Intelligent training-oriented operation-monitoring business knowledge map dynamic construction method and system

Abstract

The method and system for dynamically constructing operation-monitoring business knowledge map for intelligent training comprise: a device entity layer, which collects multi-modal data, constructs a device health quantification model, combines a primary-secondary device attribute system and a correlation relationship, and forms a device entity knowledge map; a business logic layer, which analyzes multi-source rules, constructs a three-level subgraph of the device entity knowledge map, realizes mapping with the device entity layer, and adjusts a logic weight based on device health; a multi-modal resource layer, which integrates heterogeneous resources and constructs a cross-modal resource network; a disturbance scene generation layer, which divides three-level scenes of a healthy state, a warning state and a fault state, generates a composite training scene by injecting a disturbance factor; and a closed-loop optimization layer, which constructs a three-dimensional evaluation model, positions an optimization object based on an optimization priority, and drives autonomous iteration. The application can promote the deep integration of multi-modal resources and actual combat scenes, and improve the multi-system collaborative disposal capability and the complex scene response level.

Description

Technical Field

[0001] This invention belongs to the field of intelligent skills training technology for operation and maintenance monitoring personnel of power grid centralized control stations, and particularly relates to a method and system for dynamically constructing operation-monitoring business knowledge maps for intelligent training. Background Technology

[0002] In modern power systems, centralized control stations undertake the critical task of centralized monitoring and operation and maintenance management of multiple substations. The safety, stability, and efficiency of their operation are crucial to the reliable power supply of the entire power grid. As the core force in the operation of centralized control stations, operation and maintenance monitoring personnel need to possess comprehensive and in-depth business knowledge and excellent operational skills to cope with complex and ever-changing power grid operating conditions and various unexpected faults.

[0003] With the rapid development of intelligent power systems, the State Grid Corporation of China is deepening the construction of substation centralized control stations, which poses a severe challenge to the grassroots personnel's ability to master the full-dimensional cognition of primary equipment structure, secondary protection logic, and auxiliary system linkage.

[0004] (i) The complexity of centralized control station operation and monitoring is increasing exponentially, encompassing core knowledge across multiple dimensions, including real-time monitoring of equipment operating parameters, collaborative fault handling procedures, and dynamic verification of protection logic. The limitations of traditional training models are becoming increasingly apparent. Knowledge is scattered across multimodal resources and is not linked to the physical state of the equipment (state characteristics, operating parameters) or business logic (protection linkage, operation procedures), resulting in fragmented knowledge. For example, the existing technology with publication number CN114299799A proposes a simulation system suitable for distribution automation training, targeting multiple technical and business areas such as primary and secondary equipment in the distribution network, distribution automation systems, distribution automation terminal operation and maintenance, and communication protocols. This method does not consider linking knowledge to the physical state of the equipment and business logic.

[0005] (ii) Resources are statically fixed and difficult to dynamically adjust with equipment upgrades and procedure updates. Furthermore, equipment modeling suffers from problems such as insufficient state mapping, simplification of associations, and delayed updates. For example, the existing technology with publication number CN117540033A provides a method for constructing and training a power grid skills-assisted knowledge system based on artificial intelligence. By constructing a power grid skills-assisted learning and testing knowledge system library, it realizes the training and simulation testing of power grid-related knowledge. However, this method lacks the ability to generate multi-dimensional disturbance scenarios and is difficult to simulate the impact of equipment state changes on business logic and complex fault scenarios, resulting in a serious disconnect between training and actual practice. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for dynamically constructing a knowledge map of operation and monitoring business for intelligent training. By constructing a multi-dimensional interconnected knowledge map of "equipment entities - business logic - training resources" and integrating it with a capability evaluation system, it enables the generation of highly realistic training scenarios driven by equipment status and closed-loop training throughout the entire process. This solves core problems such as knowledge fragmentation and the disconnect between scenarios and actual practice in the refined management of centralized control stations, and improves the integration of training content with operation and monitoring business and the effectiveness of practical training.

[0007] The present invention adopts the following technical solution.

[0008] This invention proposes a dynamic knowledge map construction system for operation and monitoring business aimed at intelligent training, comprising an equipment entity layer, a business logic layer, a multimodal resource layer, a disturbance scenario generation layer, and a closed-loop optimization layer. At the equipment entity layer, multimodal data is collected and standardized to construct a quantitative model of equipment health. Combined with the primary and secondary equipment attribute system and their relationships, an equipment entity knowledge map is formed. In the business logic layer, multi-source rules are parsed using natural language processing technology to construct a three-level subgraph of the device entity knowledge map. A dynamic association engine is used to realize the mapping with the device entity layer, and the logical weights are dynamically adjusted based on the device health to make the rules adapt to changes in device status. The multimodal resource layer integrates heterogeneous resources including text, images, videos, and virtual 3D models. After normalization, it constructs a cross-modal resource network to realize the association between resources and device entities and business logic, as well as resource invocation and interaction. The disturbance scenario generation layer divides the scenarios into three levels: healthy state, warning state, and fault state, based on the equipment health status. By injecting three types of disturbance factors—physical, logical, and human factors—it generates composite training scenarios that match the difficulty of the scenarios with the equipment status and the trainees' abilities. The closed-loop optimization layer collects operational behavior data, system interaction data, and capability assessment data to construct a three-dimensional evaluation model. Based on optimization priorities, it identifies optimization targets and drives the autonomous iteration of the device entity layer, business logic layer, and disturbance scenario generation layer.

[0009] More preferably, in the device entity layer, the device health quantification model is as follows:

[0010] in, , , and These represent the temperature reliability index. Operation frequency fatigue index Defect Severity Index and environmental impact index The weighting coefficients are set according to the actual situation, and .

[0011] More preferably, the attribute system refers to a system with specific physical devices / devices as nodes, where each node contains three types of core information: a unique identifier, static attributes, and dynamic attributes. Static attributes include the inherent characteristics of primary equipment and the functional and technical characteristics of secondary equipment. Dynamic attributes include the real-time operating status of primary equipment and the operating status and functional output of secondary equipment. The relationships include those between primary devices, those between primary devices and secondary devices, and those between secondary devices.

[0012] More preferably, the device entity layer is updated based on three types of signals, including real-time operating data changes, power grid topology adjustments, and parameter configuration modifications, to update the device's dynamic attributes, relationships, and health status; the update is achieved through Neo4j's Cypher statements.

[0013] More preferably, in the business logic layer, the parsing process includes: The process involves word segmentation of technical terms related to equipment entity protection actions; identification of equipment entities, action instructions, and constraints within these technical terms; extraction of logical relationships between equipment entities, including the sequence of operation steps, triggering conditions, and prohibition conditions; and conversion of the extracted logical relationships into rule expressions. A three-level subgraph of the device entity knowledge map is constructed based on rule expressions, including the operation process subgraph, the protection logic subgraph, and the linkage strategy subgraph.

[0014] More preferably, the operation flow sub-graph takes the decomposition of typical operation steps as the core, breaks down the operation into independent logical nodes, connects the nodes through sequential execution relationship, and clarifies the preconditions and operation basis between nodes; The protection logic subgraph, based on the action mechanism of the protection device, extracts protection settings, action thresholds, and exit commands as nodes, connects the criterion nodes through a "logical AND / OR" relationship, and defines the associated attributes of delay time and return coefficient. The linkage strategy subgraph, designed for multi-device collaboration scenarios, uses triggering conditions, execution actions, and upgrade strategies as nodes, connecting conditions and actions through a "causal triggering" relationship.

[0015] More preferably, in the closed-loop optimization layer, the three-dimensional evaluation model uses scenario authenticity deviation as the basic indicator, logic validity deviation as the core indicator, and training fit deviation as the fit indicator. It calculates a weighted sum of scenario authenticity deviation, logic validity deviation, and training fit deviation to obtain the optimization priority.

[0016] More preferably, the scene realism deviation refers to the degree of consistency between the virtual device response and the on-site data; The logical validity deviation is characterized by the ratio of the number of misjudgments by the rule engine to the total number of triggers; The training fit deviation is calculated using the standard deviation of trainees' ability scores.

[0017] More preferably, when the optimization priority is higher than the set threshold, the dominant influencing factors are identified by the numerical proportions of scenario authenticity deviation, logic validity deviation, and training suitability deviation. The specific level in the device entity layer, business logic layer, or disturbance scenario layer corresponding to the dominant factor is taken as the primary optimization target. The optimization model is trained by calling the historical training data associated with the level, and the core parameters within the level are associated with the student's operation characteristics. The deviation value of the corresponding dimension is reduced by correcting the parameters until the P value drops below the threshold, thus completing this round of targeted iteration.

[0018] This invention also proposes a method for dynamically constructing an operation-monitoring business knowledge map for intelligent training, including: Step 1: Collect multimodal data from the time, physical, and state dimensions, including time series data of each device's operation, environmental parameters during operation, secondary device parameters, full life cycle data, and constraint data that the device meets during operation. Step 2: Construct a device health measurement model based on multimodal data; the device health measurement model outputs health levels, including healthy state, warning state, and fault state; Step 3: Based on the power grid model and topology, extract equipment entity nodes, define attributes and relationships; based on the equipment entity nodes, defined attributes and relationships, transform the collected multimodal data into an equipment entity knowledge map in a graph database; Step four: Use parsing methods to process the rule text, construct rule expressions for device entities and their relationships, and perform structured modeling of the device entity knowledge map based on the rule expressions to obtain a three-level sub-graph, including an operation flow sub-graph, a protection logic sub-graph, and a linkage strategy sub-graph; trigger the logical rules of the structured modeled device entity knowledge map based on the device status and device health level. Step 5: Integrate text resources, image resources, video resources, and virtual 3D model resources as multimodal resources and perform normalization processing; construct a multimodal network based on the multimodal resources; intelligently call upon the multimodal resources in the multimodal network according to the training scenario requirements and user operation behavior; Step 7: Construct an association evaluation model based on multimodal interaction data, and realize the evolution of the device entity knowledge map based on the output value of the association evaluation model.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a multi-dimensional knowledge map of "equipment entity - business logic - training resources", and integrates all elements of equipment, business rules and multi-modal resources to form a three-dimensional knowledge network. This solves the problems of fragmented entity knowledge and disconnected business logic in traditional training and builds a full-dimensional cognitive system of "structure - state - logic".

[0020] 2. The multidimensional associated knowledge map of this invention dynamically binds multimodal training resources with equipment entity status and business logic nodes. Relying on the automatic iteration mechanism driven by both equipment status changes and procedure updates, it realizes real-time updates of resources as equipment parameters and procedures are adjusted, ensuring the timeliness of content.

[0021] 3. This invention combines a competency-oriented closed-loop mechanism with quantitative assessments of job performance based on job standards to accurately pinpoint skill gaps in different positions, thereby enabling precise resource allocation and achieving "on-demand supply." Based on real-time equipment status and existing resources, it dynamically generates composite scenarios incorporating physical, logical, and human-caused disturbances, covering faults from simple to complex, promoting the transformation of scenarios from pre-set to adaptive, and solving the problem of scenario isolation and disconnect from real-world application. It integrates graph databases, natural language processing, and other technologies to construct a full-chain architecture, achieving fault deduction and experience transfer through knowledge map coupling, promoting the transformation of training from general to specialized, and providing efficient talent training support for smart grid development. Attached Figure Description

[0022] Figure 1 This is a system structure diagram of the dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to the present invention; Figure 2 This is a schematic diagram of the "three-dimensional mapping + closed-loop optimization" integrated operation and monitoring knowledge map system of the present invention; Figure 3 This is a schematic diagram of the overall framework of the operation-monitoring knowledge map construction system of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0024] This invention aims to construct an integrated operation and monitoring business knowledge map for intelligent training. By integrating business knowledge from power system operation monitoring, fault analysis, and equipment maintenance, it clarifies the multiple attributes (operating parameters, operating status, health, etc.) and relationships (physical connections, secondary logical linkages, etc.) of equipment entities, and links various training resources such as text manuals and practical videos to build a multi-dimensional knowledge map of "equipment entity - business logic - training resources." Utilizing advanced information technology and artificial intelligence algorithms, it achieves dynamic updates and intelligent matching of knowledge and resources, enabling training content to be adjusted in real time as equipment is upgraded and procedures are updated. This provides centralized control station operation and maintenance personnel with personalized training support tailored to their job requirements and aligned with real-time equipment status and complex operating conditions. It promotes the deep integration of multi-modal resources and practical scenarios, helping grassroots personnel establish comprehensive understanding, improve multi-system collaborative handling capabilities and complex scenario response levels, and ensure the safe and stable operation of the power grid.

[0025] The technical solution of the present invention is as follows: This invention proposes a dynamic knowledge map construction system for operation and monitoring business aimed at intelligent training, such as... Figure 1 As shown, it includes a device entity layer, a business logic layer, a multimodal resource layer, a disturbance scenario generation layer, and a closed-loop optimization layer: At the equipment entity layer, multimodal data is collected and standardized to construct a quantitative model of equipment health. Combined with the primary and secondary equipment attribute system and their relationships, an equipment entity knowledge map is formed. In the device entity layer, the device health metric model is as follows:

[0026] in, , , and These represent the temperature reliability index. Operation frequency fatigue index Defect Severity Index and environmental impact index The weighting coefficients are set according to the actual situation, and .

[0027] The attribute system refers to a system with specific physical equipment / devices as nodes, where each node contains three types of core information: a unique identifier, static attributes, and dynamic attributes. Static attributes include the inherent characteristics of primary equipment and the functional and technical characteristics of secondary equipment. Dynamic attributes include the real-time operating status of primary equipment and the operating status and functional output of secondary equipment. The relationships include those between primary devices, those between primary devices and secondary devices, and those between secondary devices.

[0028] The device entity layer is updated based on three types of signals: real-time operational data changes, power grid topology adjustments, and parameter configuration modifications. This updates the device's dynamic attributes, relationships, and health status; the updates are achieved using Neo4j's Cypher statements.

[0029] In the business logic layer, multi-source rules are parsed using natural language processing technology to construct a three-level subgraph of the device entity knowledge map. A dynamic association engine is used to realize the mapping with the device entity layer, and the logical weights are dynamically adjusted based on the device health to make the rules adapt to changes in device status. In the business logic layer, the parsing process includes: The process involves word segmentation of technical terms related to equipment entity protection actions; identification of equipment entities, action instructions, and constraints within these technical terms; extraction of logical relationships between equipment entities, including the sequence of operation steps, triggering conditions, and prohibition conditions; and conversion of the extracted logical relationships into rule expressions. A three-level subgraph of the device entity knowledge map is constructed based on rule expressions, including the operation process subgraph, the protection logic subgraph, and the linkage strategy subgraph.

[0030] The operation flow sub-graph takes the decomposition of typical operation steps as the core, breaks down the operation into independent logical nodes, connects the nodes through sequential execution relationship, and clarifies the preconditions and operation basis between nodes. The protection logic subgraph, based on the action mechanism of the protection device, extracts protection settings, action thresholds, and exit commands as nodes, connects the criterion nodes through a "logical AND / OR" relationship, and defines the associated attributes of delay time and return coefficient. The linkage strategy subgraph, designed for multi-device collaboration scenarios, uses triggering conditions, execution actions, and upgrade strategies as nodes, connecting conditions and actions through a "causal triggering" relationship.

[0031] The multimodal resource layer integrates heterogeneous resources including text, images, videos, and virtual 3D models. After normalization, it constructs a cross-modal resource network to realize the association between resources and device entities and business logic, as well as resource invocation and interaction. The disturbance scenario generation layer divides the scenarios into three levels: healthy state, warning state, and fault state, based on the equipment health status. By injecting three types of disturbance factors—physical, logical, and human factors—it generates composite training scenarios that match the difficulty of the scenarios with the equipment status and the trainees' abilities. The closed-loop optimization layer collects operational behavior data, system interaction data, and capability assessment data to construct a three-dimensional evaluation model. Based on optimization priorities, it identifies optimization targets and drives the autonomous iteration of the device entity layer, business logic layer, and disturbance scenario generation layer.

[0032] In the closed-loop optimization layer, the three-dimensional evaluation model uses scenario authenticity deviation as the basic indicator, logic validity deviation as the core indicator, and training fit deviation as the fit indicator. It calculates a weighted sum of scenario authenticity deviation, logic validity deviation, and training fit deviation to obtain the optimization priority.

[0033] The scenario realism deviation refers to the degree of consistency between the virtual device response and the on-site data. The logical validity deviation is characterized by the ratio of the number of misjudgments by the rule engine to the total number of triggers; The training fit deviation is calculated using the standard deviation of trainees' ability scores.

[0034] When the optimization priority is higher than the set threshold, the numerical proportions of scenario authenticity deviation, logic validity deviation, and training suitability deviation are used to identify the dominant influencing factors. Based on the dominant factors, the specific levels in the device entity layer, business logic layer, or disturbance scenario layer are identified as the primary optimization targets. The historical training data associated with that level is used to train the optimization model. The core parameters within the level are correlated with the student's operational characteristics. By correcting the parameters, the deviation values ​​of the corresponding dimensions are reduced until the P-value drops below the threshold, thus completing this round of targeted iteration.

[0035] This invention also proposes a method for dynamically constructing an operation-monitoring business knowledge map for intelligent training, including: Step 1: Collect multimodal data from the time, physical, and state dimensions, including time series data of each device's operation, environmental parameters during operation, secondary device parameters, full life cycle data, and constraint data that the device meets during operation. Step 2: Construct a device health measurement model based on multimodal data; the device health measurement model outputs health levels, including healthy state, warning state, and fault state; Step 3: Based on the power grid model and topology, extract equipment entity nodes, define attributes and relationships; based on the equipment entity nodes, defined attributes and relationships, transform the collected multimodal data into an equipment entity knowledge map in a graph database; Step four: Use parsing methods to process the rule text, construct rule expressions for device entities and their relationships, and perform structured modeling of the device entity knowledge map based on the rule expressions to obtain a three-level sub-graph, including an operation flow sub-graph, a protection logic sub-graph, and a linkage strategy sub-graph; trigger the logical rules of the structured modeled device entity knowledge map based on the device status and device health level. Step 5: Integrate text resources, image resources, video resources, and virtual 3D model resources as multimodal resources and perform normalization processing; construct a multimodal network based on the multimodal resources; intelligently call upon the multimodal resources in the multimodal network according to the training scenario requirements and user operation behavior; Step 7: Construct an association evaluation model based on multimodal interaction data, and realize the evolution of the device entity knowledge map based on the output value of the association evaluation model.

[0036] Example 1 This invention addresses the problems of fragmented knowledge, disconnect between scenarios and actual combat, and insufficient training resources in the integrated operation and monitoring of centralized control stations. It constructs a knowledge map system based on "three-dimensional mapping + closed-loop optimization," such as... Figure 2 As shown, through the deep integration of equipment entities, business logic, and training resources, intelligent management of the entire chain from data collection to capability enhancement is achieved.

[0037] The system's core architecture adopts a five-layer progressive design, with each layer organically connected to the algorithm model through data interfaces, forming a dynamically collaborative technical system. (1) Equipment Entity Layer: As the data foundation of the knowledge map, multimodal data is collected through the "time-object-field" three-dimensional perception network. After standardization processing, a quantitative model of equipment health is constructed. Combined with the attribute system and association rules of "primary-secondary equipment", an equipment entity knowledge map containing equipment static attributes, dynamic status and logical relationships is formed, providing accurate digital mapping of equipment for upper-layer applications. (2) Business logic layer: As the hub connecting the equipment entity and the training application, it uses natural language processing technology to structure multi-source rules such as procedures and contingency plans, and constructs a three-level sub-graph of "operation process - protection logic - linkage strategy". It uses a dynamic association engine to realize the mapping with the equipment entity layer, and dynamically adjusts the logic weight based on the equipment health status to ensure that the rules adapt to changes in equipment status and provide calculable logical support for training scenarios.

[0038] (3) Multimodal resource layer: Integrates heterogeneous resources such as text, images, videos, and virtual 3D models. After normalization, it constructs a cross-modal resource network through a triple mapping of "resource metadata - device entity node - logical node" to achieve accurate association between resources and device entities and business logic, and supports scenario-based resource invocation and interaction.

[0039] (4) Disturbance scenario generation layer: Based on the equipment health status, the three-level scenarios are divided into healthy state, early warning state and fault state. By injecting three types of disturbance factors, namely physical, logical and human factors, a composite training scenario close to actual combat is generated to achieve dynamic adaptation of scenario difficulty with equipment status and trainee ability. (5) Closed-loop optimization layer: By collecting operational behavior data, system interaction data, and capability assessment data, a three-dimensional assessment model is constructed. Based on the optimization priority, the optimization object is located, driving the device entity layer, business logic layer, and scenario layer to iterate autonomously, forming a closed loop of "assessment-optimization-verification" to ensure that the system continuously adapts to training needs.

[0040] This architecture breaks through the limitations of "data silos" and "static resources" in traditional training systems. Through the closed-loop flow of "data-knowledge-scenario-capability", it significantly improves the ability of central control station operation and maintenance personnel to handle complex faults.

[0041] Specifically, the technology for constructing the device entity layer is as follows.

[0042] The device entity layer is the data foundation of the knowledge map. Through multimodal data fusion and health measurement, it realizes the digital expression of the physical attributes and operating rules of devices, and achieves accurate mapping of physical devices to digital models, providing core basis for business logic association and scenario generation.

[0043] 1. Multimodal data acquisition and processing To comprehensively capture equipment status, the system deploys a three-dimensional sensing network encompassing time, object, and field: Time dimension: For example, continuously monitor contact wear and bearing vibration amplitude through vibration sensors; combine parameters such as voltage, current, and operation frequency collected by SCADA system to build a time series database of equipment operating status; deploy microphones to record abnormal noises of equipment and identify potential faults such as discharge and jamming through voiceprint features.

[0044] Physical dimension: For example, infrared thermal imagers are used to collect the surface temperature distribution of equipment and extract features such as temperature gradient and hot spot temperature rise; insulation status parameters are obtained through SF6 gas sensors and oil chromatography analyzers; environmental temperature and humidity sensor data are correlated to collect environmental parameters such as temperature, humidity, air pressure, and pollution level to form a state perception system coupled with multiple physical fields; the collected multi-source heterogeneous data need to be standardized.

[0045] Status dimension: For example, parsing the protection device configuration file to obtain secondary equipment parameters such as differential protection braking coefficient and reclosing time in real time; extracting equipment model, commissioning time, maintenance records, historical defects, etc. through the equipment ledger system; obtaining the power grid topology model through the dispatch management system to obtain equipment network connection relationships, real-time power flow data, dispatch command execution records, and extracting information such as the operating constraints of equipment in the power grid.

[0046] 2. Equipment Health (SOH) Quantitative Model A quantitative model for equipment health (SOH) is constructed based on multimodal data. Equipment health is a core indicator relating equipment status to training scenarios. A multi-parameter fusion model (e.g., based on the analytic hierarchy process) is built using this model. Figure 3 As shown in the figure, this comprehensively reflects the degree of equipment aging and the risk of failure:

[0047] in, , , and These represent the temperature reliability index. Operation frequency fatigue index Defect Severity Index and environmental impact index The weighting coefficients are set according to the actual situation, and As a preferred embodiment of the present invention, , , and Set them to 0.4, 0.3, 0.2, and 0.1 respectively.

[0048] Specifically, the temperature reliability index ( The index, calculated by the deviation between infrared thermometry values ​​and design thresholds, shows a linear decrease in reliability when the internal temperature exceeds the upper limit of the equipment's normal operating temperature; typically, reliability decreases by 0.5% for every 1°C increase. When the temperature exceeds the equipment's warning temperature threshold, the index decreases more rapidly, typically decreasing by 4% for every 1°C increase. When the temperature exceeds the danger temperature threshold... It exhibits exponential decay, reflecting the nonlinear growth of the insulation aging rate. Range [0,1].

[0049] Operation frequency fatigue index ( The mechanical operating components (such as contacts and operating mechanisms) of equipment like circuit breakers and disconnectors suffer from cumulative fatigue. Each opening and closing operation leads to minor wear (such as contact bounce and spring elasticity decay). Their performance degradation follows an exponential decay law of "operation count - reliability"—that is, as the number of operations increases, the probability of mechanical failure increases exponentially, and reliability decreases exponentially. Based on the ratio of the number of opening and closing operations to the rated life of the operating mechanism, an exponential decay model is adopted. (N is the cumulative number of operations).

[0050] Defect Severity Index (DSI) ): Assign values ​​based on the historical defect levels obtained from the ledger (0.8 for general defects, 0.5 for serious defects, and 0.3 for urgent defects), and 1 for no defects.

[0051] Environmental Impact Index (EPI) ): Used to quantify the comprehensive impact of external environmental factors on the health status of equipment. Its core is to reflect the indirect effects of environmental conditions on the insulation performance and heat dissipation efficiency of equipment through the weighted fusion of multi-dimensional environmental parameters. It is usually calculated using a weighted average (usually temperature weight is 0.4, humidity weight is 0.3, and pollution level weight is 0.3).

[0052] The model updates every 5 minutes through edge computing nodes, outputting a health level (healthy state SOH>0.8, warning state 0.6≤SOH≤0.8, fault state SOH<0.6), providing trigger conditions for the generation of subsequent training scenarios.

[0053] 3. Equipment Knowledge Map Construction Technology This system uses a three-layer mapping mechanism of "data-entity-relationship" to transform the collected multi-source data into a knowledge map of equipment entities in a graph database based on the power grid model and topology. By extracting equipment entity nodes, defining attributes and relationships, the system realizes the transformation of power grid equipment from physical state to structured and semantic expression, and supports the correlation analysis and logical reasoning of equipment state.

[0054] (1) Specific composition of equipment physical nodes Equipment entity nodes are based on a "primary equipment - secondary equipment" classification system, with specific physical equipment / devices as nodes. Each node contains three core types of information: a unique identifier, static attributes, and dynamic attributes, as defined below: ① Primary equipment attributes Static attributes: used to describe the inherent characteristics of the equipment, including but not limited to: model specifications, rated voltage, rated current, rated capacity, insulation class, etc.

[0055] Dynamic attributes: used to reflect the real-time operating status of the equipment, including but not limited to: real-time voltage, current, power, temperature, pressure, opening and closing status, operation / maintenance / standby status, alarm information, and state of health (SOH).

[0056] ②Secondary equipment attributes Static attributes: used to define the functions and technical characteristics of the equipment, including but not limited to: equipment function type (differential protection / overcurrent protection / data acquisition, etc.), model, rated power supply, communication protocol, sampling accuracy, action delay range, protection range, etc.

[0057] Dynamic attributes: used to characterize the operating status and functional output of the device, including but not limited to: power on / off status, communication status (normal / interrupted), protection settings, number of actions, and interaction status with other secondary devices.

[0058] (2) Rules for constructing relationships between equipment entities The relationships are established based on the physical connections of the power grid, functional logic, and operational coordination. Structured storage is achieved through a graph database triple model, with specific types and construction methods as follows: ① Interrelationships between primary equipment Power grid topology data is extracted from the OMS system, and main wiring diagrams and equipment bay information are obtained from the PMS system. Connection relationships are automatically generated using topology analysis algorithms. Electrical connection: Indicates that equipment is directly connected through a conductor (e.g., "xx busbar - connection - xxx circuit breaker", "xx circuit breaker - connection - xxx disconnector", etc.). Hierarchical inclusion: Represents the inclusion relationship between equipment and its components (such as "xx transformer - includes - xx high voltage winding", "xx circuit breaker - includes - xx spring operating mechanism", etc.).

[0059] ② The relationship between primary and secondary equipment The protection configuration sheet and secondary wiring diagram are parsed from the PMS system, and the "device-equipment" association table is extracted from the substation automation system. Functional associations are generated through rule engine matching, and constructed based on protection configuration, monitoring range, and control logic, including: Protection relationship: Represents the protection function of secondary equipment on primary equipment (e.g., "xx differential protection device - protection - xx transformer"). The associated attributes include protection type (differential / overcurrent / distance), action output, protection setting range, etc. Monitoring Relationship: Characterizes the status acquisition of primary equipment by secondary equipment, and the associated attributes include acquisition parameters (voltage / current / temperature), sampling frequency, and data transmission path; Control relationship: Represents the operation control of the secondary equipment on the primary equipment, and the associated attribute is the control method (remote control / local control).

[0060] ③ Interrelationships between secondary equipment By analyzing the logical linkage rules defined in the protection regulations, a logical linkage relationship based on functional coordination is constructed. This aims to characterize functional coordination (e.g., "busbar protection device - blocking - line protection device"), and the associated attributes include linkage conditions (e.g., "blocking line reclosing during busbar fault"), delay parameters, etc.

[0061] (3) Dynamic update mechanism of equipment entity layer The dynamic update mechanism at the device entity layer automatically iterates entity attributes, connectivity, and health by sensing real-time changes in the power grid. It operates within a closed loop of "trigger condition - update content - execution logic - collaborative feedback," ensuring real-time mapping with the physical power grid. This mechanism is triggered by three types of signals: changes in real-time operational data, power grid topology adjustments, and parameter configuration modifications. Based on these signals, the system dynamically updates three key aspects: device dynamic attributes (such as voltage, phase angle, and power flow), connectivity (e.g., adjusting connection status or adding / deleting relationships during topology changes), and health indicators (recalculating SOH and sub-indices at regular intervals, with linked updates when defect status changes). The execution process employs an "event-driven + transaction control" model, using Neo4j's Cypher statements to update attributes and relationships. Furthermore, this mechanism collaborates with upper layers, pushing state changes to the business logic layer to trigger rule verification and sending health migration signals to the scenario generation layer to adjust scenario difficulty, ultimately forming a cross-layer closed loop. This ensures both the "digital twin" characteristic with physical devices and provides reliable data support for upper-layer applications.

[0062] Through the aforementioned attribute system and association rules, the equipment knowledge map can achieve full-dimensional digital mapping of power grid equipment's "static characteristics - dynamic status - logical associations," providing a structured knowledge foundation for subsequent rule parsing in the business logic layer and dynamic generation of training scenarios.

[0063] Furthermore, the business logic layer is constructed as follows.

[0064] As the intermediate hub between the equipment entity layer and the training application layer, the business logic layer should be constructed with "structured procedures, computable logic, and dynamic associations" as its core, which not only reflects the technical connection with the equipment entity layer, but also provides clear logical support for upper-layer applications.

[0065] First, a foundation is laid through "multi-source rule-based structured parsing." Texts such as industry standards, operating procedures, emergency plans, historical cases, and typical operation tickets are processed using natural language processing technology, employing a four-step parsing method: word segmentation, entity recognition, relation extraction, and rule generation. First, a dedicated lexicon is built for professional terms such as "differential protection" and "switching operation," and word segmentation is performed using a bidirectional LSTM model, achieving an accuracy of 98.5%. Then, a BiLSTM+CRF model is used to identify "operation objects" (such as "circuit breaker" and "protection plate"), "action commands" (such as "opening" and "closing"), and "constraints" (such as "action time limit ≤ 3 minutes" and "voltage level ≥ 220kV") in the text, achieving an entity recognition F1 score of 93%. Finally, logical relationships between entities are extracted based on an attention mechanism, including "operation..." The system extracts data into several categories, including "operation sequence" (e.g., voltage testing → grounding → tagging), "condition trigger" (e.g., oil temperature > 85℃ → start cooler), and "prohibited operation" (e.g., switching operation is prohibited during thunderstorms). The extracted entities and relationships are then transformed into "IF-THEN" rule expressions. For example, from unstructured text, we can extract "IF operation object = 'grounding switch' AND status = 'closed' THEN prohibited operation = 'close disconnect switch'", and from protection procedures, we can extract "IF differential current > 0.6 × rated current THEN protection action = 'trip circuit breaker'".

[0066] Secondly, a "hierarchical logical knowledge map" is constructed to achieve structured modeling. Based on the parsed rules, a three-level subgraph is built using Neo4j.

[0067] (1) Operation flow sub-graph: Taking the decomposition of typical operation steps as the core, the operation is broken down into independent logical nodes. The nodes are connected by the "sequential execution" relationship, clarifying the preconditions and operation basis between nodes, and realizing the structured mapping of the operation flow.

[0068] (2) Protection logic subgraph: Based on the action mechanism of the protection device, the protection setting, action threshold and exit command are extracted as nodes. The judgment nodes are connected through the "logical AND / OR" relationship. The associated attributes such as delay time and return coefficient are defined to model the logical chain of protection action.

[0069] (3) Linkage Strategy Subgraph: For multi-device collaboration scenarios, the triggering conditions, execution actions, and upgrade strategies are used as nodes. The conditions and actions are connected through the "cause-and-effect triggering" relationship to depict the collaborative response rules across devices.

[0070] Finally, a "dynamic association engine" is used to achieve real-time binding with the equipment entity layer. When the status of the equipment entity layer changes (such as "the oil temperature of a transformer reaches 90℃"), the trigger condition "oil temperature > 85℃" in the business logic map is automatically matched to activate the corresponding linkage strategy; at the same time, the association weight is dynamically adjusted based on the equipment health status (SOH) (such as when SOH < 0.6, the weight of the maintenance process rule is increased by 40%), ensuring that the logic rules are adapted to the equipment status in real time.

[0071] Furthermore, the multimodal resource layer integration technology is as follows.

[0072] The multimodal resource layer serves as a carrier for training scenarios and knowledge transfer. By integrating heterogeneous resources such as text, images, videos, and virtual 3D models, it achieves deep integration with the device entity layer and business logic layer, providing intuitive and scenario-based support for training applications. Its core technologies focus on three aspects: "resource normalization processing - cross-modal association modeling - dynamic adaptation and invocation".

[0073] 1. Types and normalization processing of multimodal resources The integrated resources cover four categories, and standardized processing is used to achieve a unified format and metadata annotation: Text resources include procedures and specifications, operation tickets, fault analysis reports, etc. After OCR recognition and NLP word segmentation, keywords are extracted, associated with device entity layer nodes and business rule nodes, and stored as a structured text library.

[0074] Image resources: including equipment appearance diagrams, wiring diagrams, infrared thermal images, etc., and annotate equipment components and defect features through image recognition, and associate related equipment entity nodes.

[0075] Video resources: covering typical operation demonstrations and fault handling processes, extracting key step frames through frame parsing, and associating nodes of the operation process sub-graph.

[0076] Virtual 3D model resources: Equipment 3D models (such as the internal structure of primary equipment and the layout of protection cabinets) and scene models (such as the primary equipment area of ​​a substation) are lightweighted and equipment entity nodes are bound to the model components to support component-level interaction.

[0077] 2. Cross-modal resource association mechanism Based on the triple mapping of "resource metadata - device entity nodes - business logic nodes", a multimodal resource network is constructed: Associating with the device entity layer: Establishing the association between resources and entities through the unique ID of the device, such as establishing a "mapping" relationship between specific devices in the virtual 3D model and specific entities in the device entity layer, and associating the "overheated area" of the infrared image with the "dynamic attribute - temperature" of the entity; Associating with the business logic layer: Matching resources with business logic nodes, such as associating the "power verification" step frame of the operation video with the "power verification" node of the operation process sub-diagram, and associating the protection setting sheet text with the "protection setting" node of the protection logic sub-diagram, etc. Inter-resource association: Cross-modal associations are established through semantic similarity algorithms, such as associating the video "main transformer cooler startup" with the text resource "high oil temperature linkage strategy", and the association strength is labeled according to the similarity.

[0078] 3. Dynamic adaptation and invocation technology Based on training scenario requirements and user behavior, resources can be intelligently pushed and displayed in combination: Scenario-triggered call: When the training scenario is "High oil temperature handling of main transformer", the system automatically retrieves related resources: virtual 3D model showing the cooler location, operation video demonstrating the startup steps, and the procedure text popping up Article 6.3 of "Transformer Operation Procedure", etc.; Interactive feedback adaptation: When students operate the "cooler" component of the virtual 3D model, the principle animation of the component is played automatically; when querying the "differential protection action" logic, the protection setting sheet text and action waveform diagram are displayed simultaneously. Through the aforementioned technologies, the multimodal resource layer achieves structured integration and intelligent access to heterogeneous resources. It retains the original characteristics of the resources while deeply integrating them into the knowledge map system through association mechanisms, providing rich resource support for the intuitive presentation of training scenarios and the multidimensional cognition of trainees.

[0079] Furthermore, the specific techniques for generating disturbed scenes are as follows.

[0080] The disturbance scenario generation layer dynamically combines equipment health with multi-dimensional disturbance factors to construct training scenarios that closely resemble real-world scenarios, solving the problem of the disconnect between traditional fixed scenarios and actual working conditions. Its core logic is as follows: using equipment health (SOH) as a baseline to divide scenario levels, and injecting three types of disturbance factors—physical, logical, and human factors—to simulate the evolution process of equipment from normal to faulty, thereby achieving dynamic adaptation and progressive complexity of training scenarios.

[0081] 1. Logical description of the health-driven scenario baseline Scene generation is triggered primarily by device health, forming a three-tiered, progressive scene system to ensure that scene difficulty matches device status. • Healthy state (SOH>0.8): Based on the normal operating status of the equipment, focus on standardized operation training, and design the scenario with the goal of memorizing the standardized process, linking basic procedures and operation video resources; • Warning state (0.6≤SOH≤0.8): Simulates the evolution of minor equipment defects. The scenario includes initial abnormal characteristics and basic handling procedures, focusing on cultivating defect identification and recording capabilities. • Fault state (SOH<0.6): Construct a complex fault scenario with multiple superimposed factors, integrating physical degradation of equipment, logical anomalies and human error, and emphasizing training in multi-system collaborative handling capabilities.

[0082] 2. Logical description of multi-dimensional perturbation injection technology By synergistically injecting three types of perturbation factors, the dynamic complexity of the scenario is adjusted. The specific logic is as follows: Physical disturbance: Based on the equipment aging model, simulate the degradation of physical parameters (such as mechanical wear and electrical performance decline), and quantify the degree of degradation through numerical calculation; Logical disturbance: Modify logical parameters such as protection settings and topology relationships to simulate logical-level anomalies such as protection malfunctions and communication failures; Human-caused disturbances: By embedding typical misoperation sequences and triggering alarms through behavior monitoring, training on standardized operating procedures is strengthened.

[0083] Furthermore, the specific implementation method of the closed-loop optimization mechanism is as follows.

[0084] The closed-loop optimization mechanism collects training process data in real time, constructing a closed-loop link of "operational feedback - defect analysis - system iteration - effect verification" to achieve dynamic evolution of each level of the knowledge map and training scenarios, ensuring the system continuously adapts to the practical needs of the central control station. Based on multimodal interaction data such as trainee operation data and scenario interaction records, a correlation evaluation model of equipment-business-scenario is built to locate system weaknesses. Through the multi-dimensional evaluation model, optimization instructions are output, driving the parameters of the equipment entity layer, business logic layer, and disturbed scenarios to achieve autonomous iteration, ensuring the system continuously meets the practical training needs of the central control station.

[0085] Data collection covered the entire training process, resulting in three core datasets: (1) Operation behavior data: By using the operation terminal and operation sequence recognition algorithm, the spatiotemporal characteristics of the student's operation (such as operation step sequence, operation duration, operation step omission rate, operation sequence error, etc.) are collected to generate operation compliance vector; (2) System interaction data: Collect equipment status response in training scenarios (such as the deviation between protection action delay and theoretical value stored in business logic layer), perception resource association accuracy, i.e. matching degree between called resource and current training scenario (such as calling "line protection" video in "main transformer cooler failure" scenario, triggering resource mapping optimization when association error rate > 8%), rule triggering success rate (such as matching accuracy of "five-prevention interlocking" rule in business logic layer). (3) Capability assessment data: The training effect is quantified through a "5-level capability matrix" (basic operation → defect identification → fault handling → collaborative decision-making → emergency innovation), and the trainee training score is output. Specifically, each capability is scored and given a certain weight, and the weighted sum is calculated as the trainee training score.

[0086] The evaluation model quantifies the system's operational status from three core dimensions, and uses weighted integration to form an optimization priority ranking. The specific dimensions are selected based on the following criteria: scenario realism deviation as a basic indicator to ensure consistency between virtual scenarios and actual working conditions; logic validity deviation as a core indicator to ensure the accuracy of business rule triggering; and training fit deviation as a fit indicator to match trainees' ability growth curves.

[0087] The optimization priority formula is:

[0088] in, The higher the value, the higher the priority for system optimization. For scene realism deviation (weight) =0.4), reflecting the degree of agreement between the virtual device response and the field data (such as the similarity of current waveforms under physical disturbances). For logical validity bias (weight) =0.3), the ratio of the number of misjudgments by the rule engine to the total number of triggers represents the accuracy of the rule engine's judgment; Training fit bias (weight) =0.3), which is calculated by the standard deviation of the students' ability scores to reflect the matching degree between the difficulty of the scenario and the students' abilities; Weight , and Set according to the actual situation, and .

[0089] when When the value exceeds the set threshold, the system disassembles... , , The numerical proportions are used to identify the dominant influencing factors. These dominant factors are then assigned to specific levels within the device entity layer, business logic layer, or disturbance scenario layer as the primary optimization targets. Historical training data associated with that level is used to train the optimization model. Core parameters within that level are dynamically correlated with trainee operational characteristics. By adjusting parameters (such as supplementing deviation coefficients and adjusting correlation weights), the deviation values ​​of the corresponding dimensions are reduced until the P-value falls below the threshold, completing this round of targeted iteration. Through this chain of "overall priority determination - influencing factor decomposition - optimization level positioning - dynamic parameter correction," a precise transformation from evaluation values ​​to specific optimization actions is achieved, ensuring that each level of the knowledge map dynamically evolves with trainee training data, continuously aligning with practical needs.

[0090] Example 2 This invention also proposes a method for dynamically constructing an operation-monitoring business knowledge map for intelligent training, the specific steps of which are as follows.

[0091] Step 1: Collect multimodal data from the time, physical, and state dimensions, including time series data of each device's operation, environmental parameters during operation, secondary device parameters, full life cycle data, and constraint data that the device meets during operation. Step 2: Construct a device health measurement model based on multimodal data; the device health measurement model outputs health levels, including healthy state, warning state, and fault state; Step 3: Based on the power grid model and topology, extract equipment entity nodes, define attributes and relationships; based on the equipment entity nodes, defined attributes and relationships, transform the collected multimodal data into an equipment entity knowledge map in a graph database; The equipment entity nodes include primary equipment and secondary equipment; each node contains three types of core information: unique identifier, static attributes, and dynamic attributes; the association relationship is established based on the physical connection of the power grid, functional logic, and operational coordination, and is structured and stored through the triplet model of the graph database; The device entity nodes, defined attributes, and relationships are updated based on three types of signals: real-time operational data changes, power grid topology adjustments, and parameter configuration modifications.

[0092] Step four: Use parsing methods to process the rule text, construct rule expressions for device entities and their relationships, and perform structured modeling of the device entity knowledge map based on the rule expressions to obtain a three-level sub-graph, including an operation flow sub-graph, a protection logic sub-graph, and a linkage strategy sub-graph; trigger the logical rules of the structured modeled device entity knowledge map based on the device status and device health level. Step 5: Integrate text resources, image resources, video resources, and virtual 3D model resources as multimodal resources and perform normalization processing; construct a multimodal network based on the multimodal resources; intelligently call upon the multimodal resources in the multimodal network according to the training scenario requirements and user operation behavior; The training scenario is generated using a disturbance scenario generation technology; the disturbance scenario generation technology constructs the training scenario by dynamically combining device health with multi-dimensional disturbance factors. Specifically, the scenario levels are divided based on the equipment health level, and the evolution process of the equipment from normal to fault is simulated by injecting three types of disturbance factors: physical, logical and human factors. Step 7: Construct an association evaluation model based on multimodal interaction data, and realize the evolution of the device entity knowledge map based on the output value of the association evaluation model; the above includes student operation data and scene interaction records; Specifically, the correlation evaluation model is characterized by an optimization priority formula, which is:

[0093] in, The higher the value, the higher the priority for system optimization. To reflect the deviation of scene realism, it reflects the degree of consistency between the virtual device response and the on-site data (such as the similarity of current waveform under physical disturbance). To assess logical validity bias, the accuracy of the rule engine's judgment is characterized by the ratio of the number of misjudgments to the total number of triggers. To address the training fit bias, the standard deviation of trainees' ability scores is calculated to reflect the match between the difficulty of the scenario and the trainees' abilities. Weight , and Set according to the actual situation, and .

[0094] when When the value exceeds the set threshold, the system disassembles... , , The numerical proportion is used to identify the dominant influencing factor. Based on the dominant factor, the specific level in the equipment entity layer, business logic layer, or disturbance scenario layer is taken as the primary optimization target. The historical training data associated with the level is called to train the optimization model. The core parameters within the level are dynamically associated with the student's operation characteristics. By correcting the parameters (such as supplementing the deviation coefficient, adjusting the association weight, etc.), the deviation value of the corresponding dimension is reduced until the P value drops below the threshold, thus completing this round of targeted iteration.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A dynamic knowledge map construction system for operation and monitoring business oriented towards intelligent training, characterized in that, It includes the device entity layer, business logic layer, multimodal resource layer, disturbance scenario generation layer, and closed-loop optimization layer: At the equipment entity layer, multimodal data is collected and standardized to construct a quantitative model of equipment health. Combined with the primary and secondary equipment attribute system and their relationships, an equipment entity knowledge map is formed. In the device entity layer, the device health metric model is as follows: in, , , and These represent the temperature reliability index. Operation frequency fatigue index Defect Severity Index and environmental impact index The weighting coefficients are set according to the actual situation, and ; In the business logic layer, multi-source rules are parsed using natural language processing technology to construct a three-level subgraph of the device entity knowledge map. A dynamic association engine is used to realize the mapping with the device entity layer, and the logical weights are dynamically adjusted based on the device health to make the rules adapt to changes in device status. The multimodal resource layer integrates heterogeneous resources including text, images, videos, and virtual 3D models, and after normalization, constructs a cross-modal resource network. The disturbance scenario generation layer divides the scenarios into three levels: healthy state, warning state, and fault state, based on the equipment health status. By injecting three types of disturbance factors—physical, logical, and human factors—it generates composite training scenarios that match the difficulty of the scenarios with the equipment status and the trainees' abilities. Among them, scene generation takes device health as the core trigger condition, forming a three-level progressive scene system to ensure that the difficulty of the scene matches the device status, including healthy state, warning state, and fault state; The synergistic injection of three types of perturbation factors enables dynamic adjustment of the scenario's complexity. The specific logic is as follows: Physical disturbance: Based on the equipment aging model, the physical parameter degradation is simulated, and the degree of degradation is quantified through numerical calculation; Logical disturbance: Modify logical parameters including protection settings and topology relationships to simulate logical-level anomalies including protection malfunctions and communication abnormalities; Human-caused disturbances: Implanting a sequence of erroneous operations, triggering alarms through behavioral monitoring, and reinforcing training in standardized operations; The closed-loop optimization layer collects operational behavior data, system interaction data, and capability assessment data to construct a three-dimensional assessment model. Based on the optimization priority, it locates the optimization object and drives the autonomous iteration of the device entity layer, business logic layer, and disturbance scenario generation layer. The three-dimensional evaluation model quantifies the system's operational status from three core dimensions, and forms an optimization priority ranking through weighted integration. The specific dimensions are selected based on: scenario realism deviation as a basic indicator; logical validity deviation as a core indicator; and training fit deviation as an fit indicator. The optimization priority formula is: in, The higher the value, the higher the priority for system optimization. This is due to a deviation from the realism of the scene. To assess logical validity bias, the accuracy of the rule engine's judgment is characterized by the ratio of the number of misjudgments to the total number of triggers. Training fit bias is calculated using the standard deviation of trainees' ability scores; weights , and Configure according to the actual situation.

2. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: The attribute system refers to a system with specific physical equipment / devices as nodes, where each node contains three types of core information: a unique identifier, static attributes, and dynamic attributes. Static attributes include the inherent characteristics of primary equipment and the functional and technical characteristics of secondary equipment. Dynamic attributes include the real-time operating status of primary equipment and the operating status and functional output of secondary equipment. The relationships include those between primary devices, those between primary devices and secondary devices, and those between secondary devices.

3. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: The device entity layer is updated based on three types of signals: real-time operational data changes, power grid topology adjustments, and parameter configuration modifications. This updates the device's dynamic attributes, relationships, and health status; the updates are achieved using Neo4j's Cypher statements.

4. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: In the business logic layer, the parsing process includes: The process involves word segmentation of technical terms related to equipment entity protection actions; identification of equipment entities, action instructions, and constraints within these technical terms; extraction of logical relationships between equipment entities, including the sequence of operation steps, triggering conditions, and prohibition conditions; and conversion of the extracted logical relationships into rule expressions. A three-level subgraph of the device entity knowledge map is constructed based on rule expressions, including the operation process subgraph, the protection logic subgraph, and the linkage strategy subgraph.

5. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 4, characterized in that: The operation flow sub-graph takes the decomposition of typical operation steps as the core, breaks down the operation into independent logical nodes, connects the nodes through sequential execution relationship, and clarifies the preconditions and operation basis between nodes. The protection logic subgraph, based on the action mechanism of the protection device, extracts protection settings, action thresholds, and exit commands as nodes, connects the criterion nodes through "logical AND / OR" relationships, and defines the associated attributes of delay time and return coefficient. The linkage strategy subgraph, designed for multi-device collaboration scenarios, uses triggering conditions, execution actions, and upgrade strategies as nodes, connecting conditions and actions through a "causal triggering" relationship.

6. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: In the closed-loop optimization layer, the three-dimensional evaluation model uses scenario authenticity deviation as the basic indicator, logic validity deviation as the core indicator, and training fit deviation as the fit indicator. It calculates a weighted sum of scenario authenticity deviation, logic validity deviation, and training fit deviation to obtain the optimization priority.

7. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: The scenario realism deviation refers to the degree of consistency between the virtual device response and the on-site data. The logical validity deviation is characterized by the ratio of the number of misjudgments by the rule engine to the total number of triggers; The training fit deviation is calculated using the standard deviation of trainees' ability scores.

8. The dynamic construction system for operation-monitoring business knowledge maps for intelligent training according to claim 1, characterized in that: When the optimization priority is higher than the set threshold, the numerical proportions of scenario authenticity deviation, logic validity deviation, and training adaptability deviation are used to identify the dominant influencing factors. The specific level in the device entity layer, business logic layer, or disturbance scenario layer corresponding to the dominant factors is taken as the primary optimization target. The model is trained and optimized by calling the historical training data associated with this level, associating the core parameters within the level with the operational characteristics of the trainees, and reducing the deviation value of the corresponding dimension by correcting the parameters until the P value drops below the threshold, thus completing this round of targeted iteration.

9. A method for dynamically constructing an operation-monitoring business knowledge map for intelligent training, applicable to the system described in any one of claims 1-8, characterized in that, include: Step 1: Collect multimodal data from the time, physical, and state dimensions, including time series data of each device's operation, environmental parameters during operation, secondary device parameters, full life cycle data, and constraint data that the device meets during operation. Step 2: Construct a device health measurement model based on multimodal data; the device health measurement model outputs health levels, including healthy state, warning state, and fault state; The device health measurement model is as follows: in, , , and These represent the temperature reliability index. Operation frequency fatigue index Defect Severity Index and environmental impact index The weighting coefficients are set according to the actual situation, and ; Step 3: Based on the power grid model and topology, extract equipment entity nodes, define attributes and relationships; based on the equipment entity nodes, defined attributes and relationships, transform the collected multimodal data into an equipment entity knowledge map in a graph database; Step four: Use parsing methods to process the rule text, construct rule expressions for device entities and their relationships, and perform structured modeling of the device entity knowledge map based on the rule expressions to obtain a three-level sub-graph, including an operation flow sub-graph, a protection logic sub-graph, and a linkage strategy sub-graph; trigger the logical rules of the structured modeled device entity knowledge map based on the device status and device health level. Step 5: Integrate text resources, image resources, video resources, and virtual 3D model resources as multimodal resources and perform normalization processing; construct a multimodal network based on the multimodal resources; intelligently call upon the multimodal resources in the multimodal network according to the training scenario requirements and user operation behavior; The training scenario is generated using a disturbance scenario generation technology; the disturbance scenario generation technology constructs the training scenario by dynamically combining device health with multi-dimensional disturbance factors. Specifically, the scenario levels are divided based on the equipment health level, and the evolution process of the equipment from normal to fault is simulated by injecting three types of disturbance factors: physical, logical and human factors. Step 7: Construct an association evaluation model based on multimodal interaction data, and realize the evolution of the device entity knowledge map based on the output value of the association evaluation model; The correlation evaluation model is characterized by an optimization priority formula, which is: in, The higher the value, the higher the priority for system optimization. This is due to a deviation from the realism of the scene. To assess logical validity bias, the accuracy of the rule engine's judgment is characterized by the ratio of the number of misjudgments to the total number of triggers. Training fit bias is calculated using the standard deviation of trainees' ability scores; weights , and Configure according to the actual situation.

Images