such as closing the circuit breaker, opening the circuit breaker, and connecting the grounding wire;

[0062] Safety regulations include physical provisions such as "test for electricity before connecting the grounding wire" and "one person operates while one person supervises".

[0063] And environmental entities, such as weather, temperature, humidity, etc.;

[0064] Next, the types of relationships between these entities are analyzed and defined, including:

[0065] Functional relationships, such as the functional relationship between "circuit breaker" and "opening and closing operation";

[0066] Safety dependencies, such as "connecting the grounding wire" depending on the safety prerequisite of "voltage testing";

[0067] The timing logic relationship, such as "voltage testing" must be performed before "grounding wire connection";

[0068] Spatial location relationships, such as the spatial inclusion relationship between "transformer" and "high-voltage room";

[0069] Then, a graph structure with entities as nodes and relations as edges is constructed. In order to capture the deep semantic association between entities and relations, graph embedding technology is used to learn the distributed vector representations of entities and relations, and generate low-dimensional dense vectors of entities and relations. These vectors can map discrete symbolic knowledge into a continuous vector space, so that semantically similar entities and relations are closer in the vector space, thereby providing quantitative semantic information for subsequent semantic annotation of 3D models.

[0070] Meanwhile, to ensure the timeliness and completeness of the knowledge graph, a dynamic update mechanism for the knowledge graph is established; version management based on timestamps and data sources is supported. When there are new work specifications, equipment standards or historical data, the system can automatically trigger incremental learning and semantic expansion processes to update the graph, ensuring that the knowledge graph always reflects the latest power distribution work knowledge.

[0071] Step 2: Semantic annotation and feature extraction of the 3D model:

[0072] This step aims to imbue the 3D model with semantic information from the domain knowledge graph, forming 3D model primitives rich in semantic information.

[0073] First, the geometric structure and topological features of the 3D model involved in power distribution work are analyzed. For example, vertex coordinates, facet normals, edge information, material textures, and skeletal animation data are extracted to generate a digital feature description of the model. These features can be multi-dimensional visual and structural features.

[0074] Next, a semantic mapping relationship is established from domain knowledge to the 3D model. Specifically, the semantic vectors of entities in the knowledge graph generated in step one are aligned across modalities with the geometric feature vectors of the 3D model. This alignment can be achieved through deep learning models, such as multimodal autoencoders or contrastive learning models, to map data from different modalities into a shared semantic space.

[0075] In the shared semantic space, a model association quantification algorithm based on semantic topological energy is used to compute entities in the knowledge graph. With 3D model Semantic association strength between The algorithm calculates using the following formula:

[0076] ;

[0077] in:

[0078] By integrating the semantic feature differences, topological association distance, and feature weights of entities and 3D models, the semantic matching degree between the two is quantified. The output range is [0,1], and the closer the value is to 1, the stronger the semantic association, providing a quantitative basis for establishing semantic mapping relationships.

[0079] This represents the core entity nodes in the knowledge graph of the power distribution operation field, including but not limited to power distribution equipment entities, operation action entities, safety standard entities, and environmental entities;

[0080] It is a 3D model in the power distribution operation 3D model library, covering power distribution equipment models, operation scene environment models and tool models, and has geometric mesh, texture mapping and skeletal animation data features;

[0081] For double summation operation, Corresponding entity Feature dimension index, Corresponding 3D model The feature dimension index, the total number of summation terms is the number of entity feature dimensions × the number of model feature dimensions, covering all combinations of feature components of the two;

[0082] For entities Low-dimensional dense embedding vectors in a shared semantic space are used to represent the deep semantic information of entities; For 3D models Low-dimensional dense embedding vectors in a shared semantic space are used to characterize the geometric structure and functional semantic fusion features of the model;

[0083] For entity-based With model feature dimension The corresponding adaptive weight coefficients constitute A weight matrix with weights dynamically adjusted based on the semantic importance of the corresponding feature dimensions;

[0084] For entities in a knowledge graph To model The shortest distance of the associated path is a non-negative integer, and the path is constructed based on the semantic relationship between the entity and the model;

[0085] and The attenuation coefficients for semantic distance and topological distance are respectively determined through experimental optimization based on the characteristics of the power distribution operation field;

[0086] For the improved Sigmoid function, ,in The Gaussian error function takes as input the combined result of the preceding operations and outputs in the range [0,1]. Its function is to normalize the semantic association strength to the standard scoring interval, which facilitates the horizontal comparison of the association strength of different entity model pairs.

[0087] Automatically assign semantic association strength to the 3D model:

[0088] Functional semantic tags, such as "circuit breaker";

[0089] Safety level labels, such as "High-voltage equipment" and "No live work allowed";

[0090] Behavioral logic labels, such as "can be opened and closed" and "requires voltage testing";

[0091] Finally, the confidence level of the annotation results is evaluated. For low-confidence annotations with semantic association strength below the preset threshold or with ambiguity, a manual review process is initiated to ensure the accuracy of the annotation.

[0092] Example 2: Construction and Consistency Maintenance of the Intelligent Model Library, and Intelligent Generation and Compliance Optimization of Work Scenarios:

[0093] This embodiment describes in detail steps three and four of the method of the present invention.

[0094] Step 3: Construction and Consistency Maintenance of the Intelligent Model Library:

[0095] First, the 3D models annotated semantically in step two are integrated to construct an intelligent model library with semantic retrieval and reasoning capabilities. Each 3D model in this library is rich in semantic information such as its functional role, safety attributes, and operational constraints in power distribution operations. To efficiently store and manage the semantic relationships between models, a graph database is used to store model entities, attributes, and their relationships. The advantage of a graph database is that it can natively support the storage and querying of graph structure data, thereby supporting efficient semantic querying and relational reasoning. For example, one can query "all circuit breaker models with high-voltage attributes and capable of opening and closing operations". To achieve rapid retrieval and location of 3D models, a hierarchical index structure based on semantic similarity is established. This index structure can perform multi-dimensional indexing based on the model's semantic tags, functional attributes, etc., thereby quickly locating the target model among massive models.

[0096] The design model library access interface supports fuzzy queries based on natural language descriptions, such as "find all switches used for high-voltage lines", and precise queries based on semantic graphs, such as "query equipment models directly associated with 'voltage testing' operations".

[0097] Establish a linkage mechanism between the model library and the domain knowledge graph; when the knowledge in the knowledge graph is updated, such as adding safety specifications or equipment parameters, the system can automatically trigger the synchronous update and consistency verification of the semantic attributes of the relevant models in the model library; for example, if the rated voltage of a device in the knowledge graph changes, the voltage attribute of the corresponding 3D model in the model library will also be automatically updated.

[0098] Regularly perform integrity scans and consistency checks on the model library to identify and repair semantic breaks, such as a model losing its proper semantic label, or logical conflicts, such as a model being labeled as both "charged" and "uncharged" at the same time, to ensure the accuracy and timeliness of the model library.

[0099] Step 4: Intelligent Generation and Compliance Optimization of Work Scenarios:

[0100] First, analyze the input work order content or training task requirements, and use natural language processing technology to identify the following:

[0101] The object being operated on, such as "Transformer No. 1";

[0102] Work procedures, such as "open the circuit breaker";

[0103] Safety measures, such as "grounding wire";

[0104] Environmental conditions, such as "operating in the rain".

[0105] Next, based on the knowledge graph, operational logic reasoning is performed to generate a work process and spatial layout scheme that complies with safety regulations. For example, based on the sequential logic that "voltage testing" must precede "grounding wire installation," the correct operational sequence is deduced. Based on the spatial relationships between equipment and safety distance requirements, a reasonable layout of the equipment in the virtual scene is determined. Then, semantic retrieval and retrieval of matching 3D models from the intelligent model library are performed, and the scene is automatically assembled based on the reasoning results. For example, based on the equipment type and quantity mentioned in the work order, the corresponding 3D model is retrieved from the model library and placed in the virtual scene according to the reasoned spatial layout scheme.

[0106] The initially generated scenario undergoes multi-dimensional evaluation and optimization to ensure it meets the requirements of power distribution operations in terms of logic, safety, and visual effects. Scenario optimization employs a scenario generation optimization function based on process constraints, which evaluates and optimizes the generated virtual scenario using the following formula. Overall quality :

[0107] ;

[0108] in:

[0109] The formula's overall function is to comprehensively quantify the compliance and adaptability of virtual scenarios from three dimensions: the degree of satisfaction of work process constraints, the degree of matching of key node states, and the penalty for safety violations. The output range is (0,1], and the closer the value is to 1, the higher the overall quality of the scenario, providing a quantitative target for scenario optimization.

[0110] This is a double product operation. The index corresponds to the work process constraints, with a total of N items, where N is the total number of work process constraints, such as operation sequence constraints, equipment layout constraints, etc. The index corresponds to the critical operation node, with a total of M items, where M is the total number of critical operation nodes, such as equipment operation nodes, safety inspection nodes, etc. The total number of product items is (N×M) items, covering all combinations of constraints and nodes; For summation operations, The index corresponds to the type of security violation, with a total of K items, where K is the total number of security violation types, such as operation order violation, safety distance violation, etc. The total number of items in the summation is K, covering all violation types;

[0111] The initial state of the generated virtual scene is determined by the automatic scene assembly result, which includes the complete elements of equipment models, spatial layout, and work process logic in the scene;

[0112] For the first The vector field represents the degree of constraint satisfaction for each process constraint. The constraint types include operation timing constraints, equipment space layout constraints, and functional association constraints. Each component of the vector field represents the degree of constraint satisfaction at the corresponding position or step. For example, whether an operation occurs at the correct time or whether a device is in the correct state.

[0113] For constraint satisfaction function based on process logic, The input is the gradient divergence value, and its function is to perform nonlinear quantization on the constraint satisfaction, which highlights the advantage of high satisfaction and suppresses extreme values ​​that deviate excessively from the reasonable range. The function value is maximized when the constraint is fully satisfied.

[0114] For the scene The Middle The state vector of a key operation node, the definition and state parameters of which are determined by the work order parsing and reasoning results, such as the open / closed state of a circuit breaker or the connected state of a grounding wire;

[0115] For the first The target status of each key operation node is determined based on the power distribution operation procedures and work order requirements, and represents the compliance status that the node should achieve.

[0116] This is the state deviation penalty function. ,in It is a hyperbolic secant function. Sensitivity parameters are set according to the operational safety level;

[0117] For the first The penalty coefficient for each type of safety violation is set according to the severity of the violation.

[0118] For the scene The indicator function for violations of security rules is determined based on security constraints and compliance check rules.

[0119] By maximizing this optimization function, the system can generate virtual scenarios that achieve optimal performance in terms of process logic, critical node states, and security compliance.

[0120] Example 3: Real-time interaction, dynamic feedback, and system performance monitoring:

[0121] This embodiment describes in detail step five of the method of the present invention, as well as system performance monitoring, scene output and fine-tuning;

[0122] Step 5: Real-time interaction and dynamic feedback:

[0123] First, in the virtual training environment, the user's operation commands on the device model in the virtual scene, such as clicking, dragging, and movement trajectory, are tracked in real time; this operation data includes the operation object, operation type, operation time, operation sequence, etc.

[0124] Next, the system performs real-time matching and comparison of the user's operation sequence with predefined compliant operation procedures in the knowledge graph. For example, it checks whether the user operates in the order of "verification - grounding - equipment operation". When the comparison result exceeds the safety threshold, indicating that the user's operation does not comply with safety regulations or poses a potential risk, the system can generate and push graded warning information and specific corrective action suggestions. For example, if the user attempts to ground the device without verifying the power supply, the system will immediately display a warning message saying "Please verify the power supply first" and highlight the verification tool. At the same time, the system records and analyzes training process data, including the user's operating habits, error types, and learning progress, to optimize user profiles and personalized training strategies. For example, for users who frequently make mistakes in a certain step, the system can automatically increase the training difficulty of that step or provide more auxiliary information.

[0125] By aggregating and analyzing users' historical operation data, a personalized capability model is built, and the difficulty and focus of training content are dynamically adjusted accordingly. For example, for users who are proficient in a certain operation, the basic training of that operation can be skipped and they can directly enter more complex scenarios.

[0126] System performance monitoring: To ensure the stable and efficient operation of the entire intelligent construction method, this invention also includes performance monitoring of the entire process of the intelligent construction method; specifically, it records key performance indicators such as knowledge graph query response time, model retrieval accuracy, scene generation time, and user operation recognition accuracy.

[0127] Based on monitoring data, a system performance baseline is established, and the allocation of computing resources and algorithm parameters for each processing stage are dynamically adjusted. For example, if the response time of a knowledge graph query is too long, the computing resources of the graph database can be increased; if the model retrieval accuracy decreases, the parameters of the semantic mapping model can be adjusted.

[0128] Establish an exception handling mechanism to diagnose, record, and recover from abnormal states that occur during system operation; for example, when a module crashes, the system can automatically restart the module and record the error log.

[0129] Scene Output and Fine-tuning: Finally, the optimized 3D virtual scene is output to a display device, such as a VR headset or large-screen monitor, via a graphical interface for immersive training. Simultaneously, a scene editing tool is provided to allow users to manually fine-tune and confirm the automatically generated scene. This allows users to adjust the device positions and environmental parameters within the scene according to specific training needs or personal habits, ensuring that the final output virtual training scene strictly complies with power distribution operation specifications while flexibly adapting to the user's specific training needs and operating habits.

[0130] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for constructing a 3D virtual model library for power distribution operations based on domain knowledge graphs, characterized in that, Includes the following steps: Step 1: Domain Knowledge Graph Construction and Semantic Enhancement: Extract key entities, attributes, and relationships from power distribution operation procedures, equipment standards, and historical work orders to construct a domain knowledge graph; By learning distributed vector representations of entities and relationships through graph embedding technology, deep semantic associations are captured; a dynamic graph update is established, and incremental learning and semantic expansion are achieved based on new specifications and data, ensuring timeliness and completeness; Step 2: Semantic annotation and feature extraction of 3D model: Analyze the geometric structure and topological features of the 3D model of power distribution operation to generate a digital description; perform cross-modal alignment between the semantic vectors of knowledge graph entities and the geometric feature vectors of the model to establish a semantic mapping relationship; based on this relationship, automatically annotate the functional roles, safety attributes and operational constraints of the model to form semantic 3D model primitives; Step 3: Building and Maintaining the Intelligent Model Library: Integrate the semantically annotated 3D models to build an intelligent model library with semantic retrieval and reasoning capabilities; use a graph database to store and manage the semantic relationships between models, supporting querying and reasoning; establish a linkage between the model library and the knowledge graph to achieve synchronous updates and consistency verification of semantic attributes; Step 4: Intelligent generation and optimization of work scenarios: Analyze work orders or training task requirements, extract work elements and safety constraints; generate work processes and layout schemes that comply with safety regulations based on knowledge graph reasoning, retrieve matching 3D models and assemble the scenario; evaluate and optimize the preliminary scenario to ensure that the logic, safety and visual effects meet the requirements of power distribution work. Step 5: Interaction and Feedback: Track user operation behavior sequences in the virtual training environment and conduct real-time compliance checks and risk assessments based on security rules and operating procedures in the knowledge graph; When a risk or violation is detected, an alert and operational guidance will be sent immediately. Record and analyze training data to optimize user profiles and personalized training strategies; The semantic mapping relationship establishment in step two employs a model association quantification algorithm based on semantic topological energy. This algorithm calculates the entity association in the knowledge graph using the following formula. With 3D model Semantic association strength between : ； in: This represents the core entity nodes in the knowledge graph of the power distribution operation field, including but not limited to power distribution equipment entities, operation action entities, safety standard entities, and environmental entities; It is a 3D model in the power distribution operation 3D model library, covering power distribution equipment models, operation scene environment models and tool models, and has geometric mesh, texture mapping and skeletal animation data features; For entities Low-dimensional dense embedding vectors in a shared semantic space are used to represent the deep semantic information of entities; For 3D models Low-dimensional dense embedding vectors in a shared semantic space are used to characterize the geometric structure and functional semantic fusion features of the model; For entity feature-based With model feature dimension The corresponding adaptive weight coefficients constitute A weight matrix with weights dynamically adjusted based on the semantic importance of the corresponding feature dimensions; For entities in a knowledge graph To model The shortest distance of the associated path is a non-negative integer, and the path is constructed based on the semantic relationship between the entity and the model; and The attenuation coefficients for semantic distance and topological distance are respectively determined through experimental optimization based on the characteristics of the power distribution operation field; For the improved Sigmoid function, ,in The Gaussian error function is used as the input, which is the combined result of the preceding operations. The output range is [0,1]. Its function is to normalize the semantic association strength to the standard scoring interval, so as to facilitate the horizontal comparison of the association strength of different entity model pairs. The scene optimization in step four employs a process-constraint-based scene generation optimization function, which evaluates and optimizes the generated virtual scene using the following formula. Overall quality ; ； in: A three-dimensional virtual scene of power distribution operations to be evaluated and optimized, including complete elements such as equipment models, spatial layout, and operation process logic in the scene; For the first The vector field represents the degree of constraint satisfaction for each process constraint. The constraint types include operation timing constraints, equipment space layout constraints, and functional association constraints. Each component of the vector field represents the degree of constraint satisfaction at the corresponding position or step. For constraint satisfaction function based on process logic, The input is the gradient divergence value; For the scene The Middle The state vectors of key task nodes, with dimensions determined according to node type, represent the actual running or operational state of the nodes in the scenario; For the first The target status of each key operation node is determined based on the power distribution operation procedures and work order requirements, and represents the compliance status that the node should achieve. This is the state deviation penalty function. ,in It is a hyperbolic secant function. Sensitivity parameters are set according to the operational safety level; For the first The penalty coefficient for each type of safety violation is set according to the severity of the violation. For the scene The indicator function for violations of security rules is based on security constraints and compliance check rules.

2. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, The construction of the domain knowledge graph in step one specifically includes: Identify and extract power distribution equipment entities, operation action entities, safety specification entities, and environmental entities from unstructured power industry text; Analyze and define the types of relationships between entities, including functional associations, security dependencies, temporal logical relationships, and spatial relationships; Construct a graph structure with entities as nodes and relations as edges, and use graph neural networks for semantic embedding learning to generate low-dimensional dense vectors of entities and relations; Establish version management based on timestamps and data sources to enable traceable incremental updates of the knowledge graph.

3. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, The semantic annotation of the 3D model in step two specifically includes: Analyze the geometric mesh, texture mapping, and skeletal animation data of the 3D model to extract visual and structural features; Calculate the similarity between entity semantic vectors and 3D model feature vectors in a shared semantic space within a knowledge graph; Based on the similarity measurement results, functional semantic tags, security level tags, and behavioral logic tags are automatically assigned to the 3D model; The confidence level of the annotation results is assessed, and a manual review process is initiated for annotations with low confidence.

4. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, The construction of the intelligent model library in step three specifically includes: Establish a hierarchical index structure based on semantic similarity to achieve rapid retrieval and localization of 3D models; Design the access interface for the model library to support fuzzy queries based on natural language descriptions and precise queries based on semantic graphs; Monitor knowledge graph change events and formulate strategies for synchronizing and updating semantic attributes of the model library; Regularly perform integrity scans and consistency checks on the model library to identify and fix semantic breaks or logical conflicts.

5. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, The intelligent generation of the work scenario in step four specifically includes: Natural language processing is performed on the input work order to identify the operation object, work steps, safety measures and environmental conditions; Infer the work path and equipment layout constraints that meet the work order requirements and comply with safety regulations from the knowledge graph; Based on the reasoning results, the required 3D model is retrieved from the model library, and its initial position and orientation in the virtual space are determined. The assembled scene is rendered, optimized, and its logic verified to output the final 3D virtual work scene.

6. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, The real-time interaction and dynamic feedback in step five specifically include: Capture user commands and movement trajectories on device models within a virtual environment; Real-time matching and comparison of user operation sequences with predefined compliance operation processes in the knowledge graph; When the comparison result exceeds the safety threshold, a graded early warning message and specific corrective action recommendations are generated and pushed out. By aggregating and analyzing users' historical operation data, a personalized capability model is constructed, and the difficulty and focus of training content are dynamically adjusted accordingly.

7. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, Also includes: The entire process of intelligent construction methods is monitored for performance, recording knowledge graph query response time, model retrieval accuracy, scene generation time, and user operation recognition accuracy. Based on monitoring data, a system performance baseline is established, and the allocation of computing resources and algorithm parameters for each processing stage are dynamically adjusted. Establish an anomaly handling mechanism to diagnose, record, and recover from abnormal states that occur during system operation.

8. The method for constructing a three-dimensional virtual model library for power distribution operations based on domain knowledge graphs according to claim 1, characterized in that, Also includes: The optimized 3D virtual scene is output to the display device through a graphical interface, and scene editing tools are provided to users to support manual fine-tuning and confirmation of the automatically generated scene. This ensures that the final output virtual training scene strictly complies with power distribution operation specifications while flexibly adapting to the user's specific training needs and operating habits.

Images