Electricity knowledge question answering system based on language model

Abstract

The present invention relates to the technical field of knowledge question and answer, specifically to an electric power knowledge question and answer system based on a language model, and the electric power knowledge question and answer system based on a language model includes a model terminology dynamic update module, a data security monitoring module, a load forecast analysis module, and a fault answer generation module. In the present invention, through term extraction and context analysis, the electric power industry knowledge base is dynamically updated, the timeliness and accuracy of the terms are improved, and the knowledge base is guaranteed to be able to reflect the latest changes in the industry in a timely manner. A multivariate time series analysis method is adopted to refine the fluctuation prediction of power demand, and through feature extraction of regional and time grouping, an accurate basis is provided for different power demand scenarios. Combined with the equipment's historical maintenance records and fault database information, accurate positioning of equipment anomalies is achieved, reducing the time for equipment fault troubleshooting, and through data access pattern analysis and layered encryption processing, the reliable operation of the system is guaranteed, greatly improving the operation and maintenance efficiency and reliability of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge question answering, and in particular to an electric power knowledge question answering system based on a language model. Background Art

[0002] The field of knowledge question answering technology involves the use of natural language processing (NLP) technology to achieve effective communication between humans and machines, so that machines can understand, process and respond to queries in the form of human language. This field combines multiple technologies such as machine learning, semantic understanding, context analysis and information retrieval, with the goal of extracting valuable information from large amounts of data to answer specific questions. It usually includes steps such as language model training, question understanding, key information extraction and answer generation. It is currently widely used, such as intelligent assistants, online customer service, professional consulting services, etc., which are all specific application scenarios of knowledge question answering technology.

[0003] Among them, the language model power knowledge question-answering system aims to apply knowledge question-answering technology to the power field to provide instant information about the power system and answer user questions. The system analyzes data and literature in the power industry and uses a well-trained language model to understand and answer questions about power equipment, energy distribution, system maintenance, etc. Its main uses include helping power engineers quickly obtain decision-making information, assisting power managers in optimizing system operations, and providing consumers with advice on energy use and power safety, thereby improving the efficiency and reliability of the power system.

[0004] Existing knowledge question-answering technologies lack a real-time update mechanism for power industry terminology and are unable to quickly synchronize new terminology or technological changes within the industry, resulting in a lag in updating the knowledge base and affecting the accuracy of answering questions. Power load forecasting mainly relies on single-dimensional data, ignoring the impact of multiple factors such as meteorological conditions and user behavior, making load forecasting results lack precision and unable to cope with complex demand fluctuation scenarios. In the process of equipment fault diagnosis, the analysis of abnormal load fluctuations is mostly limited to the current data of the equipment, and fails to effectively combine historical maintenance records and fault databases. The degree of automation of fault diagnosis is low, making it difficult to accurately locate problems, resulting in poor efficiency in formulating maintenance plans. In addition, existing technologies only rely on basic verification in terms of data security, fail to effectively prevent abnormal behavior in data access, have security loopholes, and are prone to data leakage and other risks, affecting the normal operation of the power system. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose an electric power knowledge question-answering system based on a language model.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: the power knowledge question-answering system based on the language model comprises:

[0007] The model terminology dynamic update module collects newly released power industry documents, extracts terms and their contextual definitions, compares existing terms in the knowledge base and analyzes the matching degree. According to the matching degree analysis results, it sets a judgment threshold to identify new terms that have not been entered, and stores them synchronously in the question-and-answer knowledge base to obtain the knowledge base update record;

[0008] The data security monitoring module uses the knowledge base to update records, performs identity and source verification on each data request, and performs layered encryption processing, records each data operation based on the layered encryption processing results, analyzes data access patterns, identifies and marks abnormal access behaviors, and obtains security monitoring logs;

[0009] The load forecasting and analysis module collects and integrates power load data, meteorological information and user behavior statistics based on the data security mode in the security monitoring log, predicts power demand fluctuations through multivariate time series analysis, groups demand fluctuations of different times and regions according to the power demand fluctuation prediction results, and performs feature extraction to generate power load forecast results;

[0010] The fault solution generation module monitors and collects real-time operation data of power equipment according to the power load forecast results, analyzes abnormal load fluctuations, integrates historical equipment maintenance records and fault database information, formulates target maintenance plans and preventive measures, and outputs fault solution plans.

[0011] As a further solution of the present invention, the matching degree analysis step is:

[0012] Retrieve newly released power industry documents from online databases and power industry websites, screen for target keywords, and generate a preliminary document list;

[0013] Based on the preliminary document list, extract key terms and contexts to ensure that the context of each term is accurately captured, and store them in a temporary database to obtain a detailed terminology and definition set;

[0014] Based on the detailed terms and their definition set, compared with the existing knowledge base, the formula is adopted. ;

[0015] Calculate the matching degree of each new term with the existing knowledge base terms , generate matching analysis results, where Represents the weight factor, which is used to adjust the influence of each term feature. and Represent the term vectors in the existing knowledge base and the newly extracted term vectors respectively, and Respectively represent the modulus of the existing knowledge base and the newly extracted term vector.

[0016] As a further solution of the present invention, the steps of obtaining the knowledge base update record are:

[0017] Based on the matching analysis result, a judgment threshold is set, terms with matching degrees lower than the judgment threshold are identified and determined as new terms, and a list of unentered new terms is generated;

[0018] Analyze the list of unrecorded new terms, verify the application scenario and technical accuracy of each new term based on the application importance of the power industry and industry standard documents, conduct document comparison and expert review, and obtain a verified new term set;

[0019] The verified new term set is synchronously updated to the question-and-answer knowledge base, the index and search functions of the knowledge base are updated, and the addition time and document source of each term are recorded to generate a knowledge base update record.

[0020] As a further solution of the present invention, the execution steps of the layered encryption process are:

[0021] Extracting data request information from the knowledge base update record, including the requester identifier and source address, determining the legitimacy and integrity of the request, accurately tracking each request, and generating a request review file;

[0022] Based on the request review file, the identity and source of the requester are verified using security protocols and verification mechanisms, and the validity of the request is confirmed using multi-factor identity authentication and source address checks to obtain security verification pass information;

[0023] The security verification information is encrypted through multiple layers, and the AES and RSA encryption standards are used to strengthen data protection, optimize the security of data in transmission and storage, and generate data encryption processing results.

[0024] As a further solution of the present invention, the steps of obtaining the security monitoring log are:

[0025] According to the data encryption processing result, record the details of each data operation, including the operator, operation type and timestamp and its hierarchical encryption result, ensure that all data interactions are tracked and recorded, and generate an operation record log;

[0026] Analyze the operation record log to identify normal and abnormal patterns of data access, using the formula, ;

[0027] Calculate standard statistics , set the abnormality judgment value ,like Value greater than , then the access behavior is judged to be abnormal, and the pattern recognition analysis results are obtained, among which, represents a data point for a single visit, Represents the average value of historical data access, Represents the variance of historical data;

[0028] According to the pattern recognition analysis results, abnormal behaviors are identified and marked, and alarms are generated synchronously to obtain security monitoring logs.

[0029] As a further solution of the present invention, the steps of predicting the power demand fluctuation are:

[0030] Based on the security monitoring log, all log data related to power load changes are screened, the timestamp and load value fields in the log are called, the integrity and consistency of the data are verified, and the power load data after security filtering is generated;

[0031] Integrate the securely filtered power load data with meteorological information and user behavior statistics, match them according to time series, form a unified time series table for the load, meteorological and user behavior data of different time periods, and generate an integrated multivariate time series data set;

[0032] The integrated multivariate time series data set is called to perform multivariate time series analysis using the formula: ;

[0033] Calculate electricity demand fluctuation forecasts , analyze the fluctuation range of power load and generate power demand fluctuation prediction results, among which, For the The power load at a given moment, for The weight correction factor of the momentary load, For time The meteorological factors within For the User behavior statistics at a certain moment, is the user behavior correction coefficient, Indicates the number of moments.

[0034] As a further solution of the present invention, the steps for obtaining the power load forecast result are:

[0035] Based on the power demand fluctuation prediction results, the demand fluctuation value of each time period is extracted, and the formula is used: ;

[0036] Calculate the average volatility for each time period , and compare them one by one with the set volatility standard, divide the time periods into three categories: high volatility, medium volatility, and low volatility, and generate time difference demand volatility groups, among which, The power demand fluctuation during this period, is the total time of the period, Indicates the total time period;

[0037] Based on the time difference demand fluctuation grouping, the regional power load value is extracted, using the formula, ;

[0038] Calculate the correlation score between the regional average load value and the demand fluctuation category during different periods , determine the load performance characteristics of each region in different time periods, and generate regional demand fluctuation groups, where, is the average load value for each area, is the time category, For time The regional load characteristics under is the load correction factor;

[0039] Combined with the time difference demand fluctuation grouping and the regional demand fluctuation grouping, feature extraction is performed, the impact of the grouping characteristics on the power load is analyzed, and the power load prediction result is generated.

[0040] As a further solution of the present invention, the steps of obtaining the troubleshooting solution are:

[0041] Based on the power load forecast result, the real-time operation data of the power equipment is called, the equipment operation load at each moment is compared with the forecast load value, the deviation value between the real-time operation load and the forecast load is calculated, the time point when the deviation exceeds the deviation threshold is extracted, and the load abnormality time mark is generated;

[0042] According to the load abnormality time mark, it is cross-compared with the historical maintenance records of the equipment and the information in the fault database. Through similar fault cases and maintenance conditions in the historical records, the equipment fault characteristics related to the current load abnormality are extracted, and combined with the equipment model, maintenance time and fault type, a preliminary fault diagnosis analysis result is generated;

[0043] Based on the preliminary analysis results of the fault diagnosis, the formula is used: ;

[0044] Calculate the comprehensive evaluation index of fault , formulate target maintenance plans and preventive measures, and output troubleshooting solutions, including: For the The severity of the fault, is the fault impact range value, The device's historical fault repair time. Prioritize preventive measures. is the correction factor for preventive measures, Indicates the fault type.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, the power industry knowledge base is dynamically updated through term extraction and context analysis, which improves the timeliness and accuracy of the terms, ensures that the knowledge base can promptly reflect the latest changes in the industry, adopts a multivariate time series analysis method, refines the prediction of power demand fluctuations, and provides an accurate basis for different power demand scenarios through feature extraction of regional and time groupings. Combined with the equipment's historical maintenance records and fault database information, accurate positioning of equipment anomalies is achieved, reducing the time for equipment troubleshooting. Through data access pattern analysis and layered encryption processing, the reliable operation of the system is guaranteed, and the automation and accuracy of power equipment fault diagnosis, maintenance plan formulation and preventive measures are comprehensively realized, greatly improving the operation and maintenance efficiency and reliability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a system flow chart of the present invention;

[0048] Figure 2 This is a flow chart of the analysis of the matching degree of the present invention;

[0049] Figure 3 A flowchart for obtaining update records of the knowledge base of the present invention;

[0050] Figure 4 This is an execution flow chart of the layered encryption process of the present invention;

[0051] Figure 5 This is a flow chart for obtaining the security monitoring log of the present invention;

[0052] Figure 6 A flow chart for predicting power demand fluctuations according to the present invention;

[0053] Figure 7 A flowchart for obtaining the power load forecasting result of the present invention;

[0054] Figure 8 The figure is a flow chart for obtaining the troubleshooting solution of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0057] See also Figure 1 , the power knowledge question answering system based on language model includes:

[0058] The model terminology dynamic update module collects newly released power industry documents, extracts terms and their contextual definitions, compares existing terms in the knowledge base and analyzes the matching degree. According to the matching degree analysis results, it sets a judgment threshold to identify new terms that have not been entered, and stores them synchronously in the question-and-answer knowledge base to obtain the knowledge base update record;

[0059] The data security monitoring module uses the knowledge base to update records, performs identity and source verification on each data request, and performs layered encryption processing. According to the layered encryption processing results, it records each data operation, analyzes data access patterns, identifies and marks abnormal access behaviors, and obtains security monitoring logs;

[0060] The load forecasting analysis module collects and integrates power load data, meteorological information and user behavior statistics based on the data security model in the security monitoring log, and predicts power demand fluctuations through multivariate time series analysis. According to the power demand fluctuation prediction results, the demand fluctuations with different time and area are grouped, and feature extraction is performed to generate power load forecast results;

[0061] The fault solution generation module monitors and collects real-time operation data of power equipment according to the power load forecast results, analyzes abnormal load fluctuations, integrates historical equipment maintenance records and fault database information, formulates target maintenance plans and preventive measures, and outputs fault solution plans.

[0062] Knowledge base update records include new term names, term definitions, term matching degrees and recognition thresholds; security monitoring logs include identity authentication results, data encryption levels, data operation records and abnormal access marking results; power load forecasting results include time fluctuation feature records, regional load distribution results, meteorological influencing factors and user behavior pattern analysis records; and troubleshooting solutions include fault types, maintenance suggestions, preventive measures and equipment maintenance records.

[0063] See also Figure 2 , the matching analysis steps are:

[0064] Retrieve newly released power industry documents from online databases and power industry websites, screen for target keywords, and generate a preliminary document list;

[0065] Document searches are conducted on the Internet and official databases, and a keyword screening method is used to preliminarily screen documents published in the past week. For example, relevant power industry documents are located through keywords such as "power transmission" and "renewable energy". This process is achieved through automated search engines and database query functions. The goal is to capture all documents that may contain innovations and policy updates in the power industry, ensuring that the collected data sets are the latest and most relevant. During this data collection process, the automated system will record the publication date, source, and keyword frequency of each document to verify its relevance to the power industry and generate a preliminary document list.

[0066] Based on the preliminary document list, extract key terms and contexts to ensure that the context of each term is accurately captured and stored in a temporary database to obtain a detailed set of terms and their definitions;

[0067] Natural language processing tools are used to continue extracting and analyzing documents from the preliminary document list, focusing on the definitions of terms in the documents and their contexts. The process includes natural language processing algorithms such as semantic analysis and word meaning association analysis. The system is able to identify and parse the special terms and their definitions in the power industry. Each term and the corresponding context will be extracted and form a structured term library. The data in this library includes the term itself, its definition, the context in which it appears, and the document location where the term first appears, ensuring that the collected term data is both complete and easy to query, and obtaining a detailed set of terms and their definitions.

[0068] Based on the detailed terminology and its definition set, compared with the existing knowledge base, the formula is adopted. ;

[0069] Calculate the matching degree of each new term with the existing knowledge base terms , generate matching analysis results, where Represents the weight factor, which is used to adjust the influence of each term feature. and Represent the term vectors in the existing knowledge base and the newly extracted term vectors respectively, and Respectively represent the modulus length of the existing knowledge base and the newly extracted term vector;

[0070] set up , get , , the calculation process is as follows:

[0071] ;

[0072] ;

[0073] The results show that the newly extracted terms match the terms in the knowledge base at a degree of 0.3536. The matching calculation helps to identify and classify terms, and further decide which new terms need to be recorded in detail or updated in the knowledge base.

[0074] See also Figure 3 , the steps to obtain the knowledge base update record are:

[0075] Based on the matching analysis results, a judgment threshold is set to identify terms with matching degrees lower than the judgment threshold, determine them as new terms, and generate a list of unrecorded new terms;

[0076] Based on the matching analysis results obtained previously, the terms with matching degrees lower than the judgment threshold are screened out. These terms are considered to be new terms that have not yet been entered into the knowledge base (the judgment threshold is set to 0.5. Combined with the above calculation results, since 0.3536 is less than 0.5, they are determined to be new terms). Then, by setting a threshold, those important but not yet widely recognized or used terms can be filtered out to generate a list of unentered terms that need further verification and analysis to ensure that these terms are new concepts in the power industry that have not yet been fully covered by the existing knowledge base.

[0077] Analyze the list of unrecorded new terms, verify the application scenarios and technical accuracy of each new term based on the application importance of the power industry and industry standard documents, conduct document comparison and expert review, and obtain a verified new term set;

[0078] An in-depth study was conducted on the list of new terms that had not been entered. The research work included a detailed investigation of the technical background, application areas and industry importance of each term. In this process, standard documents and technical reports of the power industry were used as the main reference materials. The expert team discussed through online and offline meetings to evaluate the technical meaning and application scope of each new term to ensure that each term was accurately interpreted and correctly classified, and the application scenarios and technical accuracy of the terms were fully verified. This process ensured the quality and practicality of the knowledge base content, thereby obtaining a verified set of new terms.

[0079] Synchronously update the verified new term sets to the Q&A knowledge base, update the index and search functions of the knowledge base, record the addition time and document source of each term, and generate a knowledge base update record;

[0080] The verified new term set is synchronously stored in the question-and-answer knowledge base. The update process includes the reconstruction of the knowledge base index and the optimization of the search function to ensure that the newly added terms can be efficiently retrieved. This storage process adopts database transaction management and data consistency guarantee mechanism. The addition of each new term is accompanied by detailed metadata records, including the source of the term, the time of addition and the associated documents. Through the update and data management of these systems, the knowledge base not only expands the content, but also improves the accuracy and response speed of data retrieval, and finally generates a detailed knowledge base update record with timestamp and source reference.

[0081] See also Figure 4 , the execution steps of the layered encryption process are:

[0082] Extract data request information from knowledge base update records, including requester ID and source address, determine the legitimacy and integrity of the request, accurately track each request, and generate a request review file;

[0083] Before any data operation is performed, the data of each request is captured in detail through the update record of the knowledge base. This data includes the identity of the requester and the IP address of the request source. The legitimacy of each request is verified through the access control list (ACL) and IP whitelist. It not only involves traditional username and password verification, but also includes digital certificates and two-factor authentication mechanisms to ensure that only authorized users can access sensitive data and ensure the confidentiality and integrity of information transmitted in the network, thereby providing a secure foundation for the audit records of data requests.

[0084] Based on the request review file, use security protocols and verification mechanisms to verify the identity and source of the requester, use multi-factor authentication and source address checks to confirm the validity of the request, and obtain security verification pass information;

[0085] After confirming the legitimacy of the data requester, the source of the requested data is checked by analyzing the source field in the packet header to determine whether the request comes from a trusted network or geographic location. Dynamic access control lists and behavioral analysis techniques are then used to further evaluate whether the behavior of the request conforms to the user's normal pattern, such as frequent login attempts or data access during non-working hours, thereby establishing the necessary security audit for the encryption processing process.

[0086] Perform multi-layer encryption on information that has passed security verification, use AES and RSA encryption standards to strengthen data protection, optimize data security during transmission and storage, and generate data encryption processing results;

[0087] Requests that have been authenticated and source checked will be subject to layered encryption processing, which includes transport layer encryption and database encryption of data. The data is encrypted using advanced encryption standards (AES and RSA) to ensure that even if the data is intercepted during transmission, the information cannot be read by unauthorized persons. In addition, field-level encryption is used for sensitive fields such as personal identity information and financial records to enhance protection and ensure the confidentiality and integrity of the data. All encryption operations are performed on the encryption server, and the results of the operations will be recorded in the encrypted processing result data to provide a basis for subsequent security audits and compliance reports.

[0088] See also Figure 5 , the steps to obtain the security monitoring log are:

[0089] According to the data encryption processing results, record the details of each data operation, including the operator, operation type and timestamp and its layered encryption results, ensure that all data interactions are tracked and recorded, and generate operation record logs;

[0090] According to the data encryption processing results, a detailed record of each data operation is obtained, including the operator identity, operation type and timestamp. The layered encryption processing results can also be monitored. Through the relevant data, a detailed operation log is established to ensure that every data transmission and access is traceable. In addition, the record can help system administrators track possible data leakage paths or illegal access behaviors. Furthermore, by analyzing the access frequency and data volume changes in the operation log, the system load changes can be predicted, and the data storage and access strategies can be optimized. The operation record log finally generated records the activities of each user in detail, improves the transparency of the system, and strengthens the protection measures for sensitive data.

[0091] Analyze operation logs to identify normal and abnormal patterns of data access, using formulas, ;

[0092] Calculate standard statistics , set the abnormality judgment value ,like Value greater than , then the access behavior is judged to be abnormal, and the pattern recognition analysis results are obtained, among which, represents a data point for a single visit, Represents the average value of historical data access, Represents the variance of historical data;

[0093] Use anomaly detection formula to analyze data access patterns. In the formula, A data point representing a single visit is the actual visit value extracted from the log file. Represents the calculated average of historical data accesses, obtained by accumulating data points and averaging them. is the variance of the historical data, which is also obtained by calculating the average of all data points and the average of the square of the difference of each point; when When the value exceeds 3 (Y value), it is considered that the data point is significantly different from the norm, indicating that the access may be abnormal behavior;

[0094] Get a data access record , historical average ,variance ,but The calculation is as follows:

[0095] ;

[0096] This result does not exceed 3, so this visit is not marked as abnormal;

[0097] Based on the results of pattern recognition analysis, abnormal behaviors are identified and marked, and alerts are generated synchronously to obtain security monitoring logs;

[0098] Using the identified data access patterns, abnormal access behaviors are further marked, and the behaviors are recorded and analyzed in detail, such as multiple consecutive login failures, unauthorized data access attempts, or unusual query behaviors on sensitive data. Each anomaly will be automatically recorded and marked in the security monitoring log by the system. In addition, by continuously monitoring these logs, the security team can detect potential threats in a timely manner. The generated security alerts not only help to respond to these threats immediately, but also provide data support for adjusting future security policies, thereby continuously improving the system's protection capabilities, ensuring the security of all critical data, and comprehensively outputting security monitoring logs.

[0099] See also Figure 6 , the prediction steps of power demand fluctuation are:

[0100] Based on the security monitoring log, filter all log data related to power load changes, call the timestamp and load value fields in the log, verify the integrity and consistency of the data, and generate securely filtered power load data;

[0101] Events related to power load are extracted from security monitoring logs. During the extraction process, the security event log data directly related to the power load changes in the logs are first screened out. These log data include key fields such as timestamp, load value, and event type. Timestamp can help match the time series of load data and other data sources. The load value can be used for subsequent demand fluctuation prediction. The event type can determine whether the load change is related to certain specific external events. During the log data extraction process, the parameters of the security mode are used for verification to ensure the integrity and consistency of the data. Specific operations include timestamp comparison and logical verification of load values ​​(such as avoiding sudden changes in load data). Finally, the log data filtered by the security mode is used as the basic input for subsequent analysis to obtain a securely filtered power load data set.

[0102] The power load data after security filtering is integrated with meteorological information and user behavior statistics, matched according to time series, and the load, meteorological and user behavior data of different time periods are formed into a unified time series table to generate an integrated multivariate time series data set;

[0103] The obtained securely filtered power load data is integrated with meteorological information and user behavior statistics. First, the temperature, humidity and other data in the meteorological information are extracted. These meteorological data are recorded in the form of time series, and each time point corresponds to a specific temperature and humidity value. At the same time, the user's consumption behavior and electricity usage pattern are extracted from the user behavior statistics. The behavior data also has a timestamp attribute, so it can be synchronized with the load data and meteorological data. Through the comparison and matching of timestamps, the data in different data sources are associated and processed based on time to ensure that the power load, meteorological information and user behavior in each period can correspond to an accurate time point, forming a unified time series table, and finally generating an integrated multivariate time series data set.

[0104] Call the multivariate time series data set and perform multivariate time series analysis using the formula, ;

[0105] Calculate electricity demand fluctuation forecasts , analyze the fluctuation range of power load and generate power demand fluctuation prediction results, among which, For the The power load at a given moment, for The weight correction factor of the momentary load, For time The meteorological factors within For the User behavior statistics at a certain moment, is the user behavior correction coefficient, Indicates the number of moments;

[0106] The formula is used to calculate the predicted value of power demand fluctuation at a specific moment. =1200 (unit: kW), =1400, =1.2, =1.1, set =4000, collection =800, )=600, set =0.05, =0.07, substitute it into the formula and perform the calculation.

[0107] Calculate the numerator part:

[0108] ;

[0109] Calculate the denominator:

[0110] ;

[0111] ;

[0112] ;

[0113] Calculation formula results:

[0114] ;

[0115] The results show that at the current time point The predicted value of electricity demand fluctuation is 40.27. The larger the value, the more severe the power load fluctuation.

[0116] See also Figure 7 , the steps to obtain the power load forecast results are:

[0117] Based on the power demand fluctuation prediction results, the demand fluctuation value of each time period is extracted, and the formula is used: ;

[0118] Calculate the average volatility for each time period , and compare them one by one with the set volatility standard, divide the time periods into three categories: high volatility, medium volatility, and low volatility, and generate time difference demand volatility groups, among which, The power demand fluctuation during this period, is the total time of the period, Indicates the total time period;

[0119] The formula is used to calculate the average power fluctuation amplitude in each time period, and the collected power demand fluctuation values ​​are =200kW, =150kW, =250kW, the time periods are =2 hours, =1 hour, =3 hours, substitute into the formula and perform the calculation.

[0120] Calculate the numerator part:

[0121] ;

[0122] Calculate the denominator:

[0123] ;

[0124] Calculate the average volatility:

[0125] ;

[0126] The results show that the average fluctuation of power demand during this period is 100kW. The fluctuation standard is high fluctuation > 120kW, medium fluctuation between 80-120kW, and low fluctuation < 80kW. 100kW falls within the medium fluctuation range, so this period is classified as medium fluctuation.

[0127] Based on the time difference demand fluctuation grouping, the regional power load value is extracted, and the formula is used. ;

[0128] Calculate the correlation score between the regional average load value and the demand fluctuation category during different periods , determine the load performance characteristics of each region in different time periods, and generate regional demand fluctuation groups, where, is the average load value for each area, is the time category, For time The regional load characteristic value under is the load correction factor;

[0129] The formula is used to calculate the correlation between the area and the time period. The power load of area A in a certain time period is =1200kW, =1500kW, the time periods are =2 hours, = 3 hours, =1100, =1300, the correction factor is =1.1, =1.2, substitute it into the formula and perform the calculation.

[0130] Calculate the numerator part:

[0131] ;

[0132] Next, calculate the denominator:

[0133] ;

[0134] ;

[0135] Calculate the relevance score:

[0136] ;

[0137] The results show that the correlation between the power load in area A and the time period is 131.1, indicating that the power load in this area varies greatly during the time period and is closely related to the load performance in the time period.

[0138] Combined with the time-difference demand fluctuation grouping and the regional demand fluctuation grouping, feature extraction is performed, the impact of grouping characteristics on power load is analyzed, and power load forecast results are generated.

[0139] Call the time difference demand fluctuation grouping and regional demand fluctuation grouping. First, extract the power load data in each group from the time difference grouping, and divide it into multiple time periods for summary. During the extraction process, the power load of each time period needs to be standardized to ensure that the data is comparable in different groups. Then, extract the meteorological data and user behavior data in each area from the regional demand grouping. The specific meteorological data includes parameters such as temperature, humidity, and wind speed. These data come from the regional meteorological monitoring system, and the user behavior data is mainly obtained from the user's electricity consumption records, including daily electricity consumption and electricity consumption patterns. Based on these extracted data, feature extraction is performed on the time and regional load fluctuations in each group. The feature extraction process includes cross-comparison of load data, meteorological data, and user behavior data. The data feature values ​​of each group are extracted and summarized one by one to generate a feature vector of the power load. Finally, based on the extracted feature vector, the load prediction characteristics of each group are obtained, and finally the power load prediction results are generated.

[0140] See also Figure 8 , the steps to obtain the troubleshooting solution are:

[0141] Based on the power load forecast results, the real-time operation data of the power equipment is called, the equipment operation load at each moment is compared with the predicted load value, the deviation value between the real-time operation load and the predicted load is calculated, the time point when the deviation exceeds the deviation threshold is extracted, and the load abnormality time mark is generated;

[0142] Based on the power load forecast results, the real-time operation data of the power equipment is called, and the equipment operation load at each moment is compared with the predicted load value. During the extraction process, the current operation load value of the equipment is first obtained from the real-time monitoring system of the power equipment (the data is recorded at intervals of seconds or minutes), and then the load forecast results are used to extract the predicted load value at each moment, and a one-to-one comparison is performed to calculate the load deviation value at each moment (the load deviation is calculated by subtracting the predicted load value from the real-time load). Then, it is determined whether these deviation values ​​exceed the pre-set deviation threshold, such as comparing the deviation value with the prediction error standard of ±5%, extracting the moment when the deviation exceeds this range, and generating a load abnormality moment mark. The mark will be used to subsequently locate the period of potential equipment abnormality to help preliminarily determine possible equipment operation problems.

[0143] According to the load abnormality time mark, it is cross-compared with the historical maintenance records of the equipment and the information in the fault database. Through similar fault cases and maintenance conditions in the historical records, the equipment fault characteristics related to the current load abnormality are extracted. Combined with the equipment model, maintenance time and fault type, the preliminary fault diagnosis analysis results are generated;

[0144] Call the load abnormality time mark and cross-check it with the equipment's historical maintenance records and the information in the fault database. First, extract the equipment's historical maintenance records from the equipment maintenance system, including past maintenance time, maintenance type, specific fault conditions and other detailed information. Then call the relevant data in the fault database to query other equipment failures of similar models to the equipment, especially those fault types that have been associated with abnormal load fluctuations. Compare the equipment history records and similar fault cases in the fault database with the current load abnormality mark to extract the characteristics of similar faults, such as overheating, short circuit or equipment aging problems caused by abnormal load fluctuations. Combined with parameters such as equipment model, maintenance time, fault type, etc., generate preliminary analysis results for fault diagnosis, and combine with historical data to further narrow the scope of equipment failures and identify possible fault types of the current equipment.

[0145] Based on the preliminary analysis results of fault diagnosis, the formula is used. ;

[0146] Calculate the comprehensive evaluation index of fault , formulate target maintenance plans and preventive measures, and output troubleshooting solutions, including: For the The severity of the fault, is the fault impact range value, The device's historical fault repair time. Prioritize preventive measures. is the correction factor for preventive measures, Indicates the fault type.

[0147] The formula is used to calculate the comprehensive evaluation index to measure the severity of the current equipment failure, the scope of impact, the historical repair time and the priority of preventive measures. This value reflects the overall priority and processing complexity of the equipment failure. =8 (represents the severity of device overheating), =6 (representing the severity of circuit aging), =0.7 (overheating affects 30% of the load capacity of the equipment), =0.5 (circuit aging affects 50% of the device's power transmission capacity), = 4 days, =0.8, =0.6, =0.1, =0.05, substitute it into the formula and perform the calculation.

[0148] Calculate the numerator part:

[0149] ;

[0150] Next, calculate the denominator:

[0151] ;

[0152] ;

[0153] ;

[0154] Calculate the comprehensive evaluation index:

[0155] ;

[0156] The results show that the maintenance plan gives priority to more serious faults with a wider impact. A higher value means that the fault has a greater weight in terms of severity, impact range or repair difficulty, has a higher priority, and needs to be repaired immediately. On the contrary, if the value is low, it means that the fault is relatively minor or can be alleviated by preventive measures. Combining historical maintenance records and the weight of preventive measures, the optimal fault resolution plan is finally output to ensure the normal operation of the equipment.

[0157] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The power knowledge question-answering system based on language model is characterized by: The system comprises: The model terminology dynamic update module collects newly released power industry documents, extracts terms and their contextual definitions, compares existing terms in the knowledge base and analyzes the matching degree. According to the matching degree analysis results, it sets a judgment threshold to identify new terms that have not been entered, and stores them synchronously in the question-and-answer knowledge base to obtain the knowledge base update record; The data security monitoring module uses the knowledge base to update records, performs identity and source verification on each data request, and performs layered encryption processing, records each data operation based on the layered encryption processing results, analyzes data access patterns, identifies and marks abnormal access behaviors, and obtains security monitoring logs; The load forecasting and analysis module collects and integrates power load data, meteorological information and user behavior statistics based on the data security mode in the security monitoring log, predicts power demand fluctuations through multivariate time series analysis, groups demand fluctuations of different times and regions according to the power demand fluctuation prediction results, and performs feature extraction to generate power load forecast results; The fault solution generation module monitors and collects real-time operation data of power equipment according to the power load forecast results, analyzes abnormal load fluctuations, integrates historical equipment maintenance records and fault database information, formulates target maintenance plans and preventive measures, and outputs fault solution plans; The steps for predicting the power demand fluctuation are: Based on the security monitoring log, all log data related to power load changes are screened, the timestamp and load value fields in the log are called, the integrity and consistency of the data are verified, and the power load data after security filtering is generated; Integrate the securely filtered power load data with meteorological information and user behavior statistics, match them according to time series, form a unified time series table for the load, meteorological and user behavior data of different time periods, and generate an integrated multivariate time series data set; The integrated multivariate time series data set is called to perform multivariate time series analysis using the formula: ； Calculate electricity demand fluctuation forecasts , analyze the fluctuation range of power load and generate power demand fluctuation prediction results, among which, For the The power load at a given moment, for The weight correction factor of the momentary load, For time The meteorological factors within For the User behavior statistics at a certain moment, is the user behavior correction coefficient, Indicates the number of moments; The steps for obtaining the power load forecast result are: Based on the power demand fluctuation prediction results, the demand fluctuation value of each time period is extracted using the formula: ； Calculate the average volatility for each time period , and compare them one by one with the set volatility standard, divide the time periods into three categories: high volatility, medium volatility, and low volatility, and generate time difference demand volatility groups, among which, The power demand fluctuation during this period, is the total time of the period, Indicates the total time period; Based on the time difference demand fluctuation grouping, the regional power load value is extracted using the formula: ； Calculate the correlation score between the regional average load value and the demand fluctuation category during different periods , determine the load performance characteristics of each region in different time periods, and generate regional demand fluctuation groups, where, is the average load value for each area, is the time category, For time The regional load characteristics under is the load correction factor; Combined with the time difference demand fluctuation grouping and the regional demand fluctuation grouping, feature extraction is performed, the impact of the grouping characteristics on the power load is analyzed, and the power load prediction result is generated.

2. The power knowledge question-answering system based on language model according to claim 1, characterized in that: The analysis steps of the matching degree are as follows: Retrieve newly released power industry documents from online databases and power industry websites, screen for target keywords, and generate a preliminary document list; Based on the preliminary document list, extract key terms and contexts to ensure that the context of each term is accurately captured, and store them in a temporary database to obtain a detailed terminology and definition set; Based on the detailed terms and their definition set, compared with the existing knowledge base, the formula is adopted: ； Calculate the matching degree of each new term with the existing knowledge base terms , generate matching analysis results, where Represents the weight factor, which is used to adjust the influence of each term feature. and Represent the term vectors in the existing knowledge base and the newly extracted term vectors respectively, and Respectively represent the modulus of the existing knowledge base and the newly extracted term vector.

3. The power knowledge question-answering system based on language model according to claim 2 is characterized in that: The steps for obtaining the knowledge base update record are: Based on the matching analysis result, a judgment threshold is set, terms with matching degrees lower than the judgment threshold are identified and determined as new terms, and a list of unentered new terms is generated; Analyze the list of unrecorded new terms, verify the application scenario and technical accuracy of each new term based on the application importance of the power industry and industry standard documents, conduct document comparison and expert review, and obtain a verified new term set; The verified new term set is synchronously updated to the question-and-answer knowledge base, the index and search functions of the knowledge base are updated, and the addition time and document source of each term are recorded to generate a knowledge base update record.

4. The power knowledge question-answering system based on language model according to claim 3 is characterized by: The execution steps of the layered encryption process are: Extracting data request information from the knowledge base update record, including the requester identifier and source address, determining the legitimacy and integrity of the request, accurately tracking each request, and generating a request review file; Based on the request review file, the identity and source of the requester are verified using security protocols and verification mechanisms, and the validity of the request is confirmed using multi-factor identity authentication and source address checks to obtain security verification pass information; The security verification information is encrypted through multiple layers, and the AES and RSA encryption standards are used to strengthen data protection, optimize the security of data in transmission and storage, and generate data encryption processing results.

5. The power knowledge question-answering system based on language model according to claim 4 is characterized in that: The steps for obtaining the security monitoring log are: According to the data encryption processing result, record the details of each data operation, including the operator, operation type and timestamp and its hierarchical encryption result, ensure that all data interactions are tracked and recorded, and generate an operation record log; Analyze the operation record log to identify normal and abnormal patterns of data access, using the formula: ； Calculate standard statistics , set the abnormality judgment value ,like Value greater than , then the access behavior is judged to be abnormal, and the pattern recognition analysis results are obtained, among which, represents a data point for a single visit, Represents the average value of historical data access, Represents the variance of historical data; According to the pattern recognition analysis results, abnormal behaviors are identified and marked, and alarms are generated synchronously to obtain security monitoring logs.

6. The power knowledge question-answering system based on language model according to claim 1 is characterized in that: The steps for obtaining the troubleshooting solution are: Based on the power load forecast result, the real-time operation data of the power equipment is called, the equipment operation load at each moment is compared with the forecast load value, the deviation value between the real-time operation load and the forecast load is calculated, the time point when the deviation exceeds the deviation threshold is extracted, and the load abnormality time mark is generated; According to the load abnormality time mark, it is cross-compared with the historical maintenance records of the equipment and the information in the fault database. Through similar fault cases and maintenance conditions in the historical records, the equipment fault characteristics related to the current load abnormality are extracted, and combined with the equipment model, maintenance time and fault type, a preliminary fault diagnosis analysis result is generated; Based on the preliminary analysis results of the fault diagnosis, the formula is used: ； Calculate the comprehensive evaluation index of fault , formulate target maintenance plans and preventive measures, and output troubleshooting solutions, including: For the The severity of the fault, is the fault impact range value, The device's historical fault repair time. Prioritize preventive measures. is the correction factor for preventive measures, Indicates the fault type.

Images