Power grid equipment health management and operation and maintenance platform based on knowledge federation and language model

Abstract

The invention discloses a power grid equipment health management and operation and maintenance platform based on knowledge federation and a language model, and relates to the technical field of power grid equipment health management and intelligent operation and maintenance. Comprising a multi-source data fusion module, a knowledge federal management system, a large language model, an intelligent analysis and prediction module, an operation and maintenance decision support module and a visual display module, and the large language model can be a DeepSeek-R1 large model. According to the method, the knowledge federation technology and the breakthrough AI capacity of the DeepSeek-R1 large model are creatively fused, remarkable technical innovation and function breakthrough are achieved in the aspects of data fusion, knowledge utilization, man-machine interaction, intelligent decision making and the like, and the limitation of an existing power grid operation and maintenance technology can be effectively overcome; the intelligent and active preventive level of operation and maintenance of power grid equipment is comprehensively improved, and key technical support is provided for constructing a safer, more reliable and more efficient intelligent power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid equipment health management and intelligent operation and maintenance, and more specifically to a power grid equipment health management and operation and maintenance platform based on knowledge federation and language model. Background Art

[0002] With the in-depth advancement of smart grid construction and the accelerated construction of new power systems, the scale of the power grid continues to expand, and the number of devices is growing exponentially. The health management and intelligent operation and maintenance level of power grid equipment are directly related to the safe and stable operation, power supply reliability and economic benefits of the power grid. However, existing intelligent operation and maintenance technologies still face many difficult-to-overcome challenges: (1) Insufficient data fusion capabilities make it difficult to effectively utilize multi-source heterogeneous data Existing intelligent O&M platforms typically focus on analyzing and applying a single data source (e.g., analyzing only operational data from SCADA systems or only inspection image data). They lack the ability to effectively integrate and comprehensively analyze heterogeneous multi-source O&M data (e.g., operational data, inspection data, historical maintenance data, environmental data, and expert knowledge). This makes it difficult to comprehensively and accurately assess equipment health and predict failure risks. Data silos persist, limiting the depth and breadth of intelligent O&M.

[0003] (2) Insufficient knowledge representation and reasoning capabilities make it difficult to effectively utilize expert experience and industry knowledge Existing intelligent O&M platforms primarily rely on traditional machine learning algorithms and statistical analysis methods. These models have limited knowledge representation and reasoning capabilities, making it difficult to effectively learn, understand, and utilize the expert experience and industry knowledge in the power grid sector. This limits the platform's intelligent decision-making capabilities and O&M strategy generation capabilities, hindering the full value of expert experience and industry knowledge.

[0004] (3) The human-computer interaction method is not smart and convenient enough, and the user experience needs to be improved Existing intelligent O&M platforms rely primarily on traditional graphical interfaces and menu-based operations for human-computer interaction. These interactions are unnatural, inefficient, and inconvenient, requiring users to spend considerable time learning and adapting to the platform's operation methods. This makes it difficult for users to quickly and efficiently obtain required information and complete O&M tasks. The user experience of intelligent O&M platforms needs further improvement.

[0005] (4) Lack of the ability to conduct quick and convenient conversational queries on the health status of massive amounts of equipment Existing intelligent O&M platforms typically lack the ability to quickly and conveniently query the health status of massive amounts of grid equipment through interactive conversations. Operators must navigate complex menus and filter data to query the health information of each device, resulting in low efficiency and difficulty meeting the rapid information retrieval requirements of large-scale grid O&M. This makes it difficult to achieve the new intelligent O&M model of interactive querying of the health status of tens of thousands of devices. Summary of the Invention

[0006] The purpose of the present invention is to solve the following technical problems in the prior art: (1) It is difficult to effectively integrate and collaboratively utilize multi-source heterogeneous operation and maintenance data, resulting in incomplete equipment health status assessment and limited prediction accuracy; (2) Traditional machine learning methods have limitations in knowledge representation and reasoning capabilities, and are unable to fully utilize expert experience and industry knowledge to improve the level of intelligent decision-making; (3) The human-computer interaction method relies on graphical interface operation and lacks natural language interaction capabilities, which affects the efficiency and experience of operation and maintenance personnel in obtaining information; (4) There is a lack of an efficient conversational query mechanism for the health of massive power grid devices, which makes it difficult to meet the needs of rapid retrieval and knowledge services for large-scale devices; This paper proposes a power grid equipment health management and operation and maintenance platform based on knowledge federation and language model. By introducing the knowledge federation architecture, data security sharing and knowledge graph construction are realized. It also combines the cognitive intelligence and natural language processing capabilities of the large language model to improve the comprehensive performance of equipment health assessment, fault prediction, operation and maintenance strategy generation, and human-computer interaction.

[0007] In order to achieve the above objectives, the present invention adopts the following technical solutions: A power grid equipment health management and operation and maintenance platform based on knowledge federation and language models, including a multi-source data fusion module, a knowledge federation management system, a large language model, an intelligent analysis and prediction module, an operation and maintenance decision support module, and a visualization display module; The multi-source data fusion module is respectively connected to the knowledge federation management system and the large language model for accessing and integrating the multi-source heterogeneous operation and maintenance data of the power grid; The knowledge federation management system is respectively connected to the large language model and the intelligent analysis and prediction module, and utilizes the knowledge federation technology to integrate the multi-source heterogeneous operation and maintenance data of the power grid and construct a knowledge graph of power grid operation and maintenance; The large language model is communicated with the intelligent analysis and prediction module and the visualization display module respectively, and uses its cognitive intelligence and natural language interaction capabilities to implement the core AI functions of power grid equipment health management and intelligent operation and maintenance; The intelligent analysis and prediction module is respectively connected to the operation and maintenance decision support module and the visualization display module, and uses a large language model and a power grid operation and maintenance knowledge graph to perform intelligent analysis and prediction of power grid equipment; The operation and maintenance decision support module is in communication with the visualization display module, and generates an operation and maintenance decision plan for the power grid equipment based on the intelligent analysis and prediction results of the intelligent analysis and prediction module, combined with the power grid operation and maintenance business needs and expert knowledge; The visualization display module is used in the human-computer interaction window of the platform to visually display the conversational query results of the large language model, the analysis and prediction results of the intelligent analysis and prediction module, and the operation and maintenance decision-making plan of the operation and maintenance decision support module.

[0008] Specifically, the multi-source data fusion module is the data entry of the platform, located in the data access layer of the platform architecture. It is responsible for collecting data from different sources and different types of power grid multi-source heterogeneous operation and maintenance data sources, and performing data cleaning, preprocessing, conversion, and fusion operations. It converts the original, heterogeneous power grid multi-source heterogeneous operation and maintenance data into a unified format, high-quality multimodal data input, and provides standard and standardized data input for the knowledge federation management system and large language model. The knowledge federation management system is the data foundation and knowledge engine of the platform. Located at the bottom layer of the platform architecture, it is responsible for the access, integration, management, and secure sharing of multi-source heterogeneous power grid operation and maintenance data, as well as the construction, storage, and maintenance of the power grid operation and maintenance knowledge graph. The knowledge federation management system provides data and knowledge support for the large language model and intelligent analysis and prediction modules. The large language model is the platform's core intelligent engine, located in the middle layer of the platform architecture. It is responsible for implementing the core AI functions of power grid equipment health management and intelligent operation and maintenance. The large language model receives multimodal data input after fusion and processing from the multi-source data fusion module, and uses its own cognitive intelligence and knowledge reasoning capabilities to perform health status assessment of power grid equipment, fault risk prediction and warning, and intelligent operation and maintenance strategy generation. The large language model also directly provides an intelligent human-computer interaction interface for conversational query of power grid equipment health to the visualization display module. The intelligent analysis and prediction module is the intelligent analysis core of the platform. Located in the intelligent analysis layer of the platform architecture, it is responsible for invoking the large language model and the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform intelligent analysis and prediction of power grid equipment. The intelligent analysis and prediction module outputs the analysis and prediction results to the operation and maintenance decision support module and the visualization display module for subsequent operation and maintenance decision support and information display. The operation and maintenance decision support module is the intelligent decision center of the platform, located in the intelligent decision layer of the platform architecture. Based on the output results of the intelligent analysis and prediction module, combined with the preset operation and maintenance strategy model and expert knowledge, it automatically generates an operation and maintenance decision plan. The operation and maintenance decision support module outputs the intelligent operation and maintenance decision plan to the visualization display module for display. The visualization display module is the human-computer interaction window of the platform, located in the user interaction layer of the platform architecture, and is responsible for the platform's user interface design and information visualization display. The visualization display module receives analysis results and decision plans from the intelligent analysis and prediction module and the operation and maintenance decision support module, as well as conversational query results from the large language model, and displays key information of power grid equipment in various forms. The visualization display module provides a human-computer interaction interface, and users can conduct conversational queries on the health of power grid equipment and intelligent interactive operations through natural language.

[0009] 1. Knowledge Federation Management System Preferably, the knowledge federation management system adopts a hybrid federated learning architecture, including a data provider node, a federated learning server, a model aggregation module and a knowledge graph construction module; The data provider nodes represent different business departments of the power grid or data sources. Each data provider node has local operation and maintenance data and computing resources, participates in local model training, and uploads model parameter updates to the federated learning server; The federated learning server, as the central coordination node of the knowledge federation, is responsible for task scheduling and model aggregation of federated learning; it has a model aggregation module and a knowledge graph construction module deployed inside it; The model aggregation module is deployed on the federated learning server and is used to receive model parameter updates uploaded by each data provider node and aggregate the model parameters according to the federated aggregation algorithm to generate a globally shared federated learning model. The knowledge graph construction module is deployed on the federated learning server side, and is used to extract knowledge fragments from multi-source heterogeneous operation and maintenance data and integrate them into a globally unified power grid operation and maintenance knowledge graph.

[0010] 2. Large Language Model Preferably, the large language model is a DeepSeek-R1 large model, and the model architecture of the DeepSeek-R1 large model includes: Transformer Encoder-Decoder Architecture: This model uses the Transformer Encoder-Decoder architecture as its core network structure. The Transformer Encoder is responsible for deep feature extraction and knowledge representation of the input multimodal operation and maintenance data. The Transformer Decoder generates intelligent analysis results and operation and maintenance decision recommendations based on the features and knowledge extracted by the Encoder. Multimodal input fusion mechanism: The model's input layer inputs power grid operation and maintenance data in multiple modalities, including text, image, time series, and structured data. A multimodal feature fusion module is designed within the model to fuse and align data features from different modalities and learn the associations and complementary information between multimodal data. Knowledge Graph Enhancement Module: This module is integrated within the model to incorporate the grid operation and maintenance knowledge graph constructed by the knowledge federation management system into the model's training and reasoning processes. The module encodes the knowledge information in the knowledge graph into knowledge vector representations and fuses these knowledge vectors with multimodal data features. Lightweight and efficient deployment optimization: The model adopts a lightweight Transformer variant structure with fewer parameters and uses model compression technologies including model pruning, model quantization, and knowledge distillation. The model also optimizes the hardware platform for inference acceleration.

[0011] 3. Intelligent Analysis and Prediction Module In the present invention, the intelligent analysis and prediction module includes: 1. Equipment health status assessment The intelligent analysis and prediction module uses a large language model and a knowledge graph of power grid operation and maintenance built by knowledge federation to perform intelligent assessments of the health status of power grid equipment. The federated learning model provides basic model capabilities trained through multi-source data fusion, enhancing the large language model's ability to understand and reason about data from different sources and types, thereby improving the comprehensiveness and accuracy of equipment health status assessments. This includes: Multimodal data feature fusion: The intelligent analysis and prediction module obtains the fused and processed multimodal inspection data, operation data, and equipment ledger data of power grid equipment from the multi-source data fusion module. Leveraging the multimodal feature fusion capabilities of the large language model, it deeply fuses and aligns data features from different modalities and extracts fused feature representations. Knowledge-graph-enhanced equipment health assessment model: The intelligent analysis and prediction module constructs a power grid equipment health assessment model based on a large language model and knowledge graph enhancement. The power grid equipment health assessment model uses fused feature representations as input, retrieves and extracts relevant knowledge information from the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system, and integrates this knowledge information into the evaluation process of the power grid equipment health assessment model using the knowledge graph enhancement module. Based on the fused features and knowledge information, the power grid equipment health assessment model performs intelligent assessments of the health status of power grid equipment, outputting the equipment's health status level and health score. Multi-dimensional quantitative assessment of equipment health: The intelligent analysis and prediction module conducts a multi-dimensional quantitative assessment of the health of power grid equipment and outputs health indicators for each dimension, including insulation status, mechanical status, electrical performance, and adaptability to the operating environment.

[0012] 2. Fault risk prediction and early warning The intelligent analysis and prediction module uses a large language model and a knowledge federation-based grid operation and maintenance knowledge graph to perform intelligent prediction and early warning of grid equipment failure risks, including: Failure probability prediction based on historical data: The intelligent analysis and prediction module uses historical data integrated by knowledge federation, combined with the time series modeling and trend prediction capabilities of the large language model, to predict the failure probability of power grid equipment over the next period of time. The large language model learns from historical data about equipment operating modes, failure patterns, and equipment degradation trends to predict the likelihood of future equipment failures and outputs the equipment failure probability prediction results. Potential defect warning based on real-time inspection data: The intelligent analysis and prediction module uses a large language model to intelligently analyze real-time drone inspection data to identify existing grid equipment defects. It also mines and identifies potential early signs or abnormal conditions that have not yet developed into obvious defects, providing early warning of potential defects. This includes: analyzing small temperature rise anomalies in infrared thermal images to warn of potential joint heating failures; and analyzing weak discharge signals in ultraviolet images to warn of potential insulation degradation risks. Fault risk reasoning and early warning based on knowledge graph: The intelligent analysis and prediction module uses the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform intelligent reasoning and early warning of power grid equipment failure risks. The model conducts knowledge reasoning on the knowledge graph to explore the potential risk factors and causal relationships of equipment failures, including: discovering the knowledge pattern that "a certain type of transformer is prone to overheating failures when running for too long in a high-load, high-temperature environment" through reasoning, and based on the current equipment's operating data and environmental data, inferring and predicting the equipment's failure risk level and generating early warning information.

[0013] 3. Intelligent operation and maintenance strategy generation The intelligent analysis and prediction module uses a large language model and a knowledge federation-based grid operation and maintenance knowledge graph to automatically generate and optimize intelligent grid equipment operation and maintenance strategies, including: Intelligent maintenance recommendation generation based on equipment health: The intelligent analysis and prediction module automatically generates intelligent maintenance recommendations for power grid equipment based on equipment health assessment results and operation and maintenance knowledge in the knowledge graph. These recommendations include maintenance type, maintenance content, maintenance time, and spare parts. The maintenance recommendation generation process considers the actual health status and failure risk level of the equipment, as well as operation and maintenance procedures and expert experience. Intelligent inspection plan development based on resource optimization: The intelligent analysis and prediction module automatically develops and optimizes intelligent inspection plans based on the health status assessment results of power grid equipment, fault risk prediction results, and operation and maintenance resource constraints. Intelligent inspection plans include inspection target selection, inspection frequency optimization, inspection route planning, and inspection personnel and equipment scheduling. The intelligent inspection plan development process takes equipment status and resource constraints into consideration. Intelligent spare parts management based on demand forecasting: The intelligent analysis and forecasting module conducts intelligent forecasting of spare parts demand based on historical fault data of power grid equipment, equipment life prediction results, and spare parts inventory information, and generates intelligent spare parts management strategies. These strategies include spare parts demand forecasting, inventory optimization suggestions, procurement plan suggestions, and replacement cycle suggestions.

[0014] 4. Conversational query of power grid equipment health The intelligent analysis and prediction module builds a core function called "dialog query of power grid equipment health" based on a large language model. Users can interact with the platform in a natural and smooth manner through natural language to quickly obtain health information and operation and maintenance knowledge of massive power grid equipment, including: Understanding user query intent in natural language: The intelligent analysis and prediction module leverages the powerful natural language understanding capabilities of a large language model to perform in-depth semantic understanding of user queries about the health status of power grid equipment posed in natural language, accurately identifying the user's query intent. Knowledge retrieval and reasoning driven by the knowledge graph: The intelligent analysis and prediction module, based on understanding the user's query intent, utilizes the power grid operation and maintenance knowledge graph built by the knowledge federation management system to perform efficient knowledge retrieval and reasoning. Based on the user's query entity and query target, the large language model quickly locates relevant entities and relationships in the knowledge graph and performs knowledge reasoning to obtain the power grid equipment health information and operation and maintenance knowledge required by the user. Natural language query result generation and visualization: The intelligent analysis and prediction module retrieves and infers grid equipment health information and operation and maintenance knowledge from the knowledge graph, and leverages the powerful natural language generation capabilities of the large language model to automatically generate concise and easy-to-understand natural language query results. These are then presented to users through the visualization module. Query results include the equipment's real-time health score, multi-dimensional health indicators, historical health trend charts, potential fault risk levels, and intelligent maintenance recommendations. The visualization module visualizes conversational query results using both text and graphics, including various charts presenting equipment health indicators and knowledge relationships.

[0015] Beneficial effects of the present invention: 1. Innovatively introduce knowledge federation technology to build a knowledge base for power grid operation and maintenance and break down data silos.

[0016] The present invention innovatively applies knowledge federation technology to the fields of power grid equipment health management and intelligent operation and maintenance, and constructs a knowledge federation management system for power grid operation and maintenance. By utilizing the distributed data fusion and secure sharing mechanism of knowledge federation, the multi-source heterogeneous operation and maintenance data of the power grid (inspection data, operation data, historical maintenance data, expert knowledge base, operation and maintenance procedures and standards, etc.) is effectively integrated, breaking down data silos and building a unified, comprehensive, and authoritative power grid operation and maintenance knowledge base. Compared with the problem of insufficient data fusion capabilities of existing intelligent operation and maintenance platforms, the present invention can more fully and effectively utilize the value of massive power grid data, providing a solid data foundation for subsequent intelligent analysis, prediction, and decision-making.

[0017] 2. The breakthrough application of a large language model (DeepSeek-R1 large model) gives the platform powerful cognitive intelligence and natural language interaction capabilities.

[0018] The present invention has made a breakthrough in applying the advanced DeepSeek-R1 large model to the power grid equipment health management and intelligent operation and maintenance platform as the core intelligent engine of the platform. It fully utilizes the breakthrough AI capabilities of the DeepSeek-R1 large model in natural language understanding (NLU), natural language generation (NLG), knowledge reasoning, multimodal information fusion, etc., and gives the platform powerful cognitive intelligence and intelligent interaction capabilities. Compared with the existing intelligent operation and maintenance platforms that have insufficient knowledge representation and reasoning capabilities, and the problems of human-computer interaction methods that are not smart and convenient enough, the present invention can more effectively learn, understand and utilize the expert experience and industry knowledge in the power grid field, achieve a smarter and more humane human-computer interaction experience, and greatly improve the platform's intelligence level and user convenience.

[0019] 3. Build a conversational query function for power grid equipment health based on a large language model (DeepSeek-R1 large model) to innovate the operation and maintenance knowledge service model.

[0020] The present invention innovatively constructs a core function of "conversational query of power grid equipment health" based on the DeepSeek-R1 large model. Users can conduct natural and smooth conversational interactions with the platform through natural language, and quickly and conveniently query the real-time health assessment results, historical health trends, potential fault risks, operation and maintenance recommendations and other information of any power grid equipment. The powerful natural language understanding and generation capabilities of the DeepSeek-R1 large model enable the platform to accurately understand the user's query intent, and quickly retrieve and extract relevant knowledge from the massive knowledge graph constructed by the knowledge federation, generating concise and easy-to-understand natural language query results, and realizing a revolutionary innovation in the operation and maintenance knowledge service model. Compared with the existing operation and maintenance platforms that lack conversational query capabilities and have low information retrieval efficiency, the present invention can greatly improve the efficiency and convenience of operation and maintenance personnel in acquiring knowledge and information, and realize smarter and more efficient operation and maintenance knowledge services.

[0021] 4. Accurately assess the health status of power grid equipment and deeply predict fault risks, improving the intelligence level of operation and maintenance decision-making.

[0022] Based on the integration of multi-source heterogeneous grid data using knowledge federation technology and the powerful intelligent analysis and reasoning capabilities of the DeepSeek-R1 large-scale model, the platform of this invention can accurately assess the health status of grid equipment and deeply predict failure risks. Compared with existing intelligent operation and maintenance platforms that have insufficient data fusion capabilities, low knowledge utilization, and limited prediction accuracy, this invention can more comprehensively and accurately assess equipment health status and predict equipment failure risks more early and accurately. This provides more scientific and reliable intelligent support for grid operation and maintenance decision-making, significantly improving the intelligence and scientific nature of grid operation and maintenance decision-making.

[0023] 5. Build the ability to automatically generate and optimize intelligent operation and maintenance strategies to improve operation and maintenance efficiency and proactive prevention.

[0024] Based on the DeepSeek-R1 large model and knowledge federation technology, the platform of the present invention can realize the automatic generation and optimization of intelligent operation and maintenance strategies for power grid equipment. The platform can automatically generate the optimal maintenance strategy, inspection plan, spare parts management strategy, etc. based on multiple factors such as the health status of the equipment, failure risk, and operation and maintenance resource constraints. Compared with the existing intelligent operation and maintenance platforms that lack the ability to generate intelligent operation and maintenance strategies and the problem that the formulation of operation and maintenance strategies mainly relies on manual experience, the present invention can greatly improve the efficiency and scientific nature of the formulation of operation and maintenance strategies, realize the transformation from passive maintenance to active preventive maintenance, and improve the work efficiency and active preventive level of power grid operation and maintenance.

[0025] In summary, the power grid equipment health management and intelligent operation and maintenance platform based on knowledge federation and large language model (DeepSeek-R1 large model) proposed in this invention innovatively integrates knowledge federation technology and the breakthrough AI capabilities of the DeepSeek-R1 large model, and has achieved significant technological innovation and functional breakthroughs in many aspects such as data fusion, knowledge utilization, human-computer interaction, and intelligent decision-making. It can effectively overcome the limitations of existing power grid operation and maintenance technologies, comprehensively improve the intelligence and proactive preventive level of power grid equipment operation and maintenance, and provide key technical support for building a safer, more reliable, and more efficient smart grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is the overall architecture of the power grid equipment health management and intelligent operation and maintenance platform of the present invention; Figure 2 This is the knowledge federated learning architecture of the present invention; Figure 3 This is the DeepSeek-R1 architecture of the present invention. DETAILED DESCRIPTION

[0027] The following will provide a clear and complete description of the concept, specific structure and technical effects of the present invention in conjunction with the embodiments and drawings, so as to fully understand the purpose, features and effects of the present invention.

[0028] Example 1 A power grid equipment health management and operation and maintenance platform based on knowledge federation and language models. This platform innovatively integrates knowledge federation technology and the powerful AI capabilities of the DeepSeek-R1 large model to build a new generation of intelligent power grid operation and maintenance platform. Its core idea is to build an intelligent power grid operation and maintenance ecosystem that is "data-driven, knowledge-enabled, intelligently decided, and human-machine collaborative."

[0029] like Figure 1 As shown in the figure, this platform is mainly composed of the following six core modules, which together build a complete power grid equipment health management and intelligent operation and maintenance platform: (1) Knowledge federation management system: The platform's data foundation and knowledge engine uses knowledge federation technology to integrate multi-source heterogeneous operation and maintenance data of the power grid, build a power grid operation and maintenance knowledge graph, realize secure data sharing, unified management of knowledge and continuous evolution, and provide comprehensive and authoritative data and knowledge support for the platform.

[0030] (2) DeepSeek-R1 large model: The core intelligent engine of the platform, which uses the powerful cognitive intelligence and natural language interaction capabilities of the DeepSeek-R1 large model to perform intelligent assessments of the health status of power grid equipment, accurately predict fault risks, and automatically generate intelligent operation and maintenance strategies. It also supports intelligent human-computer interaction for conversational queries on the health status of 10,000-level equipment, providing the platform with core intelligent analysis and decision-making capabilities.

[0031] (3) Multi-source data fusion module: The data entry of the platform is responsible for accessing and integrating different types of power grid operation and maintenance data from different sources, such as drone inspection data, SCADA operation data, PMS equipment inventory data, GIS geographic information data, historical defect data, expert knowledge base, operation and maintenance procedures and standards, etc., providing rich and multi-dimensional data input for the knowledge federation management system and DeepSeek-R1 large model.

[0032] (4) Intelligent analysis and prediction module: The intelligent analysis core of the platform uses technologies such as the DeepSeek-R1 large model and knowledge graph to realize core intelligent analysis functions such as intelligent assessment of the health status of power grid equipment, accurate prediction of fault risks, equipment life prediction, and degradation trend analysis, providing the platform with accurate and reliable equipment status perception and risk warning capabilities.

[0033] (5) Operation and maintenance decision support module: The platform's intelligent decision center, based on the output results of the intelligent analysis and prediction module, combines the power grid operation and maintenance business needs and expert knowledge, and automatically generates intelligent operation and maintenance decision plans such as maintenance recommendations for power grid equipment, intelligent inspection plans, and spare parts management strategies, providing scientific decision support and operational guidance for power grid operation and maintenance personnel.

[0034] (6) Visual display module: The platform's human-computer interaction window provides a friendly and intuitive user interface. It uses a variety of forms such as large data screens, three-dimensional visualization models, data reports, and knowledge graphs to visually display key information such as the health status, fault risks, and operation and maintenance recommendations of power grid equipment. It also supports users to conduct conversational queries on the health status of 10,000-level equipment through natural language, achieving a convenient and efficient human-computer interaction experience.

[0035] The core technical feature of this platform lies in the innovative integration of two cutting-edge technologies: knowledge federation and the DeepSeek-R1 large-scale model. This platform builds a new generation of intelligent power grid operation and maintenance platform, achieving the following key technical advantages: (1) Data fusion and knowledge sharing based on knowledge federation: Use knowledge federation technology to effectively integrate multi-source heterogeneous operation and maintenance data of the power grid, break data silos, build a unified and comprehensive power grid operation and maintenance knowledge base, and provide solid data and knowledge support for the platform's intelligent applications.

[0036] (2) Cognitive intelligence and natural language interaction based on the DeepSeek-R1 large model: The breakthrough application of the DeepSeek-R1 large model as the core intelligent engine gives the platform powerful cognitive intelligence and natural language interaction capabilities, significantly improving the platform's intelligence level and user experience.

[0037] (3) Innovative knowledge service model for conversational query of the health status of 10,000-level equipment: Innovatively construct the core function of "conversational query of the health status of 10,000-level equipment" to achieve a smarter, more efficient and more convenient operation and maintenance knowledge service model, greatly improving the efficiency and intelligence level of operation and maintenance work.

[0038] (4) Accurate equipment health status assessment and in-depth fault risk prediction: Based on deep data analysis and knowledge mining of knowledge federation and DeepSeek-R1 large model, accurate assessment of the health status of power grid equipment and in-depth prediction of fault risks are achieved, providing a scientific basis for proactive preventive maintenance of the power grid.

[0039] (5) Automatic generation and optimization of intelligent operation and maintenance strategies: The platform can automatically generate optimal maintenance recommendations and intelligent operation and maintenance strategies based on equipment health status, failure risks, and operation and maintenance resource constraints, thereby improving the scientificity and intelligence of operation and maintenance decisions.

[0040] Through the above technical solutions, the platform of the present invention can achieve smarter, more efficient, more convenient, safer and more reliable power grid equipment health management and intelligent operation and maintenance, build a new generation of intelligent operation and maintenance system for power grid enterprises, improve the inherent safety level and operation and maintenance efficiency of the power grid, reduce operation and maintenance costs, and provide strong support for the high-quality development of the power grid. Example 2 This embodiment further elaborates on the power grid equipment health management and intelligent operation and maintenance platform based on the above embodiment. The overall architecture of the power grid equipment health management and intelligent operation and maintenance platform based on knowledge federation and DeepSeek-R1 large model is as follows: Figure 1 The platform adopts a modular and layered design and consists of six core modules: the knowledge federation management system, the DeepSeek-R1 large model, the multi-source data fusion module, the intelligent analysis and prediction module, the operation and maintenance decision support module, and the visualization display module. These modules work together to realize the core functions of power grid equipment health management and intelligent operation and maintenance.

[0041] like Figure 1 As shown in the figure, the functions and relationships of each core module are as follows: (1) Knowledge federation management system The platform's data foundation and knowledge engine, located at the bottom layer of the platform architecture, are responsible for accessing, integrating, managing, and securely sharing multi-source, heterogeneous grid operations and maintenance data, as well as constructing, storing, and maintaining the grid operations and maintenance knowledge graph. The knowledge federation management system provides comprehensive, authoritative, and high-quality data and knowledge support for the DeepSeek-R1 large model and intelligent analysis and prediction modules, and is the cornerstone of the platform's intelligent operations and maintenance.

[0042] (2) DeepSeek-R1 large model The platform's core intelligent engine, located in the middle layer of the platform architecture, is responsible for implementing the core AI functions of power grid equipment health management and intelligent operation and maintenance. The DeepSeek-R1 large model receives multimodal data input from the multi-source data fusion module after fusion and processing. It leverages its powerful cognitive intelligence and knowledge reasoning capabilities to intelligently assess the health status of power grid equipment, accurately predict fault risks, and automatically generate intelligent operation and maintenance strategies. The DeepSeek-R1 large model also provides an intelligent human-computer interaction interface for conversational queries on the health of 10,000-level devices directly to the visualization display module, serving as the core driver for the platform's intelligent operation and maintenance decision-making and intelligent user interaction.

[0043] (3) Multi-source data fusion module The platform's data entry, located in the data access layer of the platform architecture, is responsible for collecting data from diverse sources and types of grid operation and maintenance data sources. It then performs data cleaning, preprocessing, conversion, and fusion operations, converting raw, heterogeneous grid operation and maintenance data into high-quality, multimodal data input in a unified format. This provides standardized and normalized data input for the knowledge federation management system and the DeepSeek-R1 large model. The multi-source data fusion module is a key enabler for the platform's data-driven intelligent operation and maintenance.

[0044] (4) Intelligent analysis and prediction module The platform's intelligent analysis core, located in the intelligent analysis layer of the platform architecture, is responsible for invoking the DeepSeek-R1 large model and the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform core intelligent analysis functions such as intelligent assessment of power grid equipment health, accurate prediction of failure risks and equipment lifespan, and degradation trend analysis. The intelligent analysis and prediction module outputs analysis and prediction results to the operation and maintenance decision support module and the visualization display module for subsequent operation and maintenance decision support and information display. The intelligent analysis and prediction module is the key support for the platform's intelligent operation and maintenance decision-making.

[0045] (5) Operation and maintenance decision support module The platform's intelligent decision-making center, located in the intelligent decision-making layer of the platform architecture, automatically generates intelligent O&M decision-making solutions, including maintenance recommendations for power grid equipment, intelligent inspection plans, and spare parts management strategies, based on the output of the intelligent analysis and prediction module, combined with pre-set O&M strategy models and expert knowledge. The O&M decision-making support module outputs these intelligent O&M decision-making solutions to the visualization module for display, assisting power grid O&M personnel in making informed decisions and ensuring efficient operations. The O&M decision-making support module is the platform's core functional module for implementing intelligent O&M management and improving O&M efficiency.

[0046] (6) Visual display module The platform's human-computer interaction window, located in the user interaction layer of the platform architecture, is responsible for the platform's user interface design and information visualization. The visualization module receives analysis results and decision-making solutions from the intelligent analysis and prediction module and the operation and maintenance decision support module, as well as conversational query results from the DeepSeek-R1 large model. It intuitively and clearly displays key information such as the health status of power grid equipment, fault risks, and operation and maintenance recommendations in a variety of formats, including large data screens, 3D visualization models, data reports, and knowledge graphs. The visualization module provides a friendly and convenient human-computer interaction interface, supporting intelligent interactive operations such as conversational queries on the health status of 10,000-level devices through natural language. The visualization module serves as a critical bridge for information exchange and functional operations between the platform and users.

[0047] Example 3 This embodiment further elaborates on the knowledge federation management system based on the above embodiment. The knowledge federation management system is the data foundation and knowledge engine of this platform. Its detailed architecture and functional components are as follows: Figure 2 shown.

[0048] 1. System Architecture like Figure 2 As shown in the figure, the knowledge federation management system adopts a hybrid federated learning architecture, which is mainly composed of the following key components: (1) Data provider node Representing different business departments or data sources of the power grid, such as transmission and maintenance departments, substation and maintenance departments, distribution and maintenance departments, dispatch centers, meteorological departments, and power research institutions. Each data provider node possesses its own local operation and maintenance data and computing resources. Data is stored locally and not uploaded to the central server to ensure data security and privacy. Data provider nodes are responsible for local data cleaning, preprocessing, and feature engineering, participate in federated learning model training, and upload model parameter updates to the federated learning server.

[0049] (2) Federated Learning Server As the central coordination node for knowledge federation, it is responsible for task scheduling and model aggregation for federated learning. It internally deploys a model aggregation module and a knowledge graph construction module. The federated learning server does not access any raw data and is only responsible for receiving model parameter updates uploaded by each data provider node. It aggregates the model parameters using federated aggregation algorithms (such as FedAvg and FedProx) to generate a globally shared federated learning model. The federated learning server is also responsible for scheduling knowledge graph construction tasks and integrating knowledge, fusing the local knowledge fragments contributed by each data provider node into a globally unified power grid operation and maintenance knowledge graph.

[0050] (3) Model aggregation module Deployed on the federated learning server side, it is responsible for aggregating the federated learning model parameters. The model aggregation module receives model parameter updates uploaded by each data provider node, and performs weighted averaging or other forms of aggregation on the model parameters according to the federated aggregation algorithm (such as Fed Avg, Fed Prox) to generate a globally shared federated learning model. The federated learning model integrates the knowledge characteristics of multi-source heterogeneous operation and maintenance data, and can be used for subsequent equipment health status assessment, fault risk prediction, and knowledge graph construction tasks. It is the key technical support for the platform to achieve cross-departmental collaborative modeling and knowledge sharing. The model aggregation module is also responsible for monitoring and managing the federated learning process, such as monitoring model training progress, evaluating model performance, and handling abnormal situations.

[0051] (4) Knowledge graph construction module Deployed on the federated learning server, it is responsible for the construction and management of the power grid operation and maintenance knowledge graph. The knowledge graph construction module extracts knowledge (e.g., entity recognition, relationship extraction, and attribute extraction) from the multi-source heterogeneous operation and maintenance data integrated by the knowledge federation, constructs local knowledge fragments, and integrates the local knowledge fragments contributed by each data provider node into a globally unified power grid operation and maintenance knowledge graph. The knowledge graph construction module is also responsible for storing (e.g., graph databases such as Neo4j and JanusGraph), updating, and maintaining the knowledge graph, as well as providing query and access interfaces for the knowledge graph.

[0052] 2. Data Sources and Federation Strategy The multi-source heterogeneous power grid operation and maintenance data integrated by the knowledge federation management system mainly includes the following types: (1) Drone inspection data: Multimodal inspection data collected by drone intelligent inspection systems, including visible light images, infrared thermal images, ultraviolet images, lidar point cloud data, gas sensor data, as well as structured and unstructured data such as defect identification results, early warning information, and inspection reports. Drone inspection data reflects the external status and operating environment of power grid equipment and is an important data source for equipment health assessment and fault risk prediction.

[0053] (2) SCADA system operation data: Real-time equipment operation data collected from the power grid SCADA system, including various operating parameters and status quantities such as voltage, current, active power, reactive power, frequency, temperature, pressure, oil level, SF6 gas pressure, as well as alarm information, event records, etc. SCADA system operation data reflects the internal operating status and electrical performance indicators of power grid equipment and is a key data source for equipment health status assessment and failure risk prediction.

[0054] (3) PMS equipment inventory data: Equipment inventory data from the power PMS equipment management system, including basic equipment information (equipment name, model, specifications, manufacturer, production date, installation date, etc.), equipment parameter information (rated voltage, rated current, rated capacity, insulation level, etc.), equipment location information (substation, line, geographic coordinates, etc.), equipment historical maintenance records, equipment defect records, spare parts information, and other structured data. PMS equipment inventory data provides static information and historical operation and maintenance information of power grid equipment, and is an important supplementary data for knowledge graph construction and equipment status assessment.

[0055] (4) GIS geographic information data: Grid geographic information data from the power GIS geographic information system, including geospatial data such as line directions, tower locations, substation distribution, geographic environment information, and meteorological environment information. GIS geographic information data provides geographic location and environmental information of power grid equipment, providing environmental background information for equipment status assessment and fault risk prediction.

[0056] (5) Historical defect data: Historical defect data of power grid equipment from historical manual inspection records, defect management systems, etc., including structured and unstructured data such as defect type, defect level, defect location, defect description, defect occurrence time, and defect handling records. Historical defect data reflects the historical failures and defect occurrences of power grid equipment, providing historical experience and data support for fault risk prediction and equipment degradation trend analysis.

[0057] (6) Expert Knowledge Base: The accumulated expertise and experience of experts in the power grid field, including structured, semi-structured, and unstructured knowledge such as knowledge on the structure and principles of power grid equipment, operation and maintenance knowledge, fault diagnosis knowledge, maintenance procedures and standards, and expert experience rules. The expert knowledge base provides the accumulated expertise and experience in the field of power grid operation and maintenance, providing an important source of knowledge for the construction of the knowledge graph and the knowledge enhancement of the DeepSeek-R1 large model.

[0058] (7) Operation and maintenance procedures and standards: These are documents and materials related to the operation and maintenance of power grid equipment issued by the State Grid Corporation of China, industry associations, and standardization organizations. Operation and maintenance procedures and standards specify normative knowledge such as the operation and maintenance process, inspection and repair standards, defect judgment standards, and safe operating procedures for power grid equipment, providing an authoritative basis for knowledge graph construction and operation and maintenance decision support.

[0059] The knowledge federation management system uses a data federation strategy to integrate multi-source data. Its main features include: (1) Data does not leave the domain, protecting data privacy and security Raw O&M data is stored locally on each data provider's node and not uploaded to a central server. Data ownership and management remain with each data provider. During federated learning, only intermediate calculation results, such as model parameter updates, are exchanged, without leaking raw data. This effectively ensures the security and privacy of sensitive grid O&M data, meeting the power industry's stringent data security requirements.

[0060] (2) Model federated learning, joint training of global shared models Model training is performed using a federated learning approach. Each data provider node trains a local model using local data and uploads updated model parameters to the federated learning server for aggregation, generating a globally shared federated learning model. The federated learning model integrates the data and knowledge of each data provider, providing more comprehensive data coverage and stronger generalization capabilities, effectively addressing data sparsity and model overfitting.

[0061] (3) Knowledge graph federation construction, integrating multi-source knowledge to form a unified view A knowledge graph federation approach is used to integrate multi-source O&M knowledge. Each data provider node extracts local knowledge fragments from local data and contributes these fragments to the federated learning server for integration, thereby constructing a globally unified power grid O&M knowledge graph. The knowledge graph federation process also adheres to the principle of data confinement to the domain, protecting intellectual property and trade secrets. By integrating multi-source knowledge, the global knowledge graph offers a more comprehensive and complete knowledge representation and enhanced knowledge reasoning capabilities.

[0062] Example 4 This embodiment further elaborates on the DeepSeek-R1 large model based on the above embodiments. The DeepSeek-R1 large model is the core intelligent engine of this platform. Its model architecture, training and fine-tuning process, and core technical features are as follows: 1. Model architecture and type like Figure 3As shown, the DeepSeek-R1 large model uses a Transformer-based deep neural network architecture and falls into the category of pre-trained language models (PLM). However, it has been customized, improved, and optimized for power grid equipment health management and intelligent operation and maintenance tasks, making it more effectively adaptable to the professional knowledge and complex application scenarios in the power grid field. The main features of the DeepSeek-R1 large model architecture include: (1) Transformer Encoder-Decoder Architecture The model uses a Transformer Encoder-Decoder architecture as its core network structure. The Encoder is responsible for deep feature extraction and knowledge representation of the input multimodal O&M data, while the Decoder generates intelligent analysis results and O&M decision recommendations based on the features and knowledge extracted by the Encoder. The Transformer architecture has powerful sequence modeling capabilities and the ability to capture long-range dependencies, effectively processing the temporal characteristics and contextual relationships of power grid O&M data.

[0063] (2) Multimodal input fusion mechanism The model's input layer supports grid operation and maintenance data input in multiple modalities, including text (e.g., O&M procedures and standards, expert knowledge base documents), image (e.g., drone inspection images), time series (e.g., SCADA system operational data), and structured data (e.g., PMS equipment inventory data). A multimodal feature fusion module is designed within the model to effectively fuse and align data features from different modalities, learning the associations and complementary information between multimodal data, and enhancing the model's comprehensive perception and understanding of the health status of grid equipment.

[0064] (3) Knowledge graph enhancement module The model integrates a knowledge graph enhancement module, which integrates the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system into the model's training and reasoning processes. This module encodes knowledge information such as entities, relationships, and attributes in the knowledge graph into knowledge vector representations and fuses these knowledge vectors with multimodal data features, enhancing the model's knowledge representation and reasoning capabilities, and improving its understanding and application of power grid domain expertise.

[0065] (4) Lightweight and efficient deployment optimization During the model's architecture design and training, the actual deployment environment and resource constraints of the power grid operation and maintenance platform were fully considered, and lightweight and efficient deployment optimizations were implemented. The model adopts a lightweight Transformer variant structure with a smaller number of parameters, and uses model compression techniques such as model pruning, quantization, and knowledge distillation to reduce model size and computational complexity. The model also performs inference acceleration optimization for common hardware platforms (such as GPUs and CPUs) to improve the model's operating efficiency and real-time performance. Lightweight and efficient deployment optimizations enable the DeepSeek-R1 large model to be efficiently and stably deployed and applied in the actual operating environment of the power grid operation and maintenance platform.

[0066] 2. Model training and fine-tuning The training process of the DeepSeek-R1 large model includes the pre-training phase and the fine-tuning phase: (1) Pre-training stage The DeepSeek-R1 large model is first pre-trained on a large-scale general corpus to learn general language knowledge and world knowledge. The pre-training corpus includes massive amounts of Internet text data, encyclopedia knowledge, book documents, etc., covering a wide range of fields and topics. The pre-training stage adopts self-supervised learning methods (such as Masked Language Model, MLM; Next Sentence Prediction, NSP). Through training on massive data, the model learns general language representation and knowledge patterns, and has strong general knowledge and language understanding capabilities, laying a solid foundation for subsequent fine-tuning in the field of power grid operation and maintenance. The pre-training process of the DeepSeek-R1 large model is completed by a professional AI model manufacturer. This solution directly uses the pre-trained DeepSeek-R1 large model as the base model.

[0067] (2) Fine-tuning stage Based on the pre-training phase, we use the multi-source heterogeneous power grid operation and maintenance data integrated by the knowledge federation management system to fine-tune the domain knowledge, making the DeepSeek-R1 large model more effectively adaptable to the specific tasks and application scenarios of power grid equipment health management and intelligent operation and maintenance. The fine-tuning phase uses supervised learning methods for model training, and the training data includes: 1) Power Grid Equipment Health Assessment Dataset: This dataset contains multimodal inspection data (visible light images, infrared thermal images, etc.) of power grid equipment and corresponding equipment health status labels (e.g., healthy, minor defect, moderate defect, severe defect). By learning from this data, the model understands the mapping between different modal inspection data and equipment health status, improving the accuracy of equipment health assessment.

[0068] 2) Power Grid Equipment Failure Risk Prediction Dataset: This dataset contains multi-dimensional data, including historical operational data (such as SCADA data), historical defect data, and environmental data, along with corresponding failure labels (e.g., normal operation, failure). By learning from this data, the model understands the correlation between different dimensional data and equipment failure risks, improving the accuracy and lead time of equipment failure risk prediction.

[0069] 3) Power Grid Operation and Maintenance Knowledge Graph: Utilizing the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system as auxiliary knowledge information, the model's knowledge representation and reasoning capabilities are enhanced. During model fine-tuning, knowledge information such as entities, relationships, and attributes in the knowledge graph is incorporated into the model training process. For example, knowledge graph embedding technology is used to encode this knowledge information into knowledge vectors, which are then integrated with input data features. The introduction of the knowledge graph can effectively enhance the model's understanding and application of power grid domain expertise.

[0070] 4) Conversational Query and Response Dataset: A conversational query and response dataset was constructed for the "10,000-level device health query" task. This dataset contains various natural language queries about the health status of power grid equipment, along with the platform's expected, accurate and concise natural language responses. By learning from this data, the model understands the user's query intent and learns how to retrieve and extract relevant knowledge from the knowledge graph, generating natural and fluent conversational responses.

[0071] During the model fine-tuning phase, a multi-task joint training strategy was employed. Multiple O&M tasks, such as equipment health assessment, fault risk prediction, intelligent O&M strategy generation, and conversational query response, were integrated into the DeepSeek-R1 large model framework for joint training. This leveraged the advantages of multi-task learning to improve the model's generalization and multi-task processing capabilities. Fine-tuning was employed during the model fine-tuning process. Based on the pre-trained model, only a small number of model parameters were fine-tuned or a small number of task-specific adaptation layers were added. This avoided training the large model from scratch, reducing computational and time costs.

[0072] Example 5 Based on the above embodiments, this embodiment further elaborates on the intelligent analysis and prediction module. The intelligent analysis and prediction module is the intelligent analysis core of this platform. It is responsible for calling the DeepSeek-R1 large model and the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to realize core intelligent analysis functions such as power grid equipment health status assessment, fault risk prediction, equipment life prediction, and degradation trend analysis.

[0073] 1. Equipment health status assessment The intelligent analysis and prediction module uses the DeepSeek-R1 large model and the grid operation and maintenance knowledge graph built by knowledge federation to perform intelligent assessments of the health status of grid equipment. Its core technologies and processes include: (1) Multimodal data feature fusion The intelligent analysis and prediction module first obtains fused and processed multimodal inspection data (visible light images, infrared thermal images, etc.), operational data (SCADA data, etc.), and equipment inventory data (PMS data, etc.) from the multi-source data fusion module. Leveraging the multimodal feature fusion capabilities of the DeepSeek-R1 large model, it deeply fuses and aligns data features from different modalities to extract fused feature representations. Multimodal data fusion fully leverages the complementary information between data from different modalities, improving the comprehensiveness and accuracy of device status perception.

[0074] (2) Knowledge graph-enhanced equipment health status assessment model The intelligent analysis and prediction module constructs a power grid equipment health status assessment model based on the DeepSeek-R1 large model and knowledge graph enhancement. The model uses fused feature representation as input, retrieves and extracts relevant knowledge information from the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system, and uses the knowledge graph enhancement module to integrate knowledge information into the model's evaluation process. Based on fused features and knowledge information, the model conducts an intelligent assessment of the health status of power grid equipment, outputting the equipment's health status level (e.g., healthy, minor defects, moderate defects, severe defects) and health score (e.g., a health score of 0-100 points). The introduction of the knowledge graph can effectively enhance the model's understanding and application of professional knowledge in the power grid field, and improve the accuracy and reliability of equipment health status assessments.

[0075] (3) Multi-dimensional quantitative assessment of equipment health The intelligent analysis and prediction module performs a multi-dimensional quantitative assessment of the health of power grid equipment. It not only provides an overall health score for the equipment but also provides a granular assessment of the equipment's health from various dimensions (such as insulation status, mechanical condition, electrical performance, and adaptability to the operating environment), outputting health indicators for each dimension. This multi-dimensional quantitative assessment provides a more comprehensive and in-depth view of the equipment's health, providing operators with more refined status information to aid in decision-making.

[0076] 2. Fault risk prediction and early warning The intelligent analysis and prediction module uses the DeepSeek-R1 large model and the power grid operation and maintenance knowledge graph built by knowledge federation to perform intelligent prediction and early warning of power grid equipment failure risks. Its core technologies and processes include: (1) Failure probability prediction based on historical data The intelligent analysis and prediction module leverages historical data, including historical operation data, defect data, and maintenance data of power grid equipment, integrated through knowledge federation. Combined with the time series modeling and trend prediction capabilities of the DeepSeek-R1 large-scale model, it predicts the probability of future failure of power grid equipment (e.g., the next month or the next year). By learning from historical data about equipment operating patterns, failure patterns, and equipment degradation trends, the model predicts the likelihood of future equipment failures and outputs a predicted failure probability for the equipment. This failure probability prediction provides data support for power grid operations and maintenance departments to proactively identify high-risk equipment and develop preventive maintenance plans.

[0077] (2) Potential defect warning based on real-time inspection data The intelligent analysis and prediction module leverages the DeepSeek-R1 large-scale model to intelligently analyze real-time drone inspection data. This module not only identifies existing grid equipment defects but also uncovers and identifies potential early warning signs or abnormal conditions that haven't yet developed into overt defects, providing early warning of potential defects. For example, by analyzing minute temperature rise anomalies in infrared thermal images, it can warn of potential connector overheating failures; by analyzing weak discharge signals in ultraviolet images, it can warn of potential insulation degradation risks. These potential defect warnings help grid operations and maintenance departments identify equipment risks earlier, enabling them to intervene before failures occur and proactively prevent them.

[0078] (3) Fault risk reasoning and early warning based on knowledge graph The intelligent analysis and prediction module utilizes the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform intelligent reasoning and early warning of power grid equipment failure risks. By performing knowledge reasoning on the knowledge graph, the model uncovers potential risk factors and causal relationships leading to equipment failures. For example, through reasoning, it discovers the knowledge pattern that "a certain type of transformer is prone to overheating failures when operated for too long in a high-load, high-temperature environment." Based on the current equipment's operating and environmental data, it infers and predicts the equipment's failure risk level and generates early warning information. Knowledge graph-based fault risk reasoning and early warning can more deeply uncover the underlying causes of failures, providing more explainable and targeted early warning information.

[0079] 3. Intelligent operation and maintenance strategy generation The intelligent analysis and prediction module uses the DeepSeek-R1 large model and the grid operation and maintenance knowledge graph built by knowledge federation to automatically generate and optimize intelligent operation and maintenance strategies for grid equipment. Its core technologies and processes include: (1) Generation of intelligent maintenance recommendations based on equipment health status The intelligent analysis and prediction module automatically generates intelligent maintenance recommendations for power grid equipment based on equipment health assessment results and the operational knowledge captured in the knowledge graph. These recommendations include maintenance type recommendations (e.g., condition-based maintenance, scheduled maintenance, emergency repair), maintenance content recommendations (e.g., replacing insulators, tightening bolts, cleaning equipment), maintenance timing recommendations (e.g., during the next power outage), and spare parts recommendations. The maintenance recommendation generation process fully considers the actual health status and failure risk level of the equipment, as well as operational and maintenance procedures and expert experience, ensuring the scientific, reasonable, and operational nature of the recommendations.

[0080] (2) Intelligent inspection plan formulation based on resource optimization The intelligent analysis and prediction module automatically develops and optimizes intelligent inspection plans based on power grid equipment health assessments, fault risk predictions, and O&M resource constraints (e.g., number of inspection personnel, number of drones, inspection time windows, etc.). Intelligent inspection plans include inspection target selection (prioritizing equipment with poor health and high failure risk), inspection frequency optimization (increasing inspection frequency for high-risk equipment and reducing it for low-risk equipment), inspection route planning (optimizing inspection routes to reduce mileage and time), and dispatching inspection personnel and equipment. The intelligent inspection plan development process fully considers equipment status and resource constraints, optimizing O&M resource allocation and maximizing inspection efficiency.

[0081] (3) Intelligent management of spare parts based on demand forecasting The intelligent analysis and prediction module uses historical fault data, equipment lifespan predictions, and spare parts inventory information from power grid equipment to intelligently predict spare parts demand and generate intelligent spare parts management strategies. These strategies include spare parts demand forecasts, inventory optimization recommendations, procurement plan suggestions, and replacement cycle recommendations. Intelligent spare parts management can help power grid operations and maintenance departments more accurately predict spare parts demand, optimize spare parts inventory management, reduce inventory costs, ensure timely spare parts supply, and enhance operations and maintenance capabilities.

[0082] 4. Conversational query of 10,000-level device health (core function) The intelligent analysis and prediction module innovatively builds a core function called "conversational query of 10,000-level device health" based on the DeepSeek-R1 large-scale model, providing convenient and efficient intelligent operation and maintenance knowledge services for power grid operators. Users can interact with the platform in a natural and smooth conversational manner through natural language, quickly obtaining health information and operation and maintenance knowledge for massive amounts of power grid equipment. Its core technologies and processes include: (1) Natural language user query intention understanding The intelligent analysis and prediction module leverages the powerful natural language understanding (NLU) capabilities of the DeepSeek-R1 large-scale model to perform deep semantic understanding of natural language queries about the health status of power grid equipment, accurately identifying the user's query intent. For example, if a user enters "Query the latest health status of main transformer No. 1," the platform accurately understands that the user is looking for the "health status" of "main transformer No. 1," specifically the "latest" health status. The powerful natural language understanding capabilities of the DeepSeek-R1 large-scale model are the key technical support for intelligent conversational querying.

[0083] (2) Knowledge retrieval and reasoning driven by knowledge graphs Based on understanding the user's query intent, the intelligent analysis and prediction module uses the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform efficient knowledge retrieval and reasoning. Based on the user's query entity (such as "No. 1 main transformer") and query target (such as "health status"), the model quickly locates relevant entities and relationships in the knowledge graph and performs knowledge reasoning to obtain the power grid equipment health information and operation and maintenance knowledge required by the user. For example, based on the "No. 1 main transformer" entity, its "health" attribute value is retrieved in the knowledge graph, and related knowledge such as its "historical health trend", "potential failure risk", and "maintenance recommendations" are inferred. The efficient knowledge retrieval and reasoning capabilities of the knowledge graph are the key knowledge source for realizing the conversational query function of the health status of 10,000-level equipment.

[0084] (3) Natural language query result generation and visualization The intelligent analysis and prediction module retrieves and infers grid equipment health information and operation and maintenance knowledge from the knowledge graph, and leverages the powerful natural language generation (NLG) capabilities of the DeepSeek-R1 large model to automatically generate concise and easy-to-understand natural language query results. These are then presented to users through the visualization module. Query results can include real-time health scores for devices, multi-dimensional health indicators, historical health trend charts, potential fault risk levels, intelligent maintenance recommendations, and other content. The powerful natural language generation capabilities of the DeepSeek-R1 large model enable the platform to present complex equipment health information and operation and maintenance knowledge to users in a natural and fluent language, significantly improving the user experience. The visualization module visualizes the results of conversational queries using a combination of graphics and text. For example, it presents equipment health indicators and knowledge relationships in various chart formats, such as dashboards, trend charts, and knowledge graphs, allowing users to more intuitively and clearly understand the query results.

[0085] Example 6 Based on the above embodiments, this embodiment further elaborates on the operation and maintenance decision support module. The operation and maintenance decision support module provides multi-dimensional intelligent decision support for power grid operation and maintenance personnel based on the output results of the intelligent analysis and prediction module, assisting operation and maintenance personnel in making scientific decisions and efficient operations.

[0086] 1. Generation of maintenance suggestions The operation and maintenance decision support module automatically generates intelligent maintenance recommendations for power grid equipment based on the grid equipment health status assessment and fault risk prediction results output by the intelligent analysis and prediction module, combined with the operation and maintenance procedures and expert experience knowledge in the grid operation and maintenance knowledge graph constructed by the knowledge federation management system. The maintenance recommendation generation process fully considers multiple factors such as the actual health status of the equipment, fault risk level, operation and maintenance cost constraints, and safety and reliability requirements, striving to generate the optimal maintenance plan to achieve accurate maintenance and maximize the value of the equipment. Maintenance recommendations mainly include: (1) Maintenance type recommendations: Based on the health status of the equipment and the risk level of failure, the most appropriate maintenance type is intelligently recommended, such as condition-based maintenance (determining the maintenance strategy based on the actual status of the equipment), scheduled maintenance (performing maintenance according to a predetermined period), emergency repair (performing emergency repair for serious failures that have occurred or are about to occur), and no maintenance (the equipment is in good condition and does not require maintenance at the moment). Maintenance type recommendations can help operation and maintenance personnel quickly determine the maintenance strategy for the equipment and avoid over-maintenance or under-maintenance.

[0087] (2) Maintenance content recommendations: Based on the defect type, defect severity, equipment structural characteristics, and maintenance procedures and standards, the system intelligently recommends detailed maintenance content, such as replacing insulators, replacing lightning arresters, tightening bolts, replacing gaskets, cleaning insulating components, changing lubricating oil, adjusting mechanical gaps, performing partial discharge tests, and performing withstand voltage tests. Maintenance content recommendations can guide maintenance personnel to clarify maintenance tasks and operating procedures, thereby improving maintenance efficiency and quality.

[0088] (3) Maintenance time recommendations: Based on the equipment failure risk prediction results, equipment operating load, grid operation mode, and maintenance resource availability, the optimal maintenance time is intelligently recommended. For example, it is recommended to perform power outage maintenance during low load periods, or to complete maintenance within the next XX days to avoid equipment failures from expanding or affecting the reliable power supply of the grid. Maintenance time recommendations can help operation and maintenance personnel arrange maintenance plans reasonably, reduce power outage time, and minimize the impact on grid operation.

[0089] (4) Spare parts recommendations: Based on the maintenance content recommendations and equipment ledger data, the system intelligently recommends the required spare parts list and quantity. Spare parts recommendations include spare part name, model, specification, quantity, inventory information, supplier information, etc. Spare parts recommendations can help operation and maintenance personnel prepare the required spare parts in advance, shorten maintenance time, and improve maintenance efficiency.

[0090] 2. 3D visualization and interaction The operation and maintenance decision support module integrates the health status assessment results, fault risk prediction results, operation and maintenance recommendations of power grid equipment with the three-dimensional model of the power grid to achieve three-dimensional visualization and interactive query of the health status of equipment, providing more intuitive and convenient decision support and operation experience for operation and maintenance personnel. The three-dimensional visualization and interactive functions mainly include: (1) Three-dimensional visualization of equipment health status: The power grid equipment health score and multi-dimensional health indicators output by the intelligent analysis and prediction module are superimposed on the three-dimensional model of the power grid using visual elements such as color, texture, and animation to intuitively display the overall health status of the equipment and the health status of each component. For example, different colors are used to represent different health levels (green for health, yellow for minor defects, and red for serious defects), heat maps are used to display the temperature distribution and thermal anomaly areas of the equipment, and arrows are used to indicate the degradation trend of the equipment. The three-dimensional visualization of the equipment health status can help operation and maintenance personnel understand the health status of power grid equipment more intuitively and quickly, and improve the efficiency of status perception.

[0091] (2) 3D visualization of fault risk: The fault risk prediction results of power grid equipment output by the intelligent analysis and prediction module are superimposed on the 3D model of the power grid using visualization elements such as heat maps, risk area highlighting, and animation simulation to intuitively display information such as the equipment's fault risk level, high-risk areas, and fault development trends. For example, different colors are used to represent different fault risk levels (green represents low risk, yellow represents medium risk, and red represents high risk), flashing animations are used to indicate high-risk areas, and arrows are used to indicate the direction of fault spread. The 3D visualization of fault risk can help operation and maintenance personnel identify high-risk equipment and areas more intuitively and promptly, take preventive measures in advance, and reduce the risk of power grid failures.

[0092] (3) 3D visualization of operation and maintenance suggestions: The intelligent maintenance suggestions generated by the operation and maintenance decision support module are superimposed on the 3D model of the power grid using visual elements such as text labels, indicator arrows, and animation demonstrations to intuitively display the specific content and operation steps of the maintenance suggestions. For example, the equipment parts recommended for maintenance are marked with text labels on the 3D model, the bolts that need to be tightened are indicated with arrows, and the operation process of replacing insulators is demonstrated with animations. The 3D visualization of operation and maintenance suggestions can help operation and maintenance personnel understand the operation and maintenance suggestions more intuitively and clearly, guide on-site operation and maintenance operations, and improve operation and maintenance efficiency and quality.

[0093] (4) Interactive query of 3D models: Users can perform interactive operations on the 3D visualization interface, such as clicking, dragging, zooming, rotating, etc., to freely browse the 3D model of the power grid and view the status of equipment from different angles and details. Users can click on any equipment component in the 3D model to quickly query the detailed health report, historical inspection data, fault risk prediction results, intelligent maintenance recommendations, and other information of the equipment. The interactive query function of the 3D model can help operation and maintenance personnel understand the health status and operation and maintenance knowledge of power grid equipment more conveniently and in-depth, thereby improving the efficiency of operation and maintenance decision-making and user experience.

[0094] Example 7 This embodiment further elaborates on the visualization module, building on the previous one. As the platform's human-computer interaction window, the visualization module is responsible for user interface design and information visualization, providing an intuitive, clear, and convenient information display and interactive operation interface for power grid managers and maintenance personnel. The visualization module primarily consists of two core components: a large-screen data display and an operation workbench interface.

[0095] 1. Data large screen display The visualization module is designed with a professional data large-screen display interface for real-time visualization of key information such as the overall health status of power grid equipment, health indicators of key equipment, early warning information of high-risk equipment, progress of operation and maintenance tasks, and system operation status, providing global monitoring and decision-making support for power grid managers. The data large-screen display mainly includes: (1) Overview of power grid equipment health: The health status of the overall power grid equipment is displayed in real time in the form of charts, numbers, etc., such as the number of healthy equipment, the number of equipment with minor defects, the number of equipment with moderate defects, the number of equipment with serious defects, the average health score of equipment, and other statistical indicators. Different colors are used to distinguish the health levels of equipment, intuitively showing the health status of the overall power grid equipment.

[0096] (2) Key Equipment Health Dashboard: Design a special health dashboard for key equipment critical to grid operation (such as main transformers, important transmission lines, and key switchgear) to centrally display detailed information such as the real-time health score, multi-dimensional health indicators, historical health trend curves, potential fault risk levels, and the latest maintenance recommendations of these key equipment, so that managers can focus on the operating status of key equipment and keep abreast of the health risks of the equipment.

[0097] (3) High-risk equipment warning list: A real-time scrolling display of the high-risk equipment warning information list detected by the current platform. The list contains key information such as the warning equipment name, warning level, defect type, warning time, and disposal status. Different colors are used to distinguish the warning levels, and flashing animations are used to prompt emergency warning information to ensure that managers can obtain high-risk equipment warning information in the first place and make timely decisions and command and dispatch.

[0098] (4) Operation and maintenance task execution monitoring: Real-time display of the execution progress and status of the currently executing operation and maintenance tasks, such as the inspection mileage, inspection completion rate, remaining inspection time of the inspection tasks, the maintenance progress, completed maintenance projects, remaining maintenance projects, etc. of the maintenance tasks, so that management personnel can grasp the overall execution status of the operation and maintenance tasks in real time and perform task scheduling and resource coordination.

[0099] (5) System operation status monitoring: Real-time display of the platform's own operation status information, such as the operation status of the knowledge federation management system, the operation status of the DeepSeek-R1 large model, data access status, data processing performance, system resource usage, network connection status, etc., to facilitate system administrators to monitor the platform's operation status in real time and promptly discover and handle system operation anomalies.

[0100] (6) Data reports and statistical charts: The data screen can also flexibly switch to display various data reports and statistical charts, such as equipment health distribution charts, defect type statistical charts, fault risk level distribution charts, operation and maintenance task statistical reports, operation and maintenance efficiency statistical reports, etc., to visualize and analyze the data of power grid equipment health management and intelligent operation and maintenance from different dimensions and granularity, providing deeper data insights for management decisions.

[0101] 2. Operation workbench interface The visualization module features a professional workbench interface designed for power grid operators, serving as their primary platform for daily work. The workbench interface is designed to be user-friendly, convenient, and efficient, with a focus on intuitive information presentation, rational functional layout, and easy operation. The workbench interface primarily includes the following functional areas: (1) Equipment health overview panel: In the main panel area of ​​the operation workbench interface, the health overview information of the power grid equipment that the current user is concerned about is displayed, such as the equipment list, equipment health score, equipment health level color code, key health indicators, the latest warning information, etc., which makes it easy for operation and maintenance personnel to quickly understand the overall health status of the equipment and quickly locate the equipment that needs special attention. The equipment list supports multi-dimensional sorting and filtering, such as sorting by health, filtering by equipment type, filtering by warning level, etc., which makes it easy for users to quickly find the target equipment.

[0102] (2) Device health details view: When the user selects a device in the device health overview panel, the operation workbench interface will automatically switch to the device health details view, which fully displays the detailed health information of the device. The details view includes: 1) Basic equipment information: displays basic information such as equipment name, model, specifications, location, line or substation, and commissioning time.

[0103] 2) Real-time equipment health score and multi-dimensional health indicators: The overall health score of the equipment and the real-time evaluation results of multi-dimensional health indicators such as insulation status, mechanical status, electrical performance, and adaptability to the operating environment are intuitively displayed in the form of dashboards and numbers.

[0104] 3) Equipment historical health trend curve: Visually display the equipment's health score and historical trends of key health indicators in the form of line charts and trend charts, making it easier for operation and maintenance personnel to understand the historical evolution and degradation trends of equipment health.

[0105] 4) Equipment failure risk prediction results: Displays information such as the equipment failure probability prediction results, failure risk level, and potential failure types within a certain period of time, helping operation and maintenance personnel to predict equipment failure risks in advance.

[0106] 5) Equipment inspection data: Integrates historical equipment inspection data, such as drone inspection images, infrared thermal images, ultraviolet images, lidar point cloud data, gas sensor data, etc., to facilitate operation and maintenance personnel to view and analyze equipment inspection data and verify equipment status assessment results.

[0107] 6) Equipment Maintenance Recommendations: Displays intelligent maintenance recommendations automatically generated by the platform, including detailed information such as maintenance type, maintenance content, maintenance time, and spare parts, to assist operation and maintenance personnel in formulating scientific maintenance plans.

[0108] 7) Equipment Knowledge Graph: Visually displays the knowledge relationships of equipment in the power grid operation and maintenance knowledge graph, such as upstream and downstream equipment, common fault types of equipment, equipment maintenance history, and relevant operation and maintenance procedures and standards of equipment, to help operation and maintenance personnel understand the relevant knowledge of equipment more comprehensively and deeply.

[0109] (3) Conversational query interface for 10,000-level equipment health: A natural language interactive entry for "conversational query for 10,000-level equipment health" is provided in a prominent position on the operating workbench interface. Users can enter natural language query questions in the text box (for example, "Query the health status of all 110kV transformers", "View the infrared image of the most recent inspection of tower #3", "Analyze the overall failure risk of XX substation equipment") and click the "Query" button. The platform can then use the natural language understanding and knowledge retrieval capabilities of the DeepSeek-R1 large model to quickly return accurate and concise query results and present them to the user in natural language. The conversational query interface is designed to be simple and intuitive, and the operation is convenient and efficient. Users can easily obtain the required information without having to learn complex operating procedures and command syntax.

[0110] (4) Task Management and Control Panel: The task management and system control panel are provided in the sidebar or toolbar area of ​​the operation workbench interface. The task management panel is used to create, edit, schedule, and monitor inspection and maintenance tasks. The control panel is used to perform system management operations such as system parameter configuration, user rights management, data export, and report generation. The control panel operation buttons and menus are simple and intuitive, making it convenient for operation and maintenance personnel to perform various system operations.

[0111] Through the above detailed description of the technical solution and the visual display in combination with the accompanying drawings, the technical principles, system architecture, core functions, key technical details and specific implementation methods of the present invention are comprehensively and clearly presented, enabling those skilled in the art to fully understand and implement the present invention.

[0112] The above is a detailed description of the embodiments of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without departing from the spirit of the present invention. These equivalents or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A power grid equipment health management and operation and maintenance platform based on knowledge federation and language model, characterized by: It includes multi-source data fusion module, knowledge federation management system, large language model, intelligent analysis and prediction module, operation and maintenance decision support module and visualization display module; The multi-source data fusion module is respectively connected to the knowledge federation management system and the large language model for accessing and integrating the multi-source heterogeneous operation and maintenance data of the power grid; The knowledge federation management system is respectively connected to the large language model and the intelligent analysis and prediction module, and utilizes the knowledge federation technology to integrate the multi-source heterogeneous operation and maintenance data of the power grid and construct a knowledge graph of power grid operation and maintenance; The large language model is communicated with the intelligent analysis and prediction module and the visualization display module respectively, and uses its cognitive intelligence and natural language interaction capabilities to implement the core AI functions of power grid equipment health management and intelligent operation and maintenance; The intelligent analysis and prediction module is respectively connected to the operation and maintenance decision support module and the visualization display module, and uses a large language model and a power grid operation and maintenance knowledge graph to perform intelligent analysis and prediction of power grid equipment; The operation and maintenance decision support module is in communication with the visualization display module, and generates an operation and maintenance decision plan for the power grid equipment based on the intelligent analysis and prediction results of the intelligent analysis and prediction module, combined with the power grid operation and maintenance business needs and expert knowledge; The visualization display module is used in the human-computer interaction window of the platform to visually display the conversational query results of the large language model, the analysis and prediction results of the intelligent analysis and prediction module, and the operation and maintenance decision-making plan of the operation and maintenance decision support module.

2. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The multi-source data fusion module is the data entry point of the platform and is located in the data access layer of the platform architecture. It is responsible for collecting data from different sources and different types of power grid multi-source heterogeneous operation and maintenance data sources, and performing data cleaning, preprocessing, conversion, and fusion operations. It converts the original, heterogeneous power grid multi-source heterogeneous operation and maintenance data into a unified format, high-quality multimodal data input, and provides standard and standardized data input for the knowledge federation management system and large language model. The knowledge federation management system is the data foundation and knowledge engine of the platform. Located at the bottom layer of the platform architecture, it is responsible for the access, integration, management, and secure sharing of multi-source heterogeneous power grid operation and maintenance data, as well as the construction, storage, and maintenance of the power grid operation and maintenance knowledge graph. The knowledge federation management system provides data and knowledge support for large language models and intelligent analysis and prediction modules; The large language model is the platform's core intelligent engine, located in the middle layer of the platform architecture. It is responsible for implementing the core AI functions of power grid equipment health management and intelligent operation and maintenance. The large language model receives multimodal data input after fusion and processing from the multi-source data fusion module, and uses its own cognitive intelligence and knowledge reasoning capabilities to perform health status assessment of power grid equipment, fault risk prediction and warning, and intelligent operation and maintenance strategy generation. The large language model also directly provides an intelligent human-computer interaction interface for conversational query of power grid equipment health to the visualization display module. The intelligent analysis and prediction module is the intelligent analysis core of the platform. Located in the intelligent analysis layer of the platform architecture, it is responsible for invoking the large language model and the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system to perform intelligent analysis and prediction of power grid equipment. The intelligent analysis and prediction module outputs the analysis and prediction results to the operation and maintenance decision support module and the visualization display module for subsequent operation and maintenance decision support and information display. The operation and maintenance decision support module is the intelligent decision center of the platform, located in the intelligent decision layer of the platform architecture. Based on the output results of the intelligent analysis and prediction module, combined with the preset operation and maintenance strategy model and expert knowledge, it automatically generates an operation and maintenance decision plan. The operation and maintenance decision support module outputs the intelligent operation and maintenance decision plan to the visualization display module for display. The visualization module is the human-computer interaction window of the platform, located in the user interaction layer of the platform architecture, responsible for the user interface design and information visualization display of the platform; The visualization display module receives analysis results and decision solutions from the intelligent analysis and prediction module and the operation and maintenance decision support module, as well as conversational query results from the large language model, and displays key information about power grid equipment in various forms. The visualization display module provides a human-computer interaction interface, allowing users to conduct conversational queries and intelligent interactive operations on the health of power grid equipment through natural language.

3. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The knowledge federation management system adopts a hybrid federated learning architecture, including data provider nodes, federated learning servers, model aggregation modules and knowledge graph construction modules; The data provider nodes represent different business departments of the power grid or data sources. Each data provider node has local operation and maintenance data and computing resources, participates in local model training, and uploads model parameter updates to the federated learning server; The federated learning server, as the central coordination node of the knowledge federation, is responsible for task scheduling and model aggregation of federated learning; It has a model aggregation module and a knowledge graph construction module deployed internally; The model aggregation module is deployed on the federated learning server and is used to receive model parameter updates uploaded by each data provider node and aggregate the model parameters according to the federated aggregation algorithm to generate a globally shared federated learning model. The knowledge graph construction module is deployed on the federated learning server side, and is used to extract knowledge fragments from multi-source heterogeneous operation and maintenance data and integrate them into a globally unified power grid operation and maintenance knowledge graph.

4. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 3 is characterized in that: In the knowledge federation management system: There are multiple data provider nodes, representing different business departments or data sources of the power grid. Each data provider node has its own local operation and maintenance data and computing resources. The data is stored locally and not uploaded to the central server. The data provider node is responsible for cleaning, preprocessing, and feature engineering of local data, participating in federated learning model training, and uploading local model parameter updates to the federated learning server; The federated learning server, as the central coordination node of the knowledge federation, is responsible for task scheduling and model aggregation of federated learning; The federated learning server does not access the original data. Its model aggregation module receives model parameter updates uploaded by each data provider node and aggregates the model parameters using a federated aggregation algorithm to generate a globally shared federated learning model. The knowledge graph construction module of the federated learning server is responsible for scheduling knowledge graph construction tasks and knowledge fusion, integrating the local knowledge fragments contributed by each data provider node into a globally unified power grid operation and maintenance knowledge graph. The model aggregation module is deployed on the federated learning server and is responsible for aggregating the federated learning model parameters; The model aggregation module receives model parameter updates uploaded by each data provider node, aggregates the model parameters according to the federated aggregation algorithm, and generates a globally shared federated learning model; The model aggregation module is also responsible for monitoring and managing the federated learning process; The knowledge graph construction module is deployed on the federated learning server and is responsible for the construction and management of the power grid operation and maintenance knowledge graph; The knowledge graph construction module extracts knowledge from the multi-source heterogeneous operation and maintenance data of the power grid integrated by the knowledge federation, constructs local knowledge fragments, and integrates the local knowledge fragments contributed by each data provider node into a global unified power grid operation and maintenance knowledge graph; the knowledge graph construction module is also responsible for the storage, update and maintenance of the knowledge graph, as well as providing query and access interfaces for the knowledge graph.

5. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The multi-source heterogeneous power grid operation and maintenance data integrated by the knowledge federation management system includes: Drone inspection data: Multimodal inspection data collected by the drone intelligent inspection system, including visible light images, infrared thermal images, ultraviolet images, lidar point cloud data, gas sensor data, defect recognition results, early warning information, and inspection reports; SCADA system operation data: Real-time equipment operation data collected from the power grid SCADA system, including operating parameters and status quantities, alarm information, and event records. The operating parameters and status quantities include voltage, current, active power, reactive power, frequency, temperature, pressure, oil level, and SF6 gas pressure; PMS equipment ledger data: Equipment ledger data from the power PMS equipment management system, including basic equipment information, equipment parameter information, equipment location information, historical equipment maintenance records, equipment defect records, and spare parts information. Basic equipment information includes equipment name, model, specifications, manufacturer, production date, and installation date. Equipment parameter information includes rated voltage, rated current, rated capacity, and insulation level. Equipment location information includes the substation to which it belongs, the line to which it belongs, and geographic coordinates. GIS geographic information data: grid geographic information data from the power GIS geographic information system, including line directions, tower locations, substation distribution, geographic environment information, and meteorological environment information; Historical defect data: Historical defect data of power grid equipment from historical manual inspection records and defect management systems, including defect type, defect level, defect location, defect description, defect occurrence time, and defect handling records; Expert knowledge base: The accumulated professional knowledge and experience of experts in the power grid field, including knowledge of power grid equipment structure and principles, operation and maintenance knowledge, fault diagnosis knowledge, maintenance procedures and standards, and expert experience rules; Operation and maintenance procedures and standards: procedures, standards, specifications, and guidelines related to grid equipment operation and maintenance issued by State Grid Corporation of China, industry associations, and standardization organizations.

6. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The large language model is the DeepSeek-R1 large model, and the model architecture of the DeepSeek-R1 large model includes: Transformer Encoder-Decoder Architecture: This model uses the Transformer Encoder-Decoder architecture as its core network structure. The Transformer Encoder is responsible for deep feature extraction and knowledge representation of the input multimodal operation and maintenance data. The Transformer Decoder generates intelligent analysis results and operation and maintenance decision recommendations based on the features and knowledge extracted by the Encoder. Multimodal input fusion mechanism: The model's input layer inputs power grid operation and maintenance data in multiple modalities, including text, image, time series, and structured data. A multimodal feature fusion module is designed within the model to fuse and align data features from different modalities and learn the associations and complementary information between multimodal data. Knowledge Graph Enhancement Module: This module is integrated within the model to incorporate the grid operation and maintenance knowledge graph constructed by the knowledge federation management system into the model's training and reasoning processes. The module encodes the knowledge information in the knowledge graph into knowledge vector representations and fuses these knowledge vectors with multimodal data features. Lightweight and efficient deployment optimization: The model adopts a lightweight Transformer variant structure with fewer parameters and uses model compression technologies including model pruning, model quantization, and knowledge distillation. The model also optimizes the hardware platform for inference acceleration.

7. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The large language model training process includes a pre-training phase and a fine-tuning phase, wherein: Pre-training phase: The large language model is first pre-trained on a large-scale general corpus to learn general language knowledge and world knowledge. The pre-training corpus includes massive amounts of internet text data, encyclopedia knowledge, and books and documents. The pre-training phase uses self-supervised learning methods. Fine-tuning phase: Based on the pre-training phase, domain knowledge is fine-tuned using the multi-source heterogeneous grid operation and maintenance data integrated by the knowledge federation management system. This allows the large language model to more effectively adapt to the specific tasks and application scenarios of grid equipment health management and intelligent operation and maintenance. Among them, the fine-tuning stage in the large language model training process adopts a multi-task joint training strategy, unifying multiple operation and maintenance tasks into the framework of the large language model for joint training. The multiple operation and maintenance tasks include health status assessment of power grid equipment, fault risk prediction and early warning, intelligent operation and maintenance strategy generation, and health status conversational query; the fine-tuning strategy is adopted in the model fine-tuning process. Based on the pre-trained model, only a small number of model parameters are fine-tuned or a small number of adaptation layers for specific tasks are added.

8. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 7 is characterized in that: The fine-tuning phase of the large language model uses a supervised learning method for model training. The training data includes: Power grid equipment health status assessment dataset: This dataset includes multimodal inspection data of power grid equipment and corresponding equipment health status labels. By learning this data, the model understands the mapping relationship between different modal inspection data and equipment health status. Power grid equipment failure risk prediction dataset: This dataset contains multi-dimensional data including historical operation data, historical defect data, and environmental data of power grid equipment, along with corresponding failure occurrence labels. By learning from this data, the model understands the correlation between different dimensional data and equipment failure risks. Power grid operation and maintenance knowledge graph: The power grid operation and maintenance knowledge graph constructed by the knowledge federation management system is used as auxiliary knowledge information to enhance the model's knowledge representation and reasoning capabilities. During model fine-tuning, the knowledge information in the knowledge graph is integrated into the model training process, including encoding the knowledge information into knowledge vectors through knowledge graph embedding technology and fusing the knowledge vectors with input data features. Conversational query and response dataset: This dataset is constructed for the "Conversational Query on Power Grid Equipment Health" task. It contains various natural language queries about the health status of power grid equipment posed by users, along with the natural language responses that the platform should return. By learning from this data, the model understands the user's query intent and learns how to retrieve and extract relevant knowledge from the knowledge graph to generate conversational responses.

9. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1 is characterized in that: The operation and maintenance decision support module automatically generates intelligent maintenance recommendations for power grid equipment based on the intelligent analysis and prediction results output by the intelligent analysis and prediction module, combined with the operation and maintenance procedures and standards and expert experience knowledge in the power grid operation and maintenance knowledge graph constructed by the knowledge federation management system; The maintenance recommendation generation process takes into account the actual health status of the equipment, the failure risk level, the operation and maintenance cost constraints, and the safety and reliability requirements; The maintenance recommendations include: Maintenance type recommendation: Based on the equipment health status and fault risk level, the most appropriate maintenance type is intelligently recommended, including condition-based maintenance, regular maintenance, emergency repair, and no maintenance. Maintenance content recommendations: Based on the defect type, defect severity, equipment structural characteristics, and maintenance procedures and standards, intelligently recommend detailed maintenance content, including: replacing insulators, replacing lightning arresters, tightening bolts, replacing gaskets, cleaning insulation components, changing lubricating oil, adjusting mechanical clearances, partial discharge testing, and withstand voltage testing; Maintenance time recommendations: Based on the equipment's failure risk prediction results, equipment operating load, grid operation mode, and maintenance resource availability, the system intelligently recommends the optimal maintenance time. This includes recommending power outages during low-load periods or completing maintenance within a certain timeframe in the future. Spare parts recommendations: Based on maintenance content recommendations and equipment ledger data, intelligently recommend the required spare parts list and quantity; spare parts recommendations include spare part name, model, specification, quantity, inventory information, and supplier information; The operation and maintenance decision support module integrates the health status assessment results, fault risk prediction results, and operation and maintenance recommendation information of power grid equipment with the three-dimensional power grid model to provide a three-dimensional visualization and interactive query of the equipment health status. The three-dimensional visualization and interaction include: 3D visualization of equipment health status: The power grid equipment health scores and multi-dimensional health indicators output by the intelligent analysis and prediction module are superimposed on the 3D power grid model using color, texture, and animation visualization elements to intuitively display the overall health status of the equipment and the health status of each component. This includes, for example, using different colors to represent different health levels, using heat maps to display the temperature distribution and thermal anomaly areas of the equipment, and using arrows to indicate equipment degradation trends. 3D Fault Risk Visualization: The power grid equipment fault risk prediction results output by the intelligent analysis and prediction module are superimposed on the 3D power grid model using heat maps, risk area highlighting, and animated simulation visualization elements. This intuitively displays the equipment's fault risk level, high-risk areas, and fault development trend information. This includes using different colors to represent different fault risk levels, flashing animations to indicate high-risk areas, and arrows to indicate the direction of fault spread. 3D visualization of maintenance recommendations: The intelligent maintenance recommendations generated by the maintenance decision support module are superimposed on the 3D grid model using text labels, arrows, and animation visualization elements to intuitively display the specific content and operation steps of the maintenance recommendations. This includes: labeling equipment components recommended for maintenance with text labels on the 3D model, indicating bolts that need to be tightened with arrows, and using animation to illustrate the operation process of replacing insulators. Interactive query of 3D models: Users can interactively operate on the 3D visualization interface, freely browse the 3D model of the power grid, and view the equipment status from different angles and details. Users can click on equipment components in the 3D model to quickly query the equipment's detailed health report, historical inspection data, fault risk prediction results, and intelligent maintenance recommendation information.

10. The power grid equipment health management and operation and maintenance platform based on knowledge federation and language model according to claim 1, characterized in that: The visualization display module serves as the human-computer interaction window of the platform and is responsible for the user interface design and information visualization display of the platform. The visualization display module includes a large-screen display of data and an operation workbench interface; The data large screen display is used to visualize in real time the overall health status of power grid equipment, health indicators of key equipment, early warning information of high-risk equipment, progress of operation and maintenance tasks, and key information of system operation status; The operation workbench interface is used as the main operation platform for daily work of operation and maintenance personnel; The data large screen display includes: Grid Equipment Health Overview: Displays the health status of the entire grid equipment in real time in the form of charts and numbers, including the number of healthy devices, the number of devices with minor defects, the number of devices with moderate defects, the number of devices with serious defects, and statistical indicators of the average health score of the equipment. Different colors are used to distinguish the health level of the equipment. Key Equipment Health Dashboard: A dedicated health dashboard is designed for key equipment critical to grid operation, centrally displaying their real-time health scores, multi-dimensional health indicators, historical health trend curves, potential failure risk levels, and the latest maintenance recommendations. High-risk equipment warning list: A real-time scrolling display of the high-risk equipment warning information list detected by the current platform. The list includes key information such as the warning equipment name, warning level, defect type, warning time, and disposal status. Different colors are used to distinguish warning levels, and flashing animations are used to indicate emergency warning information. Operation and maintenance task execution monitoring: Real-time display of the execution progress and status of the currently executing operation and maintenance tasks, including inspection mileage, inspection completion rate, remaining inspection time, maintenance progress, completed maintenance items, and remaining maintenance items; System operation status monitoring: Real-time display of the platform's own operation status information, including the operation status of the knowledge federation management system, the operation status of the large language model, data access status, data processing performance, system resource usage, and network connection status; Data reports and statistical charts: The data screen flexibly switches to display various data reports and statistical charts, including equipment health distribution charts, defect type statistics charts, fault risk level distribution charts, operation and maintenance task statistics reports, and operation and maintenance efficiency statistics reports. It provides visual analysis of power grid equipment health management and intelligent operation and maintenance data from different dimensions and granularity. The operation workbench interface includes: Equipment health overview panel: The main panel area of ​​the operation workbench interface centrally displays the health overview information of the power grid equipment currently being viewed by the user, including the equipment list, equipment health score, equipment health level color code, key health indicators, and the latest warning information; the equipment list supports multi-dimensional sorting and filtering, including sorting by health level, filtering by equipment type, and filtering by warning level; Device health details view: When the user selects a device in the device health overview panel, the operation workbench interface automatically switches to the device health details view, which fully displays the detailed health information of the device; Conversational query interface for power grid equipment health: A natural language interactive portal for "conversational query for power grid equipment health" is prominently located on the operation workbench interface. Users can enter a natural language query in a text box and click the "Query" button. The platform leverages the natural language understanding and knowledge retrieval capabilities of the large language model to quickly return accurate and concise query results, presenting them to the user in natural language. Task management and control panel: The sidebar or toolbar area of ​​the operation workbench interface provides task management and system control panel. The task management panel is used to create, edit, schedule, and monitor inspection and maintenance tasks. The control panel is used to perform system parameter configuration, user permission management, data export, and report generation system management operations. The detailed information view includes: Basic equipment information: displays basic information such as equipment name, model, specifications, location, line or substation to which it belongs, and commissioning time; Real-time equipment health score and multi-dimensional health indicators: The dashboard and digital form intuitively display the overall health score of the equipment and the real-time evaluation results of the multi-dimensional health indicators of insulation status, mechanical status, electrical performance, and operating environment adaptability; Device historical health trend curve: Visually display the device's health score and historical change trends of key health indicators in the form of line charts and trend charts; Equipment failure risk prediction results: Displays the equipment failure probability prediction results, failure risk level, and potential failure type information within a certain period of time; Equipment inspection data: Integrates historical equipment inspection data, including drone inspection images, infrared thermal images, ultraviolet images, lidar point cloud data, and gas sensor data; Equipment maintenance suggestions: Displays intelligent maintenance suggestions automatically generated by the platform, including detailed information on maintenance type, maintenance content, maintenance time, and spare parts recommendations; Equipment knowledge graph: Visually displays the knowledge association relationship of equipment in the power grid operation and maintenance knowledge graph, including upstream and downstream equipment, common fault types of equipment, maintenance history of equipment, and relevant operation and maintenance procedures and standards of equipment.

Images