LLM-based power distribution network knowledge graph construction and fault intelligent diagnosis method

Abstract

The invention relates to an LLM-based power distribution network knowledge graph construction and fault intelligent diagnosis method. The method comprises the steps of constructing a knowledge system, constructing a normal operation state model and an anomaly detection threshold value, performing fault diagnosis through an LSTM-BP cascade network, and performing fault classification. The abnormal parameters and the fault classification result are input into an LLM model, the fault type is judged, corresponding diagnosis information is obtained, and the system generates a scheduling instruction and an emergency decision suggestion according to the diagnosis information; and generating a detailed diagnosis report. According to the system, statistical analysis and deep learning algorithms are adopted for operation parameters such as voltage and current after preprocessing, and automatic detection and fault classification of abnormal parameters are realized in combination with a CNN recognition module; a preset rule base in the knowledge system is combined to accurately determine a fault type, generate a scheduling instruction and an emergency decision suggestion, and assist operation and maintenance personnel in rapid response and dynamic management; finally, a detailed diagnosis report is generated, thereby improving the fault handling efficiency and the guarantee capability of safe and stable operation of a power grid.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault diagnosis technology, specifically to a method for constructing a distribution network knowledge graph and intelligent fault diagnosis based on LLM. Background Technology

[0002] With the continuous development of smart power, distribution networks, due to their extremely complex structure, electromagnetic relationships, and fault causes, are moving towards data-driven, intelligent reasoning, and real-time diagnosis to improve their monitoring efficiency. Distribution network fault detection currently relies heavily on expert experience and pre-defined rules, combined with statistical analysis and data mining techniques to achieve fault location.

[0003] These methods can be effective in scenarios with relatively simple faults or well-known patterns. Existing technologies rely heavily on real-time monitoring and expert experience, neglecting a large amount of unstructured information such as equipment manuals, design documents, and historical maintenance records. The rich fault characteristics and maintenance experience contained in this data can further improve the condition monitoring and operation and maintenance level of the distribution network. Specifically: 1) Existing technologies rely heavily on manually formulated rules and expert experience. When faced with complex or unknown faults, the rules are difficult to fully cover, and the diagnostic results lack flexibility and accuracy.

[0004] 2) Current technologies often struggle to integrate multi-source data when constructing knowledge graphs, and their update mechanisms are not flexible enough, making it difficult for knowledge graphs to reflect the latest status and fault modes of the power distribution network system in a timely manner.

[0005] 3) In the existing technology, the sources of information such as equipment manuals, design documents, historical maintenance records and real-time sensor data are different and the formats are not uniform. Data noise and missing values ​​are common problems, resulting in a large workload for data preprocessing and difficulty in ensuring accuracy. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM, so as to overcome the shortcomings of the prior art.

[0007] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM, comprising the following steps: Step S01: Obtain structured and unstructured data from the power distribution network system, store the structured and unstructured data in a database, and build a knowledge system; Step S02: Input the preprocessed voltage and current operating parameters into the anomaly identification module. The anomaly identification module combines statistical analysis methods and deep learning algorithms to construct a normal operating state model and anomaly detection threshold. When the collected parameter data exceeds the preset normal range or abnormal fluctuations occur, the system automatically identifies it as an abnormal parameter and performs fault diagnosis and fault classification through an LSTM-BP cascaded network. Step S03: Input the abnormal parameters and fault classification results into the LLM model, combine them with the relevant knowledge preset rule base in the knowledge system, determine the fault type, and obtain the corresponding diagnostic information. The system generates scheduling instructions and emergency decision suggestions based on the diagnostic information. Step S04: Through the formatted output unit, perform multi-level comprehensive analysis of the processing results of each module, and generate a detailed diagnostic report using the preset report template.

[0008] The beneficial effects of this invention are as follows: Addressing the urgent need for fault diagnosis and management in distribution networks under the current context of intelligent and digital development of power systems, this invention proposes a comprehensive method integrating knowledge construction, abnormal parameter identification, decision support, and diagnostic report generation. The system first uses a structured database to store power system data, operation logs, and equipment status information, and utilizes text embedding technology to construct a vector database, enabling efficient construction and intelligent retrieval of power system knowledge. Subsequently, by applying statistical analysis and deep learning algorithms to preprocessed operating parameters such as voltage and current, combined with a CNN recognition module, it achieves automatic detection of abnormal parameters and fault classification. Next, the abnormal parameters and fault classification results are input into an LLM model, and combined with a pre-set rule base in the knowledge system, the fault type is accurately determined, generating dispatch instructions and emergency decision suggestions to assist maintenance personnel in rapid response and dynamic management. Finally, through a formatted output unit, a detailed diagnostic report is automatically generated, including fault description, fault location, fault cause, impact range, emergency measures, and prevention suggestions, thereby significantly improving fault handling efficiency and the ability to ensure the safe and stable operation of the power grid.

[0009] Based on the above technical solution, the present invention can be further improved as follows.

[0010] Furthermore, step S01 specifically includes the following steps: Step S11: Obtain structured and unstructured data of the power distribution network system; Structured data includes equipment status data, real-time monitoring data, and historical fault record data; Unstructured data includes operation logs, operational and maintenance diagnostic experience, and fault descriptions; Step S12: Based on the PostgreSQL database, design the data table structure according to the data characteristics of the power system distribution network, and process the structured and unstructured data separately: For structured data, it is stored in different data tables and indexed to obtain a structured database; For unstructured text data, LLM is used to perform text semantic vectorization processing to generate a corresponding vector database. A knowledge system is constructed by mapping structured databases to vector databases.

[0011] Furthermore, step S02 specifically includes the following steps: Step S21: Collect key operating parameters, including voltage, current, and frequency, from sensors in the power distribution network system in real time as features to form an input dataset. ; Step S22: Calculate the input dataset The eigenvalue of each feature Based on the mean of features Calculate the standard deviation of each feature The calculation formula is as follows: ; In the formula, It is the sample size. It is the first The original feature values ​​of each sample; Step S23: For each sample, the original feature values Data standardization is performed, and standardized eigenvalues ​​are calculated using formulas. ; Step S24: Standardize the eigenvalues The data is fed into an LSTM-BP for recognition training to obtain a large classification and recognition model. Step S25: Input the key operating parameters, including voltage, current and frequency, into the classification and identification model, and output the fault classification results.

[0012] Furthermore, the fault classification results in step S25 include abnormal parameters and the labels corresponding to the abnormal parameters; Abnormal parameters include abnormal voltage rise data, abnormal voltage fall data, abnormal current rise data, and abnormal current fall data identified for abnormal current and voltage waveforms. The tags include the fault category, fault time, fault location, faulty equipment, and weather at the fault location corresponding to each abnormal parameter.

[0013] Furthermore, step S03 specifically includes the following steps: Step S31: Input the fault classification results output in step S25 into the intelligent decision-making system, and use the LLM large language model for analysis to generate possible explanations of fault causes based on the knowledge system; Step S32: Combine rule-based reasoning and prediction to assess the fault development trend and impact range. Use the fault development trend and impact range as input to trigger the intelligent reasoning and decision support module to generate corresponding fault diagnosis and scheduling optimization schemes.

[0014] Furthermore, the rule-based reasoning prediction in step S32 includes: LLM first stores the input anomaly description text vectors in a vector database and constructs an efficient vector index structure; the vector database supports similarity search based on cosine similarity and other distance metrics, the calculation formula of which is: ; Based on retrieved historical cases, LLM generates fault cause reasoning: according to equipment operation specifications and rule base regulations, it determines whether the anomaly conforms to the established fault mode; rule reasoning uses a rule base formulated by experts to perform logical analysis on the abnormal data; it analyzes waveform characteristics to assess the fault development trend and impact range.

[0015] Furthermore, the standardized eigenvalues ​​in step S23 The calculation formula is as follows: ; In the formula, It is the first The original feature values ​​of each sample; The mean of the corresponding feature; is the standard deviation of the corresponding feature.

[0016] Furthermore, step S04 specifically includes the following steps: Step S41: Classify information according to different information priorities; based on the Large Language Model (LLM), convert the analysis results into a highly readable text report; Step S42: Based on the fault type, the system automatically matches preset diagnostic templates including short circuit faults, abnormal voltage fluctuations, and equipment abnormal alarms, and fills in the sorted key information; Step S43: Use Large Language Model (LLM) combined with Natural Language Generation (NLG) technology to automatically generate a diagnostic report that meets industry standards; Step S44: The system sends the diagnostic report to the maintenance personnel via PDF, API push, SMS or email notification, and automatically archives it to form a queryable historical fault database, providing data support for subsequent fault analysis. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the LLM-based intelligent diagnostic method for power distribution networks according to the present invention; Figure 2 This is a flowchart of the power frequency signal preprocessing of the present invention; Figure 3 This is a flowchart of the abnormal parameter identification process of the present invention; Figure 4 This is a diagram of the LSTM network structure of the present invention. Detailed Implementation

[0018] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0019] like Figures 1-4 As shown in Example 1, a method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM includes the following steps: Step S01: Knowledge System Construction: Obtain structured and unstructured data from the power distribution network system, store the structured and unstructured data in a database, and construct a knowledge system; Step S02: Abnormal Parameter Identification: The preprocessed voltage and current operating parameters are input into the abnormal identification module. The abnormal identification module combines statistical analysis methods and deep learning algorithms to construct a normal operating state model and anomaly detection threshold. When the collected parameter data exceeds the preset normal range or abnormal fluctuations occur, the system automatically identifies it as an abnormal parameter and performs fault diagnosis and fault classification through an LSTM-BP cascaded network. Step S03: Decision Support System: Input the abnormal parameters and fault classification results into the LLM model, combine them with the relevant knowledge preset rule base in the knowledge system, determine the fault type, and obtain the corresponding diagnostic information. Based on the diagnostic information, the system generates dispatch instructions and emergency decision suggestions to assist power operation and maintenance personnel in taking effective measures in a timely manner to achieve rapid response and dynamic management of distribution network faults.

[0020] Step S04: Diagnostic report generation: Through the formatted output unit, the processing results of each module are comprehensively analyzed at multiple levels, and a detailed diagnostic report is generated using the preset report template.

[0021] To address the urgent need for fault diagnosis and management in distribution networks under the current context of intelligent and digital development of power systems, this paper proposes a comprehensive method integrating knowledge construction, abnormal parameter identification, decision support, and diagnostic report generation. The system first uses a structured database to store power system data, operation logs, and equipment status information, and then utilizes text embedding technology to construct a vector database, enabling efficient construction and intelligent retrieval of power system knowledge. Subsequently, by applying statistical analysis and deep learning algorithms to preprocessed operating parameters such as voltage and current, combined with a CNN recognition module, it achieves automatic detection of abnormal parameters and fault classification. Next, the abnormal parameters and fault classification results are input into an LLM model, and combined with a pre-set rule base in the knowledge system, the system accurately determines the fault type, generates dispatch instructions and emergency decision suggestions, assisting maintenance personnel in rapid response and dynamic management. Finally, through a formatted output unit, a detailed diagnostic report is automatically generated, including fault description, fault location, fault cause, impact range, emergency measures, and prevention suggestions, thereby significantly improving fault handling efficiency and the ability to ensure the safe and stable operation of the power grid.

[0022] Example 2 is a further improvement based on Example 1, and its details are as follows: Step S01 specifically includes the following steps: Step S11: Obtain structured and unstructured data of the power distribution network system; Structured data includes equipment status data (including voltage, current, and power), real-time monitoring data, and historical fault record data; Unstructured data includes operation logs, operational and maintenance diagnostic experience, and fault descriptions; Data acquisition can be achieved through interfaces of real-time data acquisition systems, sensor networks, and historical databases. Step S12: Based on the PostgreSQL database, design the data table structure according to the data characteristics of the power system distribution network, and process the structured and unstructured data separately: For structured data, it is stored in different data tables and indexed to obtain a structured database, thereby optimizing query efficiency. For unstructured text data, LLM is used to perform text semantic vectorization processing to generate a corresponding vector database. A knowledge system is constructed by mapping structured databases to vector databases.

[0023] Unstructured data includes runtime log text data, maintenance and diagnostic experience text data, and fault description text data; let a piece of text be... Using embedding functions Its high-dimensional vector representation is obtained: ; In the formula, It is a vector; this vector It captures semantic information in the text, which facilitates subsequent similarity comparison and knowledge retrieval.

[0024] By organically combining structured and vector databases, a unified knowledge system is formed. Structured data provides precise parameter and status information, while vector databases supplement the semantic information of unstructured text data. The system has a built-in unified query interface that supports cross-database retrieval and correlation analysis. Maintenance personnel or intelligent diagnostic modules can quickly locate relevant information by inputting keywords or fault descriptions and leveraging the similarity search function of the vector database.

[0025] Example 3 is a further improvement based on Example 1, and its details are as follows: Step S02 specifically includes the following steps: Step S21: Collect key operating parameters, including voltage, current, and frequency, from sensors in the power distribution network system in real time as features to form an input dataset. ; Step S22: Calculate the input dataset The eigenvalue of each feature Based on the mean of features Calculate the standard deviation of each feature This is used to measure the degree of dispersion of eigenvalues, and the calculation formula is as follows: ; In the formula, It is the sample size. It is the first The original feature values ​​of each sample; Step S23: For each sample, the original feature values Data standardization is performed, and standardized eigenvalues ​​are calculated using formulas. To eliminate amplitude differences between different signals, so that all signals can be analyzed on the same scale and standardized eigenvalues ​​can be obtained; Step S24: Standardize the eigenvalues The data is fed into LSTM-BP (Long Short-Term Memory Cascaded Backpropagation Neural Network) for recognition training to obtain a large classification and recognition model; Step S25: Input key operating parameters, including voltage, current, and frequency, into the classification and identification model, and output the fault classification results. In specific implementations, key operating parameters also include voltage and current waveforms; LSTM networks are a variant of RNNs that can, to some extent, address the problems of vanishing gradients and long-term reliance on previous information. Their structure is as follows: Figure 3 As shown; a BP neural network is a hierarchical network structure that connects multiple neurons according to certain rules. It consists of three parts: an input layer, hidden layers, and an output layer. Layers are connected by corresponding weights and thresholds. Training data is used to obtain various connection weights and thresholds for classification and recognition on a test set. Its structure is as follows. Figure 4 As shown.

[0026] Example 4 is a further improvement on Example 3, and its details are as follows: The fault classification result in step S25 includes abnormal parameters and the labels corresponding to the abnormal parameters; Abnormal parameters include abnormal voltage rise data, abnormal voltage fall data, abnormal current rise data, and abnormal current fall data identified for abnormal current and voltage waveforms. The tags include the fault category, fault time, fault location, faulty equipment, and weather at the fault location corresponding to each abnormal parameter.

[0027] It provides key data such as the severity of the fault, the scope of its impact, trend prediction, and intelligent decision-making suggestions for decision support systems.

[0028] Example 5 is a further improvement on Example 4, and its details are as follows: Step S03 specifically includes the following steps: The decision support system, based on the results of the abnormal parameter identification module, combined with a knowledge system and a large language model (LLM), provides intelligent fault assessment, emergency response suggestions, and dispatch decision support to achieve rapid response and dynamic management of the distribution network.

[0029] Step S31: Input the fault classification results output in step S25 into the intelligent decision-making system, and use the LLM large language model for analysis. Based on the knowledge system (such as comparing historical fault record data and fault descriptions in the knowledge system), generate possible explanations of fault causes. Step S32: Combine rule-based reasoning and prediction to assess the fault development trend and impact range. Use the fault development trend and impact range as input to trigger the intelligent reasoning and decision support module to generate corresponding fault diagnosis and scheduling optimization schemes.

[0030] Example 6 is a further improvement on Example 5, and its details are as follows: The associative rule-based reasoning prediction in step S32 includes: LLM first stores the input anomaly description text vectors in a vector database and constructs an efficient vector index structure; the vector database supports similarity search based on cosine similarity and other distance metrics, the calculation formula of which is: ; Based on retrieved historical cases, LLM generates fault cause reasoning: 1. Rule matching: According to the equipment operation specifications and rule base regulations, determine whether the anomaly conforms to the established fault mode; rule reasoning uses an expert-developed rule base to perform logical analysis on the abnormal data. For example, if the voltage drop is >20% and the duration is >3 seconds, it may be a load change; if the current suddenly increases and is accompanied by voltage fluctuations, it may be a short circuit fault; 2. Pattern reasoning: Analyze waveform characteristics to assess the fault development trend and impact range.

[0031] Example 7 is a further improvement based on any one of Examples 3 to 6, and its details are as follows: Standardized eigenvalues ​​in step S23 The calculation formula is as follows: ; In the formula, It is the first The original feature values ​​of each sample; The feature mean of the corresponding feature; This represents the standard deviation of the corresponding feature. Through the above data standardization calculation, different features can be scaled to the same scale, thereby eliminating dimensional differences between features.

[0032] Example 8 is a further improvement based on Example 1, and its details are as follows: Step S04 specifically includes the following steps: In an LLM-based intelligent diagnostic system for distribution networks, the diagnostic report generation module performs multi-level comprehensive analysis of the processing results from various analysis modules and outputs a complete diagnostic report in a structured format. This report covers fault description, fault location, fault cause, impact range, emergency measures, and preventative recommendations, providing power operation and maintenance personnel with intuitive, accurate, and actionable fault handling guidelines.

[0033] Step S41: Classify the diagnostic information (including model recognition results, knowledge graph reasoning results, expert rule suggestions, etc.) according to different information priorities (such as emergency faults, potential risks, optimization suggestions); based on the Large Language Model (LLM), convert the diagnostic information into a highly readable text report; In practice, the classification includes: Basic information: such as fault type, time of occurrence, equipment name, and geographical location; Waveform characteristics: including current and voltage value changes and rate analysis of abnormal waveforms; Reasoning Result Category: Possible causal chains inferred from knowledge graph associations and expert rule engines; Response suggestions: Referencing historical cases and rules, generate response plans and early warning prompts.

[0034] The diagnostic information includes a series of comprehensive diagnostic information derived from collaborative reasoning by classification models, expert systems, and knowledge graphs, such as fault identification categories, abnormal parameters, cause chains, possible consequences, and coping strategies.

[0035] LLM is used to perform natural language generation and semantic completion on the categorized and organized diagnostic information, generating a fault diagnosis text report with a clear structure, consistent terminology, and easy readability. The report content includes: 1. Fault Description and Background; 2. Summary of raw data (standardized waveforms, characteristic indicators); 3. Fault cause inference chain; 4. Operation and maintenance suggestions and risk warnings; 5. Other additional information (such as weather, power grid topology context); Step S42: Based on the fault type, the system automatically matches preset diagnostic templates including short circuit faults, abnormal voltage fluctuations, and equipment abnormal alarms, and fills in the sorted key information; Step S43: Using a large language model (LLM) combined with natural language generation (NLG) technology, an industry-standard diagnostic report is automatically generated. This report covers fault description, fault location, fault cause, scope of impact, emergency measures, and prevention suggestions to improve the readability and professionalism of the report and provide power operation and maintenance personnel with intuitive, accurate, and actionable fault handling basis. Step S44: The system sends the diagnostic report to the maintenance personnel via PDF, API push, SMS or email notification, and automatically archives it to form a queryable historical fault database, providing data support for subsequent fault analysis.

[0036] Based on the fault identification results and knowledge graph reasoning process, combined with the diagnostic information provided by the decision support system, a structured, multi-dimensional fault diagnosis text report is generated through LLM (Large Language Model).

[0037] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for constructing a knowledge graph of a distribution network and for intelligent fault diagnosis based on LLM, characterized in that, Includes the following steps: Step S01: Obtain structured and unstructured data from the power distribution network system, store the structured and unstructured data in a database, and construct a knowledge system; Step S02: Input the preprocessed voltage and current operating parameters into the anomaly identification module. The anomaly identification module combines statistical analysis methods and deep learning algorithms to construct a normal operating state model and anomaly detection threshold. When the collected parameter data exceeds the preset normal range or abnormal fluctuations occur, the system automatically identifies it as an abnormal parameter and performs fault diagnosis and fault classification through an LSTM-BP cascaded network. Step S03: Input the abnormal parameters and the results of the fault classification into the LLM model, combine them with the relevant knowledge preset rule base in the knowledge system, determine the fault type, and obtain the corresponding diagnostic information. The system generates scheduling instructions and emergency decision suggestions based on the diagnostic information. Step S04: Through the formatted output unit, perform multi-level comprehensive analysis of the processing results of each module, and generate a detailed diagnostic report using the preset report template.

2. The method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM as described in claim 1, characterized in that, Step S01 specifically includes the following steps: Step S11: Obtain structured and unstructured data of the power distribution network system; The structured data includes equipment status data, real-time monitoring data, and historical fault record data; The unstructured data includes operation logs, operation and maintenance diagnostic experience, and fault descriptions; Step S12: Based on the PostgreSQL database, design the data table structure according to the data characteristics of the power system distribution network, and process the structured data and the unstructured data separately: For structured data, it is stored in different data tables and indexed to obtain a structured database; For unstructured text data, LLM is used to perform text semantic vectorization processing to generate a corresponding vector database. A knowledge system is constructed by mapping the structured database to the vector database.

3. The method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM as described in claim 2, characterized in that, Step S02 specifically includes the following steps: Step S21: Collect key operating parameters, including voltage, current, and frequency, from sensors in the power distribution network system in real time as features to form an input dataset. ; Step S22: Calculate the input dataset The eigenvalue of each feature Based on the mean of the features Calculate the standard deviation of each feature The calculation formula is as follows: ； In the formula, It is the sample size. It is the first The original feature values ​​of each sample; Step S23: For each sample, the original feature values Data standardization is performed, and standardized eigenvalues ​​are calculated using formulas. ; Step S24: Standardize the feature values The data is fed into an LSTM-BP for recognition training to obtain a large classification and recognition model. Step S25: Input the key operating parameters, including voltage, current and frequency, into the classification and identification model, and output the fault classification results.

4. The method for constructing a knowledge graph and intelligently diagnosing faults in a distribution network based on LLM as described in claim 3, characterized in that, The fault classification result in step S25 includes abnormal parameters and the labels corresponding to the abnormal parameters. The abnormal parameters include abnormal voltage rise data, abnormal voltage fall data, abnormal current rise data, and abnormal current fall data identified for abnormal current and voltage waveforms. The tags include the fault category, fault time, fault location, fault equipment, and weather at the fault location corresponding to each of the abnormal parameters.

5. The method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM as described in claim 4, characterized in that, Step S03 specifically includes the following steps: Step S31: Input the fault classification result output in step S25 into the intelligent decision system, and use the LLM large language model to analyze it, and generate possible explanations of fault causes based on the knowledge system; Step S32: Combine rule-based reasoning and prediction to assess the fault development trend and impact range. Use the fault development trend and impact range as input to trigger the intelligent reasoning and decision support module to generate corresponding fault diagnosis and scheduling optimization schemes.

6. The method for constructing a knowledge graph and intelligently diagnosing faults in a distribution network based on LLM as described in claim 5, characterized in that, The rule-based reasoning prediction in step S32 includes: LLM first stores the input anomaly description text vector in the vector database and constructs an efficient vector index structure; the vector database supports similarity search based on cosine similarity and other distance metrics, the calculation formula of which is: ； Based on retrieved historical cases, LLM generates fault cause reasoning: according to equipment operation specifications and rule base regulations, it determines whether the anomaly conforms to the established fault mode; rule reasoning uses a rule base formulated by experts to perform logical analysis on the abnormal data; it analyzes waveform characteristics to assess the fault development trend and impact range.

7. A method for constructing a knowledge graph and intelligently diagnosing faults in a distribution network based on LLM, as described in any one of claims 3 to 6, characterized in that, The standardized feature values ​​in step S23 The calculation formula is as follows: ； In the formula, It is the first The original feature values ​​of each sample; The feature mean of the corresponding feature; is the standard deviation of the corresponding feature.

8. The method for constructing a knowledge graph of a distribution network and intelligently diagnosing faults based on LLM according to claim 1, characterized in that, Step S04 specifically includes the following steps: Step S41: Classify the information according to different information priorities; based on the Large Language Model (LLM), convert the diagnostic information into a highly readable text report; Step S42: Based on the fault type, the system automatically matches preset diagnostic templates including short circuit faults, abnormal voltage fluctuations, and equipment abnormal alarms, and fills in the sorted key information; Step S43: Use Large Language Model (LLM) combined with Natural Language Generation (NLG) technology to automatically generate a diagnostic report that meets industry standards; Step S44: The system sends the diagnostic report to the maintenance personnel via PDF, API push, SMS or email notification, and automatically archives it to form a queryable historical fault database, providing data support for subsequent fault analysis.

Images