Intelligent power log analysis system based on knowledge graph and large language model

Abstract

The invention relates to a power log intelligent analysis system based on a knowledge graph and a large language model. The power log intelligent analysis system comprises a data collection and preprocessing module, a knowledge graph construction and management module, a context understanding and correlation analysis module, a personalized report generation module and a multi-modal data integration module. According to the method, information such as equipment, events and operations in the power logs is structured through the knowledge graph and is associated with domain knowledge, so that deep understanding and association analysis of log contents are realized, and the problem of insufficient context understanding in a traditional method is solved. By utilizing the strong generation capability of a large language model and based on context information provided by a knowledge graph, a personalized report meeting roles and requirements of a user can be automatically generated, and the user experience and practicability of log analysis are improved; log data in various formats are integrated into the knowledge graph, a comprehensive summary report containing information such as charts and timelines is generated through a large language model, and the problem that heterogeneous data are difficult to integrate is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and more specifically, relates to an intelligent analysis system for power logs based on a knowledge graph and a large language model. Background Art

[0002] In the power industry, power log data is important information that records the operating status of the power system, equipment operation conditions, fault events, etc. These log data play a vital role in the daily maintenance, fault diagnosis and operation optimization of the power system. However, with the continuous expansion of the scale and complexity of the power system, the volume and complexity of log data are also increasing dramatically. These log data often have the following characteristics: Huge data volume: Log data in modern power systems often presents large-scale and high-frequency characteristics. Millions of logs may be generated every day, involving numerous devices, operations, and events. Traditional manual analysis methods have been unable to cope with such a huge amount of data.

[0003] Diverse data formats: Power log data includes not only plain text information, but also charts, sensor data, and other formats. The integration and analysis of these heterogeneous data are difficult and often require a combination of multiple tools and technologies for processing.

[0004] Information dispersion: Log data is usually stored in a dispersed and unstructured form, lacking effective association and context information. This information dispersion often requires operators to spend a lot of time to find and organize relevant information when performing fault diagnosis and system optimization.

[0005] High professionalism: The power system log contains a large number of professional terms and equipment names, which are difficult for non-professionals to understand, resulting in a high threshold for interpreting and analyzing the log information.

[0006] Under the current technical background, although some solutions have been developed to conduct preliminary analysis and summary of power log data, there are still obvious shortcomings: Limited automation: Existing power log analysis systems mostly use rule-driven methods, relying on preset rules and templates, and are difficult to adapt to the dynamically changing power system environment. When encountering new equipment or fault types, these systems often require manual intervention and rule updates, resulting in low analysis efficiency.

[0007] Insufficient contextual understanding: Traditional log analysis methods can usually only analyze a single or a small number of logs independently, and lack the ability to understand and process the contextual relationship between logs. This limitation leads to insufficient accuracy and comprehensiveness of log analysis, making it difficult to effectively identify complex system problems.

[0008] Lack of personalized support: Existing log analysis systems usually output reports in a fixed format, which makes it difficult to generate personalized analysis results based on the needs of different users. This makes it difficult for users at different levels to obtain information, often facing the problem of redundant or insufficient information, which affects decision-making efficiency.

[0009] In summary, the existing technology has problems such as low efficiency due to the limited degree of automation of power log data, insufficient accuracy and comprehensiveness of log analysis due to insufficient context understanding, and redundancy or insufficiency of data at different levels due to the lack of personalized report generation. Summary of the invention

[0010] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of the present invention to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0011] Therefore, the purpose of the present invention is to provide an intelligent analysis system for power logs based on knowledge graphs and large language models, which can solve the problems in the prior art of inefficiency caused by the limited degree of automation of power log data, insufficient accuracy and comprehensiveness of log analysis caused by insufficient context understanding, and redundancy or insufficiency of data at different levels caused by the lack of personalized report generation.

[0012] In order to solve the above technical problems, the present invention provides an intelligent analysis system for power logs based on knowledge graphs and large language models, including a data collection and preprocessing module, a knowledge graph construction and management module, a context understanding and association analysis module, a personalized report generation module and a multimodal data integration module; the data collection and preprocessing module is used to collect raw data from various log sources in the power system, clean, denoise and structure the data, pass the standardized data to the context understanding and association analysis module, and pass the cleaned data to the knowledge graph construction and management module; the knowledge graph construction and management module is used to map the key entities and relationships of the power system to the knowledge graph to form a structured power field knowledge base, and the fields in the knowledge graph are mapped to the knowledge graph. Domain knowledge and context understanding interact with the association analysis module; context understanding and association analysis module: used to use the large language model and knowledge graph to perform semantic analysis and context understanding of the log content on the standardized data processed by the data collection and preprocessing module, and use the relationships and rules in the knowledge graph to perform association analysis on the log data, extract key events and potential problems, and send the results to the personalized report generation module; personalized report generation module: used to call the large language model to generate personalized log analysis reports based on the user's role and needs, and send the personalized reports to the user; multimodal data integration module: used to integrate different types of data into a unified knowledge graph, and transmit multimodal information to the context understanding and association analysis module.

[0013] Furthermore, the data collection and preprocessing module includes a log data collection module, a data cleaning and denoising module, a data format standardization module and a time series processing module; the log data collection module is used to receive data from different equipment logs, operation logs, fault logs and sensor data; the data cleaning and denoising module is used to clean records containing noise, redundant information and incomplete records, including removing redundant data, filling missing data, and processing outliers; the data format standardization module is used to standardize these data in different formats, including field extraction and normalization, data type conversion; the time series processing module is used to perform time series processing on log data, including event sorting and time window division.

[0014] By adopting the above technical solutions, this case was able to extract key contextual information from the original log data and build a structured knowledge graph, providing a solid foundation for subsequent intelligent analysis and personalized report generation.

[0015] Furthermore, the knowledge graph construction and management module includes a knowledge graph definition and initialization module, a log data mapping to knowledge graph module, a knowledge graph expansion and update module and an integration module with the domain knowledge base; the knowledge graph definition and initialization module: used to define a knowledge graph structure of the power field, including entities, relationships and attributes; the log data mapping to knowledge graph module: used to map the processed log data to the knowledge graph module, including entity recognition, relationship extraction and knowledge graph mapping of entities and relationships; the knowledge graph expansion and update module: used for the knowledge graph to be dynamically updated and expanded to maintain its reflection of the current power system status, including dynamic entity addition, relationship update and expansion and knowledge graph optimization; the domain knowledge base integration module: used to connect with an external knowledge base, introduce more domain knowledge, and associate knowledge with real-time log data.

[0016] By adopting the above technical solutions, the context understanding and association analysis module in this case can deeply integrate the advantages of large language models and knowledge graphs, and combine graph algorithms to efficiently parse and analyze complex power log data, thereby achieving accurate parsing of log content and accurate identification of key events. This lays a solid foundation for subsequent personalized report generation and system decision support.

[0017] Furthermore, the context understanding and association analysis module includes a semantic parsing module for log data, a context information extraction module driven by a knowledge graph, and an association analysis and problem detection module for log information; the semantic parsing module for log data includes text parsing and information extraction, semantic understanding and context modeling, and semantic enhancement driven by a knowledge graph; the context information extraction module driven by a knowledge graph includes extraction of associated entities and events, causal reasoning combining a large language model with a graph algorithm, and abnormal event detection; the association analysis and problem detection module for log information includes multi-log source information fusion, key event annotation and priority sorting, and problem analysis and suggestion generation.

[0018] By adopting the above technical solutions, the context understanding and association analysis module in this case can deeply integrate the advantages of large language models and knowledge graphs, and combine graph algorithms to efficiently parse and analyze complex power log data, thereby achieving accurate parsing of log content and accurate identification of key events. This lays a solid foundation for subsequent personalized report generation and system decision support.

[0019] Furthermore, the personalized report generation module includes a user demand and role identification module, a large language model personalized generation mechanism module based on SystemPrompt, a report content customization module based on user demand and a multi-format output module for log analysis reports; User demand and role identification module: including user role classification, demand identification and preference setting; Large language model personalized generation mechanism module based on System Prompt: including customized settings of System Prompt, context-aware generation, diversified information presentation and multi-language and professional terminology support; Report content customization module based on user demand: including dynamic content customization, hierarchical content presentation and report structure adaptation.

[0020] By adopting the above technical solutions, the personalized log analysis and report generation module in this case can generate personalized analysis reports that meet the needs of different users by combining knowledge graphs, large language models, and customized System Prompt. This not only improves the efficiency of power log analysis and user experience, but also enhances the accuracy and practicality of the report, providing strong technical support for the efficient operation and decision support of the power system.

[0021] Furthermore, the multimodal data integration module includes a multimodal data fusion and association module, a knowledge graph application module in multimodal data analysis and a comprehensive summary report generation module; the multimodal data fusion and association module: including multimodal data collection and preprocessing, knowledge graph mapping of multimodal data and knowledge enhancement of data fusion; the knowledge graph application module in multimodal data analysis: including cross-modal information integration of knowledge graphs, collaborative analysis of large language models and graph algorithms, and anomaly detection and event linkage analysis; the comprehensive summary report generation module: including comprehensive analysis of multimodal data, multi-format presentation of cross-modal results and customized report output and distribution.

[0022] By adopting the above technical solutions, the multimodal data integration and comprehensive analysis module in this case can efficiently integrate and analyze different types of power log data, providing a more comprehensive and in-depth system status analysis. This not only enhances the breadth and depth of log analysis, but also improves the early detection and response capabilities for complex power system problems, providing strong technical support for the stable operation of the power system.

[0023] In summary, the beneficial effects of the present invention are: The present invention performs structured processing on log information through knowledge graphs and uses a large language model to deeply understand the context, which can effectively associate scattered log information and provide more accurate and comprehensive analysis results.

[0024] By utilizing the powerful generation capability of the large language model and based on the contextual information provided by the knowledge graph, the present invention enables the system to automatically generate personalized reports that meet the user's roles and needs, that is, to generate personalized log analysis reports based on the user's roles and preferences, so that technical personnel and management can obtain report content that better meets their needs, thereby improving decision-making efficiency.

[0025] The present invention integrates log data in various formats into the knowledge graph, and generates a comprehensive summary report containing charts, timelines and other information through a large language model, thereby improving the structuring ability of the knowledge graph and the generation ability of the large language model, effectively integrating various log information such as text, charts, sensor data, etc., and providing a comprehensive analysis report containing multiple formats, thereby improving the comprehensiveness and practicality of log analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 It is a flow chart of the overall solution of the present invention; Figure 2 Constructing a structural diagram for data preprocessing and knowledge graphing of the present invention; Figure 3 This is a structural diagram of the context understanding and association analysis module of the present invention; Figure 4 This is a structural diagram of the personalized log analysis and report generation module of the present invention. Implementation

[0028] The following is combined with Figure 1-4 The present invention is further described in detail with specific embodiments.

[0029] Example 1, reference Figure 1 In order to solve the problems in the prior art, such as the low efficiency caused by the limited automation of power log data, the lack of accuracy and comprehensiveness of log analysis caused by insufficient context understanding, and the redundancy or insufficiency of data at different levels caused by the lack of personalized report generation, the present invention discloses a power log intelligent analysis system based on knowledge graph and large language model, including a data collection and preprocessing module, a knowledge graph construction and management module, a context understanding and association analysis module, a personalized report generation module, and a multimodal data integration module.

[0030] Data collection and preprocessing module: This module is responsible for collecting raw data from various log sources in the power system (such as equipment logs, operation logs, fault logs, etc.). Since log data is diverse in form and complex in content, the preprocessing module needs to clean, denoise, and structure the data to ensure the integrity, accuracy, and availability of the original log data, open up all links of data input, and prepare standardized data for analysis.

[0031] Knowledge graph construction and management module: This module maps the key entities (such as equipment, fault types, operating procedures, etc.) and relationships of the power system into the knowledge graph to form a structured power domain knowledge base. The knowledge graph not only contains static domain knowledge, but can also be dynamically updated according to new log data to ensure the system's ability to process the latest information, so as to provide an expandable and dynamically updateable knowledge base through the structuring of power system domain knowledge, and support the contextual understanding and in-depth analysis of log information.

[0032] Contextual understanding and association analysis module: Based on the preprocessed log data, this module realizes semantic parsing and contextual understanding of log content through the combination of large language model and knowledge graph. The system uses the relationships and rules in the knowledge graph to perform association analysis on log data, extract key events and potential problems from it, and achieve the purpose of combining knowledge graph and large language model to realize intelligent parsing and association of log data, and extract information that is critical to system operation and fault diagnosis.

[0033] Personalized report generation module: This module is responsible for calling the large language model to generate personalized log analysis reports based on the user's role and needs. By working in collaboration with the context understanding module, the report content can be more in line with user needs, providing multi-level information from technical details to decision support, and achieving the purpose of generating exclusive log analysis reports based on the needs and roles of different users, helping users quickly obtain the required key information and improving decision-making efficiency.

[0034] Multimodal data integration module: Power log data includes not only text information, but also charts, sensor data and other forms. This module integrates different types of data into a unified knowledge graph to ensure that when generating reports, a comprehensive analysis containing all relevant information can be provided, so as to achieve the purpose of processing and integrating log data in multiple formats and ensure that the analysis report can fully present the system operation status and problem diagnosis results.

[0035] The design of the overall system architecture aims to achieve intelligent analysis and personalized summary of power log data through close collaboration of various modules to meet the diverse needs of different users.

[0036] Figure 2This is a schematic diagram of the implementation process of the data collection and preprocessing module and the knowledge graph construction and management module. The power log intelligent summary and analysis system preprocesses the raw data before log analysis and report generation, and builds a structured knowledge graph to support subsequent context understanding and association analysis, including log data collection module and preprocessing, and knowledge graph construction.

[0037] The log data collection module and preprocessing include log data collection module, data cleaning and denoising module, data format standardization module and time series processing module.

[0038] Log data collection module: The system first collects data from various power log sources, which may come from different equipment logs, operation logs, fault logs, and sensor data. Since these data come from a wide range of sources and in different formats, the system needs to support multiple data input methods, such as text files, databases, real-time streaming data, etc.

[0039] Data cleaning and denoising module: The collected raw log data usually contains noise, redundant information, and incomplete records. In order to ensure the high quality of the data, the system needs to clean the raw data, including: (1) Remove redundant data: such as duplicate log records or irrelevant system messages; Fill in missing data: For important log entries, try to complete them through context or reference to historical data; Outlier processing: Identify and process abnormal log records to ensure data consistency and reliability.

[0040] Data format standardization module: Power log data may contain different formats, such as structured data (such as CSV), semi-structured data (such as JSON, XML) and unstructured data (such as plain text). The system needs to standardize these data in different formats and convert them into a unified structured format for subsequent processing. This process includes: Field extraction and normalization: Extract key fields such as timestamp, device ID, event description, etc. from semi-structured and unstructured data and perform normalization.

[0041] Data type conversion: Convert log data in different formats into a unified structured format so that they can be associated in the knowledge graph.

[0042] Time series processing module: Electricity logs usually have strong time series characteristics, so the system needs to perform time series processing on log data. Specifically, it includes: (1) Event sorting: Sort log events based on timestamps to ensure that the order and causal relationships of events can be correctly identified during subsequent analysis.

[0043] (2) Time window division: Divide log data into different time windows according to analysis requirements to facilitate local analysis or long-term trend analysis.

[0044] The knowledge graph construction includes the knowledge graph definition and initialization module, the log data mapping to the knowledge graph module, the knowledge graph expansion and update module and the integration module with the domain knowledge base.

[0045] Definition and initialization module of knowledge graph: The knowledge graph is the core component of the system for context understanding and association analysis. When initializing the knowledge graph, the system can build a basic graph framework by integrating the existing knowledge base or expert experience knowledge in the power field. First, the system needs to define a knowledge graph structure in the power field, which includes the following main elements: (1) Nodes: such as equipment (generator, transformer, etc.), fault type (short circuit, overload, etc.), operation steps (start, stop, etc.), etc.

[0046] (2) Edges: The relationship between entities, such as "Device A has fault B", "Operation step C causes device A to shut down", etc.

[0047] (3) Attributes: Specific attribute information of entities and relationships, such as the model of the equipment, the severity of the fault, the execution time of the operation steps, etc.

[0048] Mapping log data to the knowledge graph module: After completing data preprocessing, the system needs to map the processed log data to the knowledge graph module. This process includes: (1) Entity recognition: Identify entities related to the power system from log text, such as equipment name, fault type, operation steps, etc.

[0049] (2) Relationship extraction: Based on the description in the log, identify and extract the relationship between entities, such as equipment downtime caused by a certain fault, or system anomalies caused by a certain operation step.

[0050] (3) Knowledge graph mapping of entities and relationships: Map the identified entities and relationships to the corresponding nodes and edges of the knowledge graph, and update the structure and content of the knowledge graph.

[0051] Knowledge graph expansion and update module: As new log data is continuously input, the knowledge graph needs to be dynamically updated and expanded to keep it reflecting the current power system status. This process includes: (1) Dynamic entity addition: When new equipment, fault types, or operation steps appear in the log, the system automatically adds them to the knowledge graph.

[0052] (2) Relationship update and expansion: Based on the latest log data, the system continuously updates the relationships between existing entities or adds new relationships to ensure the real-time and accuracy of the knowledge graph.

[0053] (3) Knowledge graph optimization: By analyzing the structure of the knowledge graph, redundant or incorrect relationships are identified and optimized and corrected.

[0054] Integration module with domain knowledge base: In order to enhance the accuracy and practicality of the knowledge graph, the system can integrate it with the existing power domain knowledge base. By connecting with the external knowledge base, the system can introduce more domain knowledge and associate this knowledge with real-time log data to build a more comprehensive knowledge graph.

[0055] Through the above steps, the system of the present invention can extract key contextual information from the original log data and construct a structured knowledge graph, providing a solid foundation for subsequent intelligent analysis and personalized report generation.

[0056] Figure 3 This is the structure diagram of the context understanding and association analysis module. In the power log intelligent summary and analysis system, the context understanding and association analysis module is the core part, which is responsible for deep semantic analysis of preprocessed log data, and combines knowledge graph and large language model to realize context understanding, and performs association analysis through graph algorithm to extract key information and identify key events from logs. The implementation of this module includes the semantic analysis module of log data, the context information extraction module driven by knowledge graph, and the association analysis and problem detection module of log information.

[0057] The semantic analysis module of log data includes text parsing and information extraction, semantic understanding and context modeling, and knowledge graph-driven semantic enhancement.

[0058] Text parsing and information extraction: The system first performs semantic parsing on the standardized log text. Through natural language processing (NLP) technology, the system can extract key information from the log text, such as device name, operation event, fault type, and timestamp. Specifically, it includes: (1) Word segmentation and part-of-speech tagging: The log text is segmented and the parts of speech are marked for further semantic understanding.

[0059] (2) Named Entity Recognition (NER): Use a large language model to identify entities in the log, such as equipment, operation steps, fault type, etc.

[0060] (3) Relationship extraction: Combined with the semantic understanding capabilities of large language models, the relationship between entities can be identified and extracted from the text, such as a certain device failure or a certain operation step leading to a certain result.

[0061] Semantic understanding and context modeling: After extracting key information, the system uses a large language model to further perform deep semantic understanding of the log text. The large language model can accurately capture the deep meaning of the log content based on context information and identify the causal relationship and logical association between log events. This process includes: (1) Context window division: Logs are divided into multiple context windows based on timestamps and event order to capture the temporal and logical associations between events.

[0062] (2) Semantic modeling: Use a large language model to perform semantic modeling on log events within the context window, understand the potential relationships between events, and build more accurate causal chains based on these relationships.

[0063] Knowledge graph-driven semantic enhancement: During the semantic parsing process, the system combines the knowledge graph to semantically enhance the parsing results. The structured knowledge in the knowledge graph provides additional contextual support for the fuzzy or uncertain information in the log text, including: (1) Entity alignment and verification: Align the entities identified by the large language model with the entities in the knowledge graph, verify their correctness and complete any attribute information that may be missed.

[0064] (2) Context supplementation: Use the relationships and rules in the knowledge graph to supplement the contextual information not explicitly stated in the log text, such as inferring the possible root cause of a certain fault.

[0065] The knowledge graph-driven contextual information extraction module includes the extraction of related entities and events, causal reasoning combining large language models with graph algorithms, and abnormal event detection.

[0066] Extraction of related entities and events: The system uses the associations in the knowledge graph to extract entities and events related to the current analysis target from the log. This association information can help identify key events in the log and their scope of influence, such as the impact that a device failure may have on the operating status of other devices.

[0067] Causal reasoning combining large language models and graph algorithms: The system uses large language models to perform preliminary causal analysis on log events, and then uses graph algorithms to further optimize and verify these relationships. For example, by using graph algorithms to identify paths or community structures in knowledge graphs, potential associations or fault propagation paths between devices can be discovered, and potential causal relationships in logs can be inferred.

[0068] (1) Path analysis: Use the shortest path algorithm or other graph search algorithms to identify possible paths for fault propagation.

[0069] (2) Community detection: Through the community detection algorithm, we can identify groups of entities that are closely related in the knowledge graph, thereby inferring the correlation between device groups and the fault transmission chain.

[0070] Abnormal event detection: Combining the anomaly detection capabilities of the large language model and the structured information of the knowledge graph, the system can more effectively detect abnormal events in logs. By comparing the current log event with the standard operating mode in the knowledge graph, the system can quickly identify abnormal operations or failure modes and provide corresponding explanations and handling suggestions.

[0071] The log information correlation analysis and problem detection module includes multi-log source information fusion, key event annotation and priority sorting, problem analysis and suggestion generation.

[0072] Information fusion from multiple log sources: In complex power systems, information from a single log source is usually not sufficient to fully reflect the operating status of the system. The system fuses information from different log sources through knowledge graphs to form a comprehensive view. For example, correlation analysis of equipment logs and fault logs can more accurately identify the root cause and impact of the fault. Cross-source semantic fusion of large language models: Using the contextual understanding ability of large language models, semantic information from different log sources is unified into a shared semantic space, thereby achieving effective fusion of cross-source information.

[0073] Key event labeling and prioritization: The system labels the identified key events and prioritizes them based on factors such as their impact, severity, and correlation. For example, the system can determine whether an event will cause a serious system failure based on the causal relationship in the knowledge graph, thereby providing early warning.

[0074] Problem analysis and suggestion generation: Based on the contextual understanding of log information, graph algorithm analysis, and the generation of large language models, the system can generate problem analysis reports and provide solution suggestions. For example, the system can identify the potential root cause of a certain failure and recommend what actions to take to avoid similar problems from happening again.

[0075] Through the above steps, the context understanding and association analysis module of the present invention can deeply integrate the advantages of large language models and knowledge graphs, and combine graph algorithms to efficiently parse and analyze complex power log data, thereby achieving accurate parsing of log content and accurate identification of key events. This lays a solid foundation for subsequent personalized report generation and system decision support.

[0076] Figure 4 The module structure diagram for personalized log analysis and report generation is generated by combining knowledge graph, large language model and customized System Prompt to generate personalized log analysis reports according to the user's role and needs. The system can provide tailored content for users of different levels and professional backgrounds to improve the accuracy, readability and practicality of the report, including user needs and role identification module, large language model personalized generation mechanism module based on System Prompt, report content customization module based on user needs, and multi-format output module for log analysis reports.

[0077] The user needs and role identification module includes user role classification, needs identification and preference setting.

[0078] User role classification: The system first classifies users by role, identifying their specific roles in the power system, such as technicians, management, operation and maintenance engineers, etc. Each role corresponds to different information needs and concerns, such as: (1) Technical staff: focus on detailed technical details, root causes of equipment failures, operational suggestions, etc.

[0079] (2) Management: Focus on the overall operating status of the system, key performance indicators, summaries of major events, etc.

[0080] (3) Operation and maintenance engineers: pay attention to equipment maintenance logs, operating procedures, system alarms, etc.

[0081] Demand identification and preference setting: Based on the user role identification, the system further identifies the user's specific needs and preferences, including the depth of report content, information presentation format (such as charts, text), and report frequency requirements. User preferences can be obtained in the following ways: (1) User-defined settings: Users can independently set the level of detail and focus of the report content.

[0082] (2) Historical behavior analysis: By analyzing the user’s past report reading and usage behavior, the system can intelligently adjust the report content to match the user’s preferences.

[0083] The large language model personalized generation mechanism module based on System Prompt includes customized settings of System Prompt, context-aware generation, diversified information presentation, and multi-language and professional terminology support.

[0084] Customized settings of System Prompt: When generating personalized reports, the system generates different System Prompts based on user roles and needs. System Prompt defines the generation style, tone, and use of professional terms of the large language model to ensure that the generated report content meets user expectations. For example: (1) Technical Personnel Prompt: The system may set a prompt for detailed technical instructions to instruct the model to use more precise technical terms and detailed operation steps.

[0085] (2) Management prompt: The system may set a high-level overview prompt to instruct the model to generate a concise and clear overview report that highlights key indicators and trends.

[0086] Context-aware generation: The system uses the context understanding capabilities of the large language model, combined with a customized SystemPrompt, to generate relevant analysis content based on user roles and needs. The large language model generates highly relevant text that meets user needs based on the context of the log data and the associated information of the knowledge graph. For example: (1) Technical staff: Generate a technical analysis report including root cause analysis, detailed operation steps and repair suggestions.

[0087] (2) Management: Generate an overview summary that includes a summary of key events, overall system performance, and future forecasts.

[0088] Diversified information presentation: The system supports multiple information presentation forms, including text, charts, timelines, etc., to meet the needs of different users. The large language model combined with the knowledge graph and System Prompt instructions can generate report content in multiple formats including chart descriptions, event timelines, etc. For example: (1) Chart generation: Generate fault trend charts, equipment performance indicator charts, etc. based on log data.

[0089] (2) Timeline generation: Generate a timeline of log events to show the sequence and causal relationship of events.

[0090] Multi-language and professional terminology support: The system supports customized processing of multi-language and professional terminology according to the instructions of System Prompt. For users of different languages, the large language model can generate corresponding report content according to the user's language preference. In addition, the system can adjust the use of terminology in the report according to the user's professional background to ensure a balance between professionalism and readability of the report content.

[0091] The report content customization module based on user needs includes dynamic content customization, hierarchical content presentation and report structure adaptation.

[0092] Dynamic content customization: The system dynamically customizes report content based on real-time user needs and log data. For example, when a device has a serious failure, the system can automatically generate a detailed failure analysis report for the relevant technicians and send an early warning notification.

[0093] Hierarchical content presentation: The report content is presented hierarchically according to different user roles. For example: (1) Technical details layer: including equipment operation steps, detailed fault analysis and repair suggestions.

[0094] (2) Management overview layer: includes the overall system operation status, key event summary and future forecasts.

[0095] Adaptive report structure: The system's report generation mechanism is adaptive and can adjust the report structure based on user feedback and usage habits. For example, the system can automatically adjust the order of chapters and content focus of the report based on the user's reading time and click behavior to optimize the user experience.

[0096] The multi-format output module of log analysis reports includes report format support and interactive report generation.

[0097] Report format support: The system supports multiple report output formats, including PDF, HTML, Word documents, etc., to meet the needs of different scenarios. For example, PDF format is suitable for formal reporting, and HTML format is suitable for online viewing and interaction.

[0098] Interactive report generation: For users who need in-depth analysis, the system supports the generation of interactive reports, allowing users to filter, sort and further query in the report. This form of report is particularly suitable for technicians and analysts who need to conduct in-depth data mining.

[0099] Through the above steps, the personalized log analysis and report generation module of the present invention can generate personalized analysis reports that meet the needs of different users by combining knowledge graphs, large language models, and customized System Prompt. This not only improves the efficiency of power log analysis and user experience, but also enhances the accuracy and practicality of the report, providing strong technical support for the efficient operation and decision support of the power system.

[0100] The multimodal data integration and comprehensive analysis module aims to form a unified analysis framework by integrating different types of power log data, such as text, charts, sensor data, etc., to provide more comprehensive and in-depth log analysis results. This module combines knowledge graphs, large language models, and graph algorithms to achieve a panoramic insight into the operating status of the power system through cross-modal data fusion and analysis, including a multimodal data fusion and association module, a knowledge graph application module in multimodal data analysis, and a comprehensive summary report generation module.

[0101] The multimodal data fusion and association module includes multimodal data collection and preprocessing, knowledge graph mapping of multimodal data and knowledge enhancement of data fusion.

[0102] Multimodal data collection and preprocessing: The system first collects multimodal data from different sources. After collection, these data are standardized, including format conversion, data cleaning, time alignment, etc., to ensure that all types of data can be used in the same analysis framework. Including but not limited to: (1) Text data: equipment logs, operation logs, fault logs, etc.

[0103] (2) Graphic data: such as performance indicator trend chart, temperature curve, etc.

[0104] (3) Sensor data: such as real-time data of current, voltage, and temperature sensors.

[0105] (4) Image / video data: such as device status images captured by surveillance cameras.

[0106] Knowledge graph mapping of multimodal data: The system maps the processed multimodal data into a unified knowledge graph. The specific steps include: (1) Entity and relationship recognition: Through large language models and image processing algorithms, key entities and relationships in multimodal data can be identified. For example, device status information can be extracted from sensor data, and key performance indicators can be identified from chart data.

[0107] (2) Association and alignment of data points: Align and associate data points from different modalities in the knowledge graph, for example, associating equipment failure information in text logs with abnormal readings in sensor data.

[0108] Knowledge enhancement of data fusion: The system uses structured knowledge in the knowledge graph to fuse and supplement multimodal data. For example, when the temperature of a device is detected to be abnormal in the sensor data, the system can search for related fault logs and operation records through the knowledge graph to enhance the understanding and inference ability of the abnormality.

[0109] The application modules of knowledge graphs in multimodal data analysis include cross-modal information integration based on knowledge graphs, collaborative analysis of large language models and graph algorithms, anomaly detection and event linkage analysis.

[0110] Cross-modal information integration based on knowledge graph: The system integrates cross-modal information through knowledge graph, unifying different types of data into a shared semantic space. Specific applications include: (1) Unification of semantic levels: Through the knowledge graph, various data types such as text, sensor data, and charts are mapped to the same semantic level, so that data of different modalities can refer to and interpret each other.

[0111] (2) Contextual association analysis: Using the existing contextual information in the knowledge graph, the system can discover potential associations between data of different modalities. For example, by comparing sensor data with operation logs, the system can infer whether a certain operation has caused a decline in device performance.

[0112] Collaborative analysis of large language models and graph algorithms: The system combines the powerful semantic understanding capabilities of large language models with the structural analysis capabilities of graph algorithms to collaboratively analyze multimodal data: (1) Semantic reasoning of large language models: Based on knowledge graphs and log text data, large language models can infer the causal relationship between cross-modal events, such as multimodal anomalies caused by a certain operation.

[0113] (2) Structured analysis of graph algorithms: Using graph algorithms, such as community detection and path search, to identify the potential patterns and structures of multimodal data in the knowledge graph. For example, the system can determine the source of abnormal sensor readings by analyzing the fault propagation path.

[0114] Anomaly detection and event linkage analysis: Through the linkage analysis function of the knowledge graph, the system can identify cross-modal abnormal events. For example: (1) Cross-modal anomaly detection: By combining abnormal fluctuations in sensor data with fault reports in log text, the system can detect potential problems earlier and provide corresponding warnings.

[0115] (2) Event linkage and response suggestions: Based on the linkage analysis of cross-modal data, the system can generate a linkage event chain and provide optimization suggestions to operators, such as how to adjust operations to avoid similar anomalies from happening again.

[0116] The generation module of the comprehensive summary report includes comprehensive analysis of multimodal data, multi-format presentation of cross-modal results, and customized report output and distribution.

[0117] Comprehensive analysis of multimodal data: The system integrates multimodal data to achieve comprehensive analysis of the power system. Using the personalized System Prompt generated by the large language model, the system can generate a comprehensive summary report based on different user needs. For example: (1) Technical staff report: including detailed technical analysis, cross-modal fault detection results and optimization suggestions.

[0118] (2) Management report: including the overall system operation status, summary of key abnormal events and trend forecast.

[0119] Multi-format presentation of cross-modal results: The system supports presenting cross-modal analysis results in multiple formats to suit the reading habits of different users. For example: (1) Charts and data tables: Display the comparative analysis results of cross-modal data, such as the cause analysis diagram of temperature anomaly.

[0120] (2) Timeline and event chain: Display the sequence and relationship of cross-modal events, such as the event chain from sensor anomaly to operational error.

[0121] Customized report output and distribution: The system customizes the output of analysis reports according to user roles and needs, and supports multi-channel distribution. For example, the system can generate a simplified report suitable for reading on mobile devices, or a detailed PDF report suitable for formal reporting.

[0122] Through the above steps, the multimodal data integration and comprehensive analysis module of the present invention can efficiently integrate and analyze different types of power log data, providing a more comprehensive and in-depth system status analysis. This not only enhances the breadth and depth of log analysis, but also improves the early detection and response capabilities for complex power system problems, providing strong technical support for the stable operation of the power system.

[0123] The above are all preferred embodiments of the present invention, and the protection scope of the invention is not limited thereby. Therefore, any equivalent changes made according to the structure, shape, principle and method of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent power log analysis system based on knowledge graph and large language model, characterized in that: Including data collection and preprocessing module, knowledge graph construction and management module, context understanding and association analysis module, personalized report generation module and multimodal data integration module; Data collection and preprocessing module: used to collect raw data from various log sources in the power system, clean, denoise and structure the data, pass the standardized data to the context understanding and association analysis module, and pass the cleaned data to the knowledge graph construction and management module; Knowledge graph construction and management module: used to map the key entities and relationships of the power system into the knowledge graph, form a structured power domain knowledge base, and interact the domain knowledge and context understanding in the knowledge graph with the association analysis module; Context understanding and association analysis module: It is used to perform semantic analysis and context understanding of log content through a large language model and knowledge graph to process the standardized data processed by the data collection and preprocessing module. It uses the relationships and rules in the knowledge graph to perform association analysis on the log data, extract key events and potential problems, and send the results to the personalized report generation module. Personalized report generation module: used to call the large language model to generate personalized log analysis reports based on the user's role and needs, and send the personalized reports to the user; Multimodal data integration module: used to integrate different types of data into a unified knowledge graph and transfer multimodal information to the context understanding and association analysis module.

2. According to claim 1, a power log intelligent analysis system based on knowledge graph and large language model is characterized in that: The data collection and preprocessing module includes a log data collection module, a data cleaning and denoising module, a data format standardization module and a time series processing module; Log data collection module: used to receive data from different equipment logs, operation logs, fault logs and sensors; Data cleaning and denoising module: used to clean records containing noise, redundant information and incomplete information, including removing redundant data, filling missing data and handling outliers; Data format standardization module: used to standardize data in different formats, including field extraction and normalization, and data type conversion; Time serialization processing module: a module used to perform time serialization processing on log data, including event sorting and time window division.

3. According to claim 1, a power log intelligent analysis system based on knowledge graph and large language model is characterized in that: The knowledge graph construction and management module includes a knowledge graph definition and initialization module, a log data mapping to knowledge graph module, a knowledge graph expansion and update module, and an integration module with a domain knowledge base; Knowledge graph definition and initialization module: used to define a knowledge graph structure in the power field, including entities, relationships and attributes; Log data mapping to knowledge graph module: used to map the processed log data to the knowledge graph module, including entity recognition, relationship extraction, and knowledge graph mapping of entities and relationships; Knowledge graph expansion and update module: used for knowledge graphs that need to be dynamically updated and expanded to maintain their reflection of the current power system status, including dynamic entity addition, relationship update and expansion, and knowledge graph optimization; Integration module of domain knowledge base: used to connect with external knowledge base, introduce more domain knowledge, and associate knowledge with real-time log data.

4. According to claim 1, a power log intelligent analysis system based on knowledge graph and large language model is characterized in that: The context understanding and association analysis module includes a semantic parsing module for log data, a knowledge graph-driven context information extraction module, and a log information association analysis and problem detection module; Semantic parsing module for log data: including text parsing and information extraction, semantic understanding and context modeling, and semantic enhancement driven by knowledge graph; Knowledge graph-driven context information extraction module: including extraction of related entities and events, causal reasoning combining large language models with graph algorithms, and abnormal event detection; Log information correlation analysis and problem detection module: including multi-log source information fusion, key event annotation and priority sorting, and problem analysis and suggestion generation.

5. According to claim 1, a power log intelligent analysis system based on knowledge graph and large language model is characterized in that: The personalized report generation module includes a user demand and role identification module, a large language model personalized generation mechanism module based on System Prompt, a report content customization module based on user demand, and a multi-format output module for log analysis reports; User needs and role identification module: including user role classification, needs identification and preference setting; Large language model personalized generation mechanism module based on System Prompt: including customized settings of System Prompt, context-aware generation, diversified information presentation, and multi-language and professional terminology support; Report content customization module based on user needs: including dynamic content customization, hierarchical content presentation and report structure adaptation; Multi-format output module for log analysis reports: includes report format support and interactive report generation.

6. According to claim 1, a power log intelligent analysis system based on knowledge graph and large language model is characterized in that: The multimodal data integration module includes a multimodal data fusion and association module, a knowledge graph application module in multimodal data analysis, and a comprehensive summary report generation module; Multimodal data fusion and association module: including multimodal data collection and preprocessing, knowledge graph mapping of multimodal data and knowledge enhancement of data fusion; Application modules of knowledge graphs in multimodal data analysis: including cross-modal information integration of knowledge graphs, collaborative analysis of large language models and graph algorithms, and anomaly detection and event linkage analysis; Comprehensive summary report generation module: including comprehensive analysis of multimodal data, multi-format presentation of cross-modal results, and customized report output and distribution.

Images