An electric power question and answer result fusion tracing method, device, medium and equipment

Abstract

The application discloses the technical field of power system intelligent question and answer and data processing, and relates to a power question and answer result fusion and tracing method, device, medium and equipment. The method comprises the following steps: structurally analyzing a received user power natural language question and answer request to obtain structured task information; acquiring multi-source results according to the structured task information; converting the multi-source results into unified result units; performing consistency verification on the unified result units to obtain verification results; performing fusion processing on the unified result units according to the verification results and preset weight rules to obtain fusion results; generating fusion interpretation results according to the fusion results; and generating credible tracing information according to the structured task information, multi-source results, unified result units, verification results and fusion results. The application can improve the fusion stability, interpretability and checkability of power question and answer results, and is convenient for business review, audit and review.

Description

Technical Field

[0001] This invention relates to the field of intelligent question-and-answer and data processing technology for power systems, specifically to a method, device, medium, and equipment for fusion and tracing of power question-and-answer data results. Background Technology

[0002] With the development of digital and intelligent power system construction, a large amount of knowledge documents, indicator data, operational data, alarm records, historical cases, and business rules have been accumulated in business scenarios such as dispatching, operation and maintenance, repair, marketing, and equipment management. In order to improve the efficiency of business personnel in obtaining information and analyzing problems, more and more power business systems are beginning to introduce natural language question-and-answer capabilities, enabling users to ask business questions through natural language and obtain knowledge question-and-answer results, indicator query results, data analysis results, or auxiliary handling suggestions.

[0003] Existing power industry natural language question-answering and data systems typically include capabilities such as knowledge-based question answering, indicator querying, data service invocation, tool service invocation, and rule-based judgment. For example, for equipment status-related questions, the system may need to retrieve equipment procedures, query real-time operating data, invoke status diagnostic tools, and combine them with threshold rules for judgment; for alarm handling-related questions, the system may need to retrieve handling manuals, query current alarm records, invoke alarm analysis tools, and combine them with business rules to generate handling suggestions; for indicator analysis-related questions, the system may need to query indicator definitions, obtain historical statistical data, and interpret them using relevant knowledge.

[0004] However, in existing systems, the results of knowledge-based question answering, indicator querying, data service, tool invocation, and rule-based judgment are typically generated separately by different modules, resulting in differences in data structure, return format, time range, indicator definitions, and conclusion expression. Knowledge bases usually return text fragments, indicator bases typically return indicator definitions, calculation methods, or threshold rules, data services typically return tabular data, time-series data, or statistical results, tool services typically return analysis reports or diagnostic conclusions, and rule engines typically return judgment results or disposal rules. Without a unified method for organizing results, the system often has to directly piece together content from different sources or rely on a generative model to summarize it, which can easily lead to answers that lack focus, inconsistent explanations, or unclear basis.

[0005] Furthermore, conflicts or inconsistencies may exist between results from different sources. For example, the business object in the user's question may not match the object corresponding to the data query result; the threshold in the knowledge document may not match the threshold in the indicator library; the data query time range may not match the user's actual intent; the source or statistical period of real-time data may not match that of historical statistical data; and the conclusions of tool analysis may not match the conclusions of rule judgment. If the system fails to identify and verify these differences before generating the final answer, it may lead to incorrect fusion, incorrect interpretation, or unreliable conclusions.

[0006] Meanwhile, the power business scenario demands high verifiability and traceability of the answers. Business personnel not only need to know the conclusions given by the system, but also need to understand which knowledge documents were used as the basis for those conclusions, which indicators were queried, what query conditions were used, which tools were invoked, what rules were adopted, and how the conclusions were formed. Existing question-and-answer systems often focus on providing natural language answers, lacking complete records of knowledge sources, data sources, indicator definitions, tool call parameters, rule basis, and integration process, making it difficult to meet the requirements for subsequent review, auditing, and post-mortem analysis.

[0007] Therefore, a result processing mechanism is needed for the power industry natural language question-answering scenario, which unifies the organization, consistency verification, fusion interpretation, and reliable traceability of multi-source results to improve the accuracy, interpretability, and verifiability of the final answer. (Summary of the Invention)

[0008] The purpose of this invention is to overcome the problems of fragmented sources, inconsistent formats, lack of consistency verification, unstable fusion interpretation, and difficulty in tracing the basis of answers in existing power natural language question-and-answer systems, which involve knowledge question-and-answer results, indicator query results, data service results, tool call results, and rule judgment results. This invention provides a method, device, medium, and equipment for the fusion and tracing of power question-and-answer results, which uniformly organizes, verifies, fuses, interprets, and traces results from multiple sources, thereby improving the accuracy, interpretability, and verifiability of power business question-and-answer results.

[0009] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution.

[0010] In a first aspect, the present invention provides a method for fusing and tracing the results of power-related question-and-answer sessions, including:

[0011] The received user electricity natural language question-and-answer request is parsed in a structured manner to obtain structured task information;

[0012] Obtain multi-source results based on the structured task information;

[0013] The multi-source results are converted into a unified result unit;

[0014] The consistency of the unified result unit is checked to obtain the check result;

[0015] The unified result unit is fused according to the verification result and the preset weighting rules to obtain the fusion result;

[0016] A fusion interpretation result is generated based on the fusion result;

[0017] Trustworthy traceability information is generated based on the structured task information, multi-source results, unified result units, verification results, and fusion results.

[0018] Optionally, the structured task information includes at least one of the following: business intent, business object, indicator conditions, time range, query conditions, statistical method, business scenario, and expected output type.

[0019] The business intent is used to characterize the type of user's power natural language question and answer request, which is either knowledge-based question and answer, indicator query, status judgment, cause analysis, alarm handling, trend analysis, or report generation.

[0020] The business object is used to represent the power plant, line, transformer, protection device, energy storage station, battery cluster, communication link or telemetry point corresponding to the user's power natural language question and answer request;

[0021] The indicator conditions are used to characterize the indicator name, indicator alias, indicator definition, or threshold rule that needs to be queried or judged.

[0022] The time range is used to characterize the real-time, current day, past week, past month, or custom time period involved in the user's electricity natural language question and answer request;

[0023] The query criteria are used to characterize the device type, voltage level, region, alarm type, statistical period, or filtering conditions.

[0024] Optionally, the multi-source results include at least two of the following: knowledge retrieval results, indicator query results, data service results, tool call results, and rule judgment results;

[0025] The step of obtaining multi-source results based on the structured task information includes:

[0026] If the structured task information includes knowledge question-and-answer intent, obtain knowledge retrieval results from the domain knowledge base, procedure document library, equipment manual library, or fault case library;

[0027] If the structured task information includes an indicator query or indicator judgment intent, obtain the indicator query results from the indicator library, indicator caliber library, or threshold rule library;

[0028] If the structured task information includes the intent to query running data, obtain the data service results from the real-time database, historical database, time-series data service, or unified model query service;

[0029] If the structured task information includes diagnostic, analytical, or report generation intent, invoke alarm analysis tools, status diagnostic tools, report generation tools, or algorithm models, and obtain the tool invocation results;

[0030] If the structured task information includes threshold judgment, handling suggestions, or compliance verification intent, call the business rules, threshold rules, verification rules, or handling rules to obtain the rule judgment result.

[0031] Optionally, the unified result unit includes result source type, result source identifier, business object, indicator name, time range, data value, conclusion text, source basis, confidence level, and traceability information;

[0032] The result source type is used to identify whether the result comes from knowledge retrieval, indicator query, data service, tool call, or rule judgment;

[0033] The result source identifier is used to identify the knowledge base number, document number, indicator library number, data service number, tool number, or rule number;

[0034] The business object is used to identify the plant, line, equipment, component, protection device, energy storage station, battery cluster, communication link or telemetry point corresponding to the result;

[0035] The indicator name is used to identify the indicator, indicator alias, or indicator code corresponding to the result;

[0036] The time range is used to identify the start and end time, statistical period, data timestamp, or generation time of the result.

[0037] The data values ​​are used to record numerical values, tables, time-series data, or statistical results obtained from data services, indicator queries, or tool analysis.

[0038] The conclusion text is used to record the conclusive content formed by knowledge retrieval, tool analysis, or rule judgment;

[0039] The source basis is used to record knowledge fragments, document locations, indicator definitions, rule clauses, data table names, or tool return bases;

[0040] The confidence level is used to indicate the degree of credibility of the result unit in the current task;

[0041] The traceability information is used to record the result generation chain, source path, call record, and generation time;

[0042] The unified result unit also includes at least one of the following: unit, scope of application, generation time, calling parameters, and exception information;

[0043] The unit is used to record the unit of measurement corresponding to the indicator value or data value;

[0044] The scope of application is used to indicate the business scenarios, equipment types, voltage levels, or data ranges to which the result applies.

[0045] The calling parameters are used to record data query conditions, tool calling parameters, rule execution parameters, or model input parameters;

[0046] The exception information is used to record data missing, call failure, insufficient permissions, low confidence or conflict markers;

[0047] The step of converting the multi-source results into a unified result unit includes:

[0048] Convert knowledge retrieval results into knowledge-based result units;

[0049] And / or,

[0050] Convert the indicator query results into indicator-type result units;

[0051] And / or,

[0052] Transform data service results into data-type result units;

[0053] And / or,

[0054] Convert tool call results into tool-type result units;

[0055] And / or,

[0056] Convert the rule judgment result into a rule-type result unit.

[0057] Optionally, the consistency verification includes business object consistency verification, time range consistency verification, indicator caliber consistency verification, data source consistency verification, and conclusion consistency verification;

[0058] The process of performing consistency verification on the unified result unit to obtain the verification result includes:

[0059] Determine whether the business object corresponding to the unified result unit is consistent with the business object in the structured task information: if they are inconsistent, generate an object conflict flag and the verification fails; if they are consistent, the verification passes.

[0060] Determine whether the time range corresponding to the unified result unit is consistent with the time range in the structured task information: if they are inconsistent, generate a time conflict flag and the verification fails; if they are consistent, the verification passes.

[0061] Determine whether the indicator name, calculation method, unit, threshold, or statistical period corresponding to the unified result unit are consistent: if inconsistent, generate an indicator caliber conflict flag and the verification fails; if consistent, the verification passes.

[0062] Determine whether the data collection source, statistical source, or update time corresponding to different data sources in the unified result unit are consistent: if they are inconsistent, a data source conflict flag is generated and the verification fails; if they are consistent, the verification passes.

[0063] Determine whether the conclusions among the knowledge basis, data results, tool analysis results, or rule judgment results in the unified result unit are consistent: if inconsistent, generate a conclusion conflict flag and the verification fails; if consistent, the verification passes.

[0064] The consistency verification also includes at least one of data integrity verification, permission availability verification, and call status verification;

[0065] The verification result includes at least one of the following: verification passed, verification failed, conflict type, conflict field, conflict source, conflict description, and handling suggestion;

[0066] The conflict types include at least one of the following: object conflict, time conflict, indicator conflict, data source conflict, conclusion conflict, missing data, low confidence, unavailable source, and abnormal tool invocation.

[0067] Optionally, the preset weighting rules include at least one of the following: source credibility, object matching degree, time matching degree, indicator caliber matching degree, data integrity, time freshness, rule priority, tool call success rate, historical feedback quality, and manual confirmation status.

[0068] The source credibility is used to indicate the reliability of the source of the result;

[0069] The object matching degree is used to represent the degree of matching between business objects in the unified result unit and business objects in the structured task information.

[0070] The time matching degree is used to represent the degree of matching between the time range in the unified result unit and the time range requested by the user;

[0071] The indicator caliber matching degree is used to indicate the degree of consistency between the indicator name, calculation method, unit, threshold, and statistical period in the unified result unit and the caliber of the indicator library;

[0072] The data integrity refers to whether the key fields, data values, source basis, and calling parameters in the unified result unit are complete.

[0073] The time freshness is used to represent the time interval between the result generation time or data update time and the current task;

[0074] The rule priority is used to indicate the priority order among different business rules or threshold rules;

[0075] The tool call success rate is used to indicate the availability of the tool service in the current or historical tasks;

[0076] The historical feedback quality is used to represent the evaluation of similar results sources by historical user feedback or manual confirmation;

[0077] The manual confirmation status is used to indicate whether the result has been confirmed by business personnel;

[0078] The fusion results include at least one of the following: main conclusion basis, auxiliary conclusion basis, conflict explanation, low confidence explanation, and explanation of not using basis;

[0079] When multiple unified result units have consistent conclusions and the credibility of their sources meets the requirements, they are used together as the basis for the main conclusion.

[0080] When a unified result unit is consistent with the direction of the main conclusion but the credibility of its source does not meet the requirements, it shall be used as a basis for auxiliary conclusions.

[0081] When there is a conflict between the unified result unit and the main conclusion, record the explanation of the conflict;

[0082] When the source credibility of a unified result unit does not meet the requirements, record a low confidence level description;

[0083] When a unified result unit has missing data, an unavailable source, or an abnormal tool call, the record will not be used based on the explanation.

[0084] And / or,

[0085] The integrated interpretation results include direct conclusions, data basis, explanation of indicator definitions, knowledge basis, cause analysis, handling suggestions, and risk warnings;

[0086] The direct conclusions are used to answer user questions;

[0087] The data cited is used to illustrate the data values, statistical results, or trend changes corresponding to the conclusions.

[0088] The description of the indicator scope is used to explain the name of the indicator, calculation method, unit, threshold or statistical period used;

[0089] The knowledge base mentioned is used to explain the referenced procedural documents, equipment manuals, or case knowledge;

[0090] The causal analysis is used to explain the possible reasons for reaching the conclusion;

[0091] The proposed actions are used to provide suggestions for inspection, review, processing, or follow-up operations.

[0092] The risk warning is used to explain the confidence level, missing data, conflict markers, or matters requiring manual confirmation.

[0093] Optionally, the trusted traceability information includes at least one of the following: original question, structured task information, result source information, query conditions, query statement, data source, indicator caliber, knowledge source, tool call information, rule basis, unified result unit information, verification result, conflict information, fusion weight, and conclusion generation record;

[0094] And / or,

[0095] It also includes: optimizing at least one of the following based on user feedback, manual confirmation results, and anomaly records: entity recognition rules, indicator matching rules, indicator alias library, caliber mapping rules, conflict verification rules, weighting rules, interpretation templates, traceability fields, and tool invocation strategies; and writing at least one of the following as feedback records into the trusted traceability information: user feedback, manual confirmation results, anomaly records, and optimization actions.

[0096] Secondly, the present invention provides a power question-and-answer result fusion and traceability device to implement the power question-and-answer result fusion and traceability method as described in any of the first aspects, including:

[0097] The question structuring module is used to perform structured parsing on the received user power natural language question and answer requests to obtain structured task information;

[0098] A multi-source result acquisition module is used to acquire multi-source results based on the structured task information;

[0099] The results standardization module is used to convert the multi-source results into a unified result unit.

[0100] A consistency verification module is used to perform consistency verification on the unified result unit and obtain the verification result.

[0101] The fusion weight calculation module is used to perform fusion processing on the unified result unit according to the verification result and the preset weight rules to obtain the fusion result;

[0102] A fusion interpretation generation module is used to generate a fusion interpretation result based on the fusion result;

[0103] The trusted traceability module is used to generate trusted traceability information based on the structured task information, multi-source results, unified result units, verification results, and fusion results.

[0104] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the power question-and-answer result fusion and tracing method as described in any of the first aspects.

[0105] Fourthly, the present invention provides a computer device, comprising:

[0106] Memory, used to store instructions;

[0107] A processor is configured to execute the instructions, causing the device to perform operations implementing the power question-and-answer result fusion and tracing method as described in any of the first aspects.

[0108] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0109] 1. This invention converts knowledge retrieval results, indicator query results, data service results, tool call results, and rule judgment results into a unified result unit, enabling results from different sources and in different formats to be organized and processed according to a unified structure, thus avoiding the problem of chaotic answer structure caused by directly splicing multiple source results.

[0110] 2. This invention performs consistency checks on business objects, time ranges, indicator definitions, data sources, and conclusions for unified result units. It can identify object conflicts, time conflicts, definition conflicts, source conflicts, and conclusion conflicts before merging and generating the final answer, thereby reducing the risk of misinterpretation caused by inconsistencies in multi-source results.

[0111] 3. This invention performs fusion processing on multi-source results based on verification results and preset weight rules, giving higher priority to results with high source credibility, high object matching, matching time range, consistent indicator caliber, and complete data during the fusion process, thereby improving the credibility and stability of the final answer;

[0112] 4. Based on the fusion results, this invention generates a fusion interpretation result that includes direct conclusions, data basis, indicator definition explanations, knowledge basis, cause analysis, handling suggestions, and risk warnings, enabling business personnel not only to obtain conclusions, but also to understand the corresponding data basis, knowledge basis, and definition basis.

[0113] 5. This invention generates credible traceability information based on structured task information, multi-source results, unified result units, verification results, and fusion results, enabling the final answer to be traced back to the original question, query conditions, data source, indicator definition, knowledge source, tool calls, rule basis, and conclusion generation record, facilitating subsequent review, auditing, and debriefing.

[0114] 6. This invention is applicable to various power business scenarios such as power equipment status analysis, alarm handling and interpretation, indicator question analysis, and operation data analysis, and can improve the business adaptability and reliable application capability of the power natural language question answering system. Attached Figure Description

[0115] Figure 1 The diagram shown is a flowchart of the power question-and-answer result fusion and traceability method provided in an embodiment of the present invention.

[0116] Figure 2 The diagram shown is a schematic diagram of the unified result unit structure provided in an embodiment of the present invention;

[0117] Figure 3 The diagram shown is a schematic diagram of a trusted traceability link provided in an embodiment of the present invention;

[0118] Figure 4 The diagram shown is a schematic diagram of the power question-and-answer result fusion and traceability device provided in an embodiment of the present invention. Detailed Implementation

[0119] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0120] It should be noted that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0121] Example 1

[0122] This embodiment discloses a method for fusing and tracing the results of electricity-related question-and-answer sessions, referencing... Figure 1 As shown, the specific steps include the following:

[0123] S1, perform structured parsing on the received user electricity natural language question and answer request to obtain structured task information;

[0124] S2, Obtain multi-source results based on the structured task information;

[0125] S3, convert the multi-source results into a unified result unit;

[0126] S4, perform a consistency check on the unified result unit to obtain the check result;

[0127] S5, perform fusion processing on the unified result unit according to the verification result and the preset weighting rules to obtain the fusion result;

[0128] S6, Generate a fusion interpretation result based on the fusion result;

[0129] S7. Generate trusted traceability information based on the structured task information, multi-source results, unified result unit, verification results, and fusion results.

[0130] In step S1, the structured task information includes one or more of the following: business intent, business object, indicator conditions, time range, query conditions, statistical method, business scenario, and expected output type.

[0131] Specifically, the business intent is used to characterize the type of user's power natural language question and answer request, which falls under knowledge-based question and answer, indicator query, status judgment, cause analysis, alarm handling, trend analysis, or report generation; the business object is used to characterize the plant, line, transformer, protection device, energy storage station, battery cluster, communication link, or telemetry point corresponding to the user's request. In this implementation, the business object includes one or more of the following: plant, line, transformer, protection device, energy storage station, battery cluster, communication link, and telemetry point; the indicator condition is used to characterize the indicator name, indicator alias, indicator caliber, or threshold rule that needs to be queried or judged; the time range is used to characterize the real-time, current day, past week, past month, or custom time period involved in the user's power natural language question and answer request; and the query condition is used to characterize the equipment type, voltage level, region, alarm type, statistical period, or filtering condition.

[0132] In step S2, the multi-source results include at least two of the following: knowledge retrieval results, indicator query results, data service results, tool call results, and rule judgment results. Specifically, if the structured task information includes a knowledge question-and-answer intent, knowledge retrieval results are obtained from a domain knowledge base, procedure document library, equipment manual library, or fault case library; if the structured task information includes an indicator query or indicator judgment intent, indicator query results are obtained from an indicator library, indicator caliber library, or threshold rule library; if the structured task information includes an operational data query intent, data service results are obtained from a real-time database, historical database, time-series data service, or unified model query service; if the structured task information includes a diagnosis, analysis, or report generation intent, alarm analysis tools, status diagnosis tools, report generation tools, or algorithm models are invoked to obtain tool call results; if the structured task information includes a threshold judgment, handling suggestion, or compliance verification intent, business rules, threshold rules, verification rules, or handling rules are invoked to obtain rule judgment results.

[0133] In step S3, this embodiment converts the multi-source results into a unified result unit, enabling results from different sources and in different formats to be processed according to the same structure. The unified result unit is used to carry the result content, business semantics, credibility, and traceability information returned by a single source.

[0134] refer to Figure 2 As shown, the unified result unit includes result source type, result source identifier, business object, indicator name, time range, data value, conclusion text, source basis, confidence level, and traceability information; the result source type is used to identify whether the result comes from knowledge retrieval, indicator query, data service, tool call, or rule judgment; the result source identifier is used to identify the knowledge base number, document number, indicator library number, data service number, tool number, or rule number; the business object is used to identify the plant, line, equipment, component, protection device, energy storage station, battery cluster, communication link, or telemetry point corresponding to the result; the indicator name is used to identify the indicator corresponding to the result. The following parameters are used: indicator alias or indicator code; time range to identify the start and end time, statistical period, data timestamp or generation time of the result; data value to record numerical values, tables, time series data or statistical results obtained from data services, indicator queries or tool analysis; conclusion text to record the conclusive content formed by knowledge retrieval, tool analysis or rule judgment; source basis to record knowledge fragments, document locations, indicator definitions, rule clauses, data table names or tool return basis; confidence level to indicate the credibility of the result unit in the current task; traceability information to record the result generation chain, source path, call record and generation time.

[0135] The unified result unit also includes one or more of the following: unit, scope of application, generation time, calling parameters, and exception information; the unit is used to record the unit of measurement corresponding to the indicator value or data value; the scope of application is used to indicate the business scenario, equipment type, voltage level, or data range to which the result is applicable; the calling parameters are used to record data query conditions, tool calling parameters, rule execution parameters, or model input parameters; and the exception information is used to record data missing, call failure, insufficient permissions, low confidence, or conflict markers.

[0136] In one implementation, the unified result unit includes fields such as sourceType, sourceId, objectId, objectName, metricId, metricName, timeRange, value, unit, conclusion, evidence, confidence, applicability, generationTime, callParameters, traceInfo, and exceptionInfo. Among these, sourceType indicates the result source type, sourceId indicates the result source identifier, objectId and objectName indicate the business object, metricId and metricName indicate the metric name, timeRange indicates the time range, value indicates the data value, unit indicates the unit, conclusion indicates the conclusion text, evidence indicates the source basis, confidence indicates the confidence level, applicability indicates the scope of application, generationTime indicates the generation time, callParameters indicates the call parameters, traceInfo indicates traceability information, and exceptionInfo indicates exception information.

[0137] Specifically, knowledge retrieval results are converted into knowledge-based result units, indicator query results are converted into indicator-based result units, data service results are converted into data-based result units, tool call results are converted into tool-based result units, and rule judgment results are converted into rule-based result units.

[0138] When a data service result is converted into a data-type result unit, the result source type is data service, the result source identifier is data service number, data table name or interface name, the data value includes real-time value, historical value, statistical value, tabular data or time series data, the calling parameters include query object, query time, filtering conditions and statistical method, and the traceability information includes data source, query statement, interface request record and return time.

[0139] When the results of a tool call are converted into tool-type result units, the result source type is tool call, the result source identifier is tool name, tool version or model number, the data values ​​include the tool output score, status quantity, anomaly type or analysis results, the conclusion text includes diagnostic conclusions, cause analysis or report content, and the call parameters include input object, time range, model parameters and tool execution parameters.

[0140] When a rule judgment result is converted into a rule-type result unit, the result source type is rule judgment, the result source identifier is rule number or rule name, the source basis includes rule clauses, threshold conditions, judgment conditions or handling rules, and the conclusion text includes rule hit result, judgment conclusion or handling suggestion.

[0141] Through the aforementioned unified result unit, results from different sources can undergo object consistency verification, time range consistency verification, indicator caliber consistency verification, data source consistency verification, conclusion consistency verification, weight fusion, and traceability information generation under the same data structure.

[0142] In step S4, after obtaining the unified result unit, this embodiment performs a consistency check on the unified result unit to identify inconsistencies between results from different sources and between the results and user issues. The consistency check includes business object consistency check, time range consistency check, indicator caliber consistency check, data source consistency check, and conclusion consistency check. The check result includes at least one of the following: check passed, check failed, conflict type, conflict field, conflict source, conflict description, and handling suggestion.

[0143] Business object consistency verification is used to determine whether the business objects in the unified result unit are consistent with the business objects in the structured task information. The business objects include power plants, lines, transformers, protection devices, energy storage stations, battery clusters, communication links, telemetry points, or other power business objects. When the object name, object code, equipment level, or region in the unified result unit does not match the business objects in the structured task information, an object conflict flag is generated.

[0144] The time range consistency check is used to determine whether the time range in a unified result unit is consistent with the time range in the structured task information. The time range includes the query start and end times, statistical period, data timestamps, alarm occurrence time, tool execution time, or knowledge entry effective time. A time conflict flag is generated when the time range of the data result does not cover the user-requested time range, or when the statistical period, sampling time, or update time is inconsistent between different result units.

[0145] The indicator consistency check is used to determine whether the indicator name, indicator code, calculation method, unit of measurement, threshold conditions, and statistical period are consistent within a unified result unit. When different sources use indicators with the same name but different calculation methods, or when the indicator unit, threshold, or statistical period is inconsistent with the user's question or the indicator library's definition, an indicator consistency conflict flag is generated.

[0146] Data source consistency verification is used to determine whether the data source corresponding to the unified result unit meets the requirements of the current task. The data sources include real-time databases, historical databases, time-series data services, statistical reports, unified model query services, or tool calculation results. When the collection source, generation method, update time, or data version of real-time data, historical data, statistical data, or tool-calculated data are inconsistent, a data source conflict flag is generated.

[0147] The conclusion consistency check is used to determine whether the conclusion texts in different unified result units support each other or are contradictory. The conclusion texts include procedural requirements obtained from knowledge retrieval, threshold judgments obtained from indicator queries, statistical conclusions obtained from data services, diagnostic results obtained from tool calls, and treatment conclusions obtained from rule judgments. When contradictory judgments, inconsistent risk levels, or inconsistent treatment recommendations occur among knowledge bases, data results, tool analysis results, or rule judgment results, a conclusion conflict flag is generated.

[0148] In this embodiment, the consistency check also includes one or more of the following: data integrity check, permission availability check, and call status check; data integrity check is used to determine whether key fields in the unified result unit are missing; permission availability check is used to determine whether the current task has permission to access the corresponding data source, knowledge source, or tool service; and call status check is used to determine whether the data service or tool service has been successfully called.

[0149] In this embodiment, the conflict types include one or more of the following: object conflict, time conflict, indicator caliber conflict, data source conflict, conclusion conflict, missing data, low confidence, unavailable source, and abnormal tool call. Object conflict indicates that the business object corresponding to the result is inconsistent with the object requested by the user; time conflict indicates that the time range, statistical period, or update time corresponding to the result is inconsistent with the user request; indicator caliber conflict indicates that the indicator name, calculation method, unit, threshold, or statistical period is inconsistent; data source conflict indicates that the data source, data version, collection source, or generation method is inconsistent; conclusion conflict indicates that there is a contradiction between the knowledge basis, data result, tool analysis result, or rule judgment result; missing data indicates that a key field in the unified result unit is missing; low confidence indicates that the confidence of the unified result unit is lower than a preset threshold; unavailable source indicates that the knowledge source, data source, tool source, or rule source is inaccessible; abnormal tool call indicates that the tool call times out, fails, or returns an abnormal result.

[0150] In step S5, after obtaining the verification result in this embodiment, the unified result unit is fused according to the verification result and the preset weight rules to obtain the fusion result, and a fusion interpretation result for business personnel is generated based on the fusion result; the preset weight rules include one or more of the following: source credibility, object matching degree, time matching degree, indicator caliber matching degree, data integrity, time freshness, rule priority, tool call success rate, historical feedback quality, and manual confirmation status.

[0151] Specifically, source credibility indicates the reliability of the result source; for example, a manually maintained indicator library, officially released procedure documents, authorized data services, and verified business rules have high source credibility. Object matching degree indicates the degree of matching between business objects in the unified result unit and business objects in the structured task information. Time matching degree indicates the degree of matching between the time range in the unified result unit and the time range requested by the user. Indicator caliber matching degree indicates the consistency between the indicator name, calculation method, unit, threshold, and statistical period in the unified result unit and the caliber in the indicator library. Data integrity indicates whether the key fields, data values, source basis, and calling parameters in the unified result unit are complete. Time freshness indicates the time interval between the result generation time or data update time and the current task. Rule priority indicates the priority order between different business rules or threshold rules. Tool call success rate indicates the availability of the tool service in the current task or historical tasks. Historical feedback quality indicates the evaluation of similar result sources by historical user feedback or manual confirmation. Manual confirmation status indicates whether the result has been confirmed by business personnel.

[0152] In one implementation, a fusion priority can be generated based on various weighting factors. For example, for unified result units with high source credibility, high object matching, high time matching, high indicator matching, and complete data, their fusion priority is increased; for unified result units with object conflicts, time conflicts, indicator conflicts, data source conflicts, conclusion conflicts, missing data, low confidence, or abnormal tool calls, their fusion priority is decreased; for unified result units that have been manually verified, their fusion priority is increased; for unified result units with unavailable sources or failed calls, they are recorded as abnormal information and are not used as the main basis for conclusions.

[0153] In one implementation, the overall weight of the unified result unit can be determined based on the source credibility weight, object matching weight, time matching weight, indicator caliber matching weight, data integrity weight, and conflict penalty coefficient.

[0154] The conflict penalty coefficient is determined based on the verification results. When there is no conflict marker in the unified result unit, the conflict penalty coefficient takes a higher value. When there are object conflicts, time conflicts, indicator conflicts, data source conflicts, or conclusion conflicts in the unified result unit, the conflict penalty coefficient is reduced. The above weight calculation method is only an example. In actual implementation, rule tables, scoring tables, model scoring, or manual configuration methods can also be used to determine the fusion priority.

[0155] After obtaining the fusion priority, the fusion weight calculation module sorts, filters and combines the unified result units according to the fusion priority to obtain the fusion result; the fusion result includes at least one or more of the following: main conclusion basis, auxiliary conclusion basis, conflict explanation, low confidence explanation and explanation of not using basis.

[0156] Specifically, when multiple unified result units reach the same conclusion and the source credibility meets the requirements, they are used together as the main conclusion basis; when some unified result units are consistent with the main conclusion but have low credibility or weak source, they are used as auxiliary conclusion basis; when some unified result units conflict with the main conclusion, conflict explanations are recorded in the fusion results; when unified result units have missing data, low confidence, unavailable source, or abnormal tool calls, risk explanations are recorded in the fusion results or explanations of not using the basis are not adopted.

[0157] In step S6, the fusion interpretation results include one or more of the following: direct conclusions, data basis, indicator definition explanations, knowledge basis, cause analysis, handling suggestions, risk warnings, and confidence level explanations.

[0158] Specifically, the direct conclusion is used to answer user questions in natural language; the data basis is used to list the indicator values, statistical values, time series changes, outliers, or tool output values ​​that support the direct conclusion; the indicator definition is used to explain the name, calculation method, unit, threshold, and statistical period of the indicator used; the knowledge basis is used to explain the procedures, equipment manuals, case libraries, or handling documents referenced; the conflict description is used to explain the source of conflicting results, conflicting fields, and handling methods; the cause analysis is used to generate possible causes by combining the data basis, knowledge basis, and rule basis; the handling suggestions are used to generate inspection, review, or handling suggestions by combining business rules, handling procedures, and tool analysis results; the risk warning is used to indicate data missing, low confidence, conflict markers, access restrictions, or matters requiring manual confirmation; and the confidence level description is used to explain the credibility of the current answer.

[0159] Through the above weighted fusion and interpretation generation process, the generated answer is no longer a simple splicing of multi-source content, but a structured business interpretation based on unified result units, consistency verification results, and fusion priorities.

[0160] In step S7, refer to Figure 3 As shown, this embodiment generates credible traceability information corresponding to the fusion interpretation result while generating the fusion interpretation result; the credible traceability information is used to record the task parsing, result acquisition, result standardization, consistency verification, weight fusion and interpretation generation chain in the process from the user's original question to the final answer generation, so that the final answer can be verified, reviewed and audited.

[0161] The credible traceability information includes at least one of the following: original question, structured task information, result source information, query conditions, query statement, data source, indicator caliber, knowledge source, tool call information, rule basis, unified result unit information, verification result, conflict information, fusion weight, and conclusion generation record.

[0162] Specifically, the original question is used to record the number of power-related natural language question-and-answer requests input by the user; the structured task information is used to record the business intent, business object, indicator conditions, time range, query conditions, statistical methods, business scenarios, and expected output types identified from the original question; the result source information is used to record the source type and source identifier corresponding to knowledge retrieval results, indicator query results, data service results, tool call results, and rule judgment results; the query conditions are used to record the object, time, filtering conditions, statistical period, and permission conditions used for data query, indicator query, or knowledge retrieval; the query statement is used to record structured query statements, domain query language, interface request parameters, or data service request parameters; the data source is used to record the source information of real-time databases, historical databases, time-series data services, statistical reports, or unified model query services; the indicator caliber is used to record the indicator name, indicator code, calculation method, unit of measurement, threshold conditions, statistical period, and scope of application; and the knowledge source is used to record the knowledge document name, document version, chapter position, and paragraph position. The system includes: Knowledge Fragment or Case Number; Tool Call Information (recording tool name, version, call parameters, return result, call time, and call status); Rule Basis (recording rule number, rule name, rule clauses, threshold conditions, judgment conditions, and rule hit results); Unified Result Unit Information (recording unified result unit fields after conversion from various source results); Verification Results (recording verification results for business object consistency, time range consistency, indicator caliber consistency, data source consistency, and conclusion consistency); Conflict Information (recording object conflicts, time conflicts, indicator caliber conflicts, data source conflicts, conclusion conflicts, missing data, low confidence, unavailable source, or abnormal tool call); Fusion Weight (recording the weight factor, fusion priority, and adoption status corresponding to each unified result unit); Conclusion Generation Record (recording the generation basis for the final direct conclusion, data basis, knowledge basis, cause analysis, handling suggestions, risk warnings, and confidence level description); and Feedback Record (recording user feedback, manual confirmation results, and subsequent optimization actions after feedback optimization).

[0163] In one implementation, credible traceability information can be organized according to a chain of final answer, integrated explanation result, integrated result, verification result, unified result unit, multi-source result, query conditions or calling parameters, data source, knowledge source, indicator caliber, tool call or rule basis.

[0164] Through the above-mentioned links, business personnel can trace back step by step from the final answer to the corresponding data basis, knowledge basis, indicator definition, query conditions, tool calls and rule basis.

[0165] Specifically, when the final answer contains a direct conclusion, the trusted traceability module records the main conclusion basis, auxiliary conclusion basis, and explanation of the basis not used for that direct conclusion; when the final answer contains data basis, the trusted traceability module records the corresponding data source, query conditions, query statement, data time range, and data return results; when the final answer contains an indicator definition, the trusted traceability module records the corresponding indicator definition, calculation method, unit, threshold, and statistical period; when the final answer contains knowledge basis, the trusted traceability module records the corresponding document name, document version, chapter position, and knowledge fragment; when the final answer contains tool analysis results, the trusted traceability module records the corresponding tool name, tool version, calling parameters, and return results; and when the final answer contains rule judgment results, the trusted traceability module records the corresponding rule number, rule clause, and hit condition.

[0166] When a conflict is detected during the consistency verification process, the trusted traceability module writes the conflict type, conflict field, conflict source, conflict description, and handling suggestions into the trusted traceability information. For example, when there is a conflict in the definition of an indicator, it records the name of the conflicting indicator, calculation method, unit, threshold, or statistical period; when there is a time conflict, it records the user's requested time range and the actual time range of the result; when there is a conflict in the conclusion, it records the knowledge basis, data results, tool analysis results, or rule judgment results that are conflicting.

[0167] Trustworthy traceability information can be stored in structured data format or output along with natural language responses. Structured data formats include log records, traceability tables, sets of link nodes, or sets of key-value pairs. Natural language output formats include explanations of the basis, source information, conflict warnings, risk warnings, and expandable traceability details.

[0168] By generating credible traceability information, the method in this embodiment can expand the results of power Q&A from simply providing an answer to providing an answer along with an explanation of the answer's source, generation process, and credibility, thereby meeting the requirements of power business scenarios for verification, auditing, and review.

[0169] After generating the fusion interpretation results and trusted traceability information, this embodiment can also continuously optimize the problem structuring, multi-source result acquisition, consistency verification, weight fusion, interpretation generation, and trusted traceability processes based on user feedback, manual confirmation results, and anomaly records.

[0170] The user feedback includes user confirmation, modification, rejection, or supplementation of the answer conclusion, data basis, indicator definition, knowledge basis, cause analysis, handling suggestions, risk warnings, and traceability information; the manual confirmation results include confirmation results from business experts, operation and maintenance personnel, scheduling personnel, or system administrators regarding the source of the results, indicator definition, conflict handling, fusion weight, and interpretation template; the anomaly records include object identification errors, indicator matching errors, missing data, unavailable sources, tool call failures, rule hit anomalies, conflict identification errors, and low-confidence answer records.

[0171] When a user reports an error in the identification of a business object, the feedback optimization module updates the entity identification rules, object alias library, or object mapping relationship based on the user's correction. For example, it adds the abbreviation of the plant, equipment, line, or protection device in the user's input to the object alias library to improve the accuracy of object identification for similar problems in the future.

[0172] When users report that the indicator matching is incorrect or the indicator definition is incorrect, the feedback optimization module updates the indicator matching rules, indicator alias library, or definition mapping rules based on the user's correction results; for example, mapping commonly used indicator abbreviations to standard indicator names, and correcting error thresholds or statistical periods to the standard definitions in the indicator library.

[0173] When users report inaccurate identification of conflicts from multiple sources, the feedback optimization module updates the conflict verification rules based on manual confirmation. For example, in scenarios where real-time data and historical statistical data coexist, comparison rules are configured between different data sources, different statistical periods, or different time granularities to reduce false or missed conflicts.

[0174] When users provide feedback that a certain source of results is more reliable or that a certain source of results is not applicable to the current business scenario, the feedback optimization module updates the weighting rules based on the manually confirmed status. For example, in a device status diagnosis scenario, the integration priority of status diagnosis tools and real-time data is increased; in a procedure Q&A scenario, the integration priority of formal procedure documents and handling manuals is increased.

[0175] When user feedback indicates that the expression of the integrated explanation results does not conform to business practices, the feedback optimization module updates the explanation template. The explanation template includes a direct conclusion template, a data basis template, an indicator definition explanation template, a knowledge basis template, a conflict explanation template, a cause analysis template, a handling suggestion template, and a risk warning template.

[0176] When the traceability information lacks fields that users need to verify, the feedback optimization module updates the traceability field configuration. For example, it adds fields such as data service interface name, query statement, knowledge fragment location, tool version, rule number, and manual confirmer or confirmation time.

[0177] The feedback optimization module can generate feedback samples from user feedback, manual confirmation results, and anomaly records. Based on these samples, it updates entity recognition rules, indicator matching rules, caliber mapping rules, conflict verification rules, weighting rules, interpretation templates, traceability fields, and tool invocation strategies. The updated rules, templates, and strategies are then used in the subsequent processing of power industry natural language question-and-answer requests.

[0178] Through the above-mentioned feedback correction and continuous optimization process, the embodiments of the present invention can continuously improve the accuracy of structured parsing, the accuracy of multi-source result fusion, the adaptability of interpretation and expression, and the integrity of reliable traceability in actual business use.

[0179] Example 2

[0180] This embodiment uses the scenario of energy storage battery consistency anomaly analysis as an example to illustrate the method of Embodiment 1.

[0181] The user's natural language Q&A request for electricity was: Has the battery consistency at the Tongliang Huaiyuan Energy Storage Station been abnormal in the past week?

[0182] The above requests are parsed in a structured manner to obtain structured task information. The business intent is to query indicators, determine status, and analyze causes. The business object is the Tongliang Huaiyuan Energy Storage Station. The indicator conditions include battery consistency, single-cell voltage difference, temperature difference, and state of charge difference. The time range is the past week. The business scenario is the analysis of the operating status of the energy storage station. The expected output type is anomaly judgment, cause explanation, and handling suggestions.

[0183] Based on the structured task information, multi-source results are obtained; specifically, the definitions of battery consistency indicators, pressure difference thresholds, temperature difference thresholds, and state of charge difference thresholds are obtained from the indicator library; the individual cell voltage, temperature, state of charge, and alarm records of each battery cluster at the Tongliang Huaiyuan Energy Storage Station for the past week are obtained from historical databases or time-series data services; operation and maintenance knowledge, equipment manuals, and historical cases related to battery consistency anomalies are retrieved from the domain knowledge base; battery status diagnostic tools are invoked to obtain battery cluster anomaly scores, abnormal cell location results, and possible cause analysis; and the rule engine is invoked to determine whether battery consistency anomalies exist based on threshold rules.

[0184] The above multi-source results are converted into unified result units; indicator query results are converted into indicator-type result units, where the source basis includes the definition of battery consistency indicators, pressure difference threshold, temperature difference threshold, and statistical period; data service results are converted into data-type result units, where the data values ​​include the maximum voltage difference of individual cells, the maximum temperature difference, the maximum state of charge difference, and the time of anomaly occurrence in the past week; knowledge retrieval results are converted into knowledge-type result units, where the source basis includes equipment manual chapters, operation and maintenance procedure clauses, and historical case fragments; tool call results are converted into tool-type result units, where the conclusion text includes the abnormal battery cluster number, anomaly score, and possible causes; rule judgment results are converted into rule-type result units, where the conclusion text includes whether the threshold is exceeded and the threshold rule that is hit.

[0185] Consistency checks are performed on the aforementioned unified result units. Specifically, it checks whether each result unit corresponds to the Tongliang Huaiyuan Energy Storage Station. If a data result corresponds to another energy storage station, an object conflict flag is generated. It also checks whether the data service results, tool call results, and user requests all correspond to the past week's time range. If the tool results only cover the data of the current day, a time conflict flag is generated. Furthermore, it checks whether the battery consistency criteria in the indicator library are consistent with the criteria used in the tool's diagnostic model. If the thresholds or statistical periods are different, an indicator criterion conflict flag is generated. Finally, it checks whether the data sources of historical data, time-series data, and alarm records are consistent. If there are differences in the collection source or update time, a data source conflict flag is generated. Finally, it checks whether the rule judgment conclusions, tool diagnostic conclusions, and knowledge basis support each other. If the rule judgment is abnormal but the tool diagnosis is normal, a conclusion conflict flag is generated.

[0186] The fusion priority of each unified result unit is determined based on the verification results and preset weight rules. If historical data, indicator definitions, and rule judgments all indicate that the maximum voltage difference of a single cell exceeds the set threshold, and the battery status diagnostic tool locates the same battery cluster, then the aforementioned data-type result unit, indicator-type result unit, rule-type result unit, and tool-type result unit will be used as the main conclusion basis. If historical cases in the knowledge retrieval results are similar to the current abnormal characteristics, they will be used as the basis for auxiliary cause analysis. If there are result units with inconsistent time ranges or missing data, their fusion priority will be reduced, and this will be recorded in the risk warning.

[0187] The fusion results generate a fusion interpretation; for example, the direct conclusion is: The Tongliang Huaiyuan Energy Storage Station has experienced abnormal battery consistency over the past week. The data basis is: Within the past week, the maximum voltage difference between individual cells in some battery clusters exceeded the battery consistency index threshold, with anomalies concentrated in specific battery clusters, accompanied by increased differences in state of charge. The index definition is: This judgment uses the individual cell voltage difference, temperature difference, and state of charge difference from the battery consistency index, calculated according to the statistical period of the past week. The knowledge basis is: Equipment operation and maintenance knowledge indicates that abnormal battery consistency is usually related to cell aging, sampling anomalies, equalization strategy failure, or temperature control differences. The cause analysis is: Combining data changes and tool diagnostic results, possible causes include performance degradation of some individual cells, sampling link anomalies, or insufficient equalization control effect. The recommended actions are: Review the sampling data of the abnormal battery clusters, check the sampling link of the battery management system, verify the execution of the equalization strategy, and conduct focused inspections of the abnormal cells. The risk warning is: If data is missing for some time periods or the tool diagnostic results only cover some battery clusters, manual review is required.

[0188] Generate credible traceability information corresponding to the above-mentioned fusion interpretation results; the credible traceability information includes the user's original question, structured task information, object identification results of Tongliang Huaiyuan Energy Storage Station, battery consistency index caliber, data query conditions for the past week, time series data service interface, source of index library, knowledge document name and chapter position, battery status diagnosis tool name and calling parameters, threshold rule number, consistency verification results, fusion weight of each unified result unit, and final conclusion generation record.

[0189] Through this embodiment, business personnel can not only obtain a judgment on whether the consistency of batteries in an energy storage station is abnormal, but also trace the data source, indicator definition, knowledge basis, tool call records, rule basis, and conflict verification results corresponding to the conclusion.

[0190] Example 3

[0191] This embodiment uses a secondary equipment alarm handling scenario as an example to illustrate the method of Embodiment 1.

[0192] The user's natural language Q&A request for electricity was: How to handle communication interruption alarms of protection devices?

[0193] The above request is parsed in a structured manner to obtain structured task information; among which, the business intent is to generate alarm handling Q&A and handling suggestions; the business object is the protection device; the indicator condition or event condition is communication interruption alarm; the time range is the current alarm or the most recent alarm; the business scenario is the operation and maintenance of secondary equipment; and the expected output type is an explanation of the alarm cause, handling steps and basis.

[0194] Based on the structured task information, multi-source results are obtained. Specifically, the meaning, scope of impact, troubleshooting steps, and handling requirements of communication interruption alarms related to protection devices are retrieved from the domain knowledge base, procedure document library, and handling manual. The current or most recent communication interruption alarm record of the protection device is obtained from the alarm database, including alarm time, alarm object, alarm level, alarm status, and recovery status. The online status of the protection device, communication link status, channel quality, and upstream communication status are obtained from real-time data services or communication status query services. The alarm analysis tool is invoked to obtain alarm correlation analysis results, possible causes, and scope of affected equipment. The rule engine is invoked to generate handling judgment results based on alarm handling rules, equipment operation rules, and communication link rules.

[0195] The above multi-source results are converted into unified result units; knowledge retrieval results are converted into knowledge-based result units, where the source basis includes the name of the handling manual, procedure clauses, and fault case fragments; alarm record and communication status query results are converted into data-based result units, where the data values ​​include alarm occurrence time, alarm status, communication link status, channel quality, and recovery status; alarm analysis tool return results are converted into tool-based result units, where the conclusion text includes possible causes, related alarms, scope of affected equipment, and suggested troubleshooting directions; rule judgment results are converted into rule-based result units, where the source basis includes the alarm handling rule number, rule conditions, and handling requirements.

[0196] Consistency checks are performed on the aforementioned unified result units. Specifically, the checks verify whether the knowledge retrieval results, alarm records, communication status results, and tool analysis results all correspond to the communication interruption alarm of the protection device. If the alarm record corresponds to other devices or other alarm types, an object conflict or indicator conflict flag is generated. The checks verify whether the alarm record time, communication status query time, and user request time are consistent. If the communication status data is not within the time range corresponding to the current alarm, a time conflict flag is generated. The checks verify whether the alarm classification in the alarm handling manual is consistent with the alarm classification in the alarm database. If the alarm names are the same but the classifications or levels are different, an indicator conflict flag is generated. The checks verify whether the tool analysis conclusion, rule judgment conclusion, and knowledge handling requirements are consistent. If the tool judges that the communication link is abnormal while the rule judges that the device is offline, a conclusion conflict flag is generated.

[0197] The fusion priority of each unified result unit is determined based on the verification results and preset weight rules. If the alarm records, real-time communication status, and alarm analysis tools all indicate that the protection device and the master station are interrupted, then the alarm data type result unit, communication status data type result unit, and tool type result unit will be used as the main conclusion basis. If the handling manual and procedure documents provide corresponding troubleshooting steps, they will be used as the basis for handling suggestions. If there are expired alarm records, mismatched historical cases, or abnormal tool calls, the fusion priority of the corresponding result unit will be reduced and recorded in the risk warning.

[0198] The fusion results generate a fusion interpretation result; for example, the generated direct conclusion is: a communication interruption alarm of the protection device indicates an anomaly in the communication link between the protection device and the monitoring or dispatch master station, requiring investigation based on the current alarm status and the communication link status. The data basis is: the current alarm record shows a communication interruption alarm of the protection device, and the communication status query results show that the corresponding link status is abnormal or the data has not been refreshed for a long time; the knowledge basis is: the handling manual stipulates that communication interruption alarms should first check the device operating status, communication port, switch link, front-end communication process, and network configuration; the cause analysis is: possible causes include device power failure, communication port anomaly, switch link anomaly, network configuration change, front-end communication process anomaly, or channel quality degradation; the handling suggestion is: first check the operating status and power status of the protection device, then check the communication port and network cable connection, then check the switch port, network connectivity, and front-end communication status, and if necessary, contact communication or automation professionals for joint investigation; the risk warning is: if the current alarm has been resolved or communication status data cannot be obtained in real time, manual confirmation based on the on-site status and the latest alarm record is required.

[0199] Generate trusted traceability information corresponding to the above fusion interpretation results; the trusted traceability information includes the user's original problem, structured task information, protection device object identification results, communication interruption alarm records, alarm database source, communication status query conditions, handling manual name and chapter position, alarm analysis tool name and calling parameters, alarm handling rule number, consistency verification results, fusion weight, and final handling suggestion generation record.

[0200] Through this embodiment, business personnel can not only obtain handling suggestions for communication interruption alarms of protection devices, but also trace the alarm records, communication status, handling manuals, tool analysis results, rule basis and conflict verification results corresponding to the handling suggestions.

[0201] Example 4

[0202] This embodiment uses the scenario of scheduling automation telemetry not refreshing as an example to illustrate the method of Embodiment 1.

[0203] The user's natural language Q&A request for electricity was: "Is there a problem with telemetry not refreshing at a certain station today? What could be the reason?"

[0204] The above requests are parsed in a structured manner to obtain structured task information. The business intent is to query runtime data, identify anomalies, and analyze causes. The business object is a certain station. The indicator conditions or event conditions are telemetry refresh status, telemetry update time, telemetry change rate, and telemetry anomaly alarm. The time range is the current day. The business scenario is to monitor the automated operation of the scheduling system. The expected output type is whether an anomaly exists, the scope of the anomaly, possible causes, and the basis for tracing.

[0205] Based on the structured task information, obtain multi-source results; specifically, obtain the latest values, update times, refresh cycles, and changes of telemetry points for a certain station on that day from a real-time database or time-series data service; obtain alarm records related to telemetry non-refresh, telemetry interruption, channel abnormality, or pre-process abnormality on that day from the alarm database; retrieve cause analysis knowledge and handling experience related to telemetry non-refresh from the domain knowledge base, operation and maintenance procedures, and historical case database; call communication link status query tools, pre-process status query tools, or telemetry quality analysis tools to obtain communication link status, pre-process status, channel quality, acquisition status, and abnormal telemetry point range; call the rule engine to determine whether there is a telemetry non-refresh problem based on the telemetry refresh cycle, data update time, and alarm rules.

[0206] The above multi-source results are converted into unified result units; real-time data or time-series data service results are converted into data-type result units, where data values ​​include the latest telemetry point value, last update time, duration without refresh, refresh cycle, and number of abnormal telemetry points; alarm records are converted into data-type result units, where the conclusion text includes whether there are telemetry non-refresh alarms, channel abnormal alarms, or preceding abnormal alarms; knowledge retrieval results are converted into knowledge-type result units, where the source basis includes operation and maintenance procedure chapters, fault case fragments, and cause analysis knowledge; tool call results are converted into tool-type result units, where the conclusion text includes communication link status, preceding process status, abnormal channels, abnormal telemetry point range, and possible causes; rule judgment results are converted into rule-type result units, where the source basis includes telemetry refresh judgment rules, data timeout rules, and alarm triggering rules.

[0207] Consistency checks are performed on the aforementioned unified result units. Specifically, each result unit is checked to ensure it corresponds to a specific station in the user request. If a telemetry result or alarm record corresponds to another station, an object conflict flag is generated. Real-time data, alarm records, tool call results, and user requests are all checked to ensure they correspond to the current day's time range. If a result is a historical fault record rather than a record from the current day, a time conflict flag is generated. The refresh cycle, timeout duration, and alarm triggering conditions in the telemetry non-refresh judgment rules are checked to ensure they are consistent with the indicator caliber or rule base. If different sources define the timeout threshold differently, an indicator caliber conflict flag is generated. The telemetry status in the real-time database, time-series database, and alarm database is checked to ensure they are consistent. If the real-time database shows no refresh but the alarm database has no corresponding alarm, a data source conflict or conclusion conflict flag is generated. The communication link tool, the pre-state tool, and the rule judgment results are checked to ensure they support each other. If the communication link is normal but the pre-state is abnormal, the conclusion difference is recorded and used as a basis for cause analysis.

[0208] The fusion priority of each unified result unit is determined based on the verification results and preset weight rules. If real-time data or time-series data services show that the update time of multiple telemetry points exceeds the refresh threshold, and the rule engine hits the telemetry non-refresh rule, then data-type result units and rule-type result units are used as the main conclusion basis for whether there is an anomaly. If the communication link status query tool or the pre-state query tool returns an abnormal status, then the tool-type result unit is used as the main basis for cause analysis. If there are descriptions of the causes of the same phenomenon in historical cases and operation and maintenance knowledge, then they are used as the basis for auxiliary cause analysis. If a data source has a delayed update time, missing data, or failed interface call, then its fusion priority is reduced and recorded in the risk warning.

[0209] The fusion results generate a fusion interpretation. For example, the direct conclusion is: a certain station has a telemetry failure to refresh issue on that day; the data basis is: the last update time of some telemetry points exceeds the preset refresh cycle, and the abnormal telemetry points are concentrated in a specified interval or a specified channel; the indicator definition is: this judgment is based on the telemetry refresh cycle, the last update time, and the timeout threshold; the knowledge basis is: operation and maintenance procedures and historical cases show that telemetry failure to refresh is usually related to channel interruption, abnormal pre-process, abnormal acquisition device, degraded communication quality, or abnormal measurement and control device; the cause analysis is: combined with the communication link status, pre-process status, and the range of abnormal telemetry points, possible causes include abnormal pre-process, interrupted channel link, abnormal communication of acquisition device, or failure of the station-end measurement and control device to send data. The handling recommendation is: it is recommended to first check the pre-process communication and channel status, then check the station-end communication link, the operating status of the measurement and control device, and related alarm records, and if necessary, contact automation and communication professionals for joint handling; the risk warning is: if some telemetry points are under planned maintenance or manually blocked, further confirmation is needed based on the operating mode and on-site status.

[0210] Generate trusted traceability information corresponding to the above fusion interpretation results; the trusted traceability information includes the user's original question, structured task information, object identification results of a certain station, time range of the day, telemetry point query conditions, real-time data service or time-series data service interface, alarm record source, telemetry refresh judgment rule number, communication link status query tool name and calling parameters, previous status query tool name and calling parameters, operation and maintenance procedures or historical case source, consistency verification results, fusion weight, and final conclusion generation record.

[0211] Through this embodiment, business personnel can not only obtain a judgment on whether a certain station has a telemetry non-refresh problem on a certain day, but also trace the telemetry data, alarm records, communication link status, pre-state, rule basis, knowledge basis and conflict verification results corresponding to the conclusion.

[0212] Example 5

[0213] Based on the same inventive concept as Embodiment 1, this embodiment discloses a power question-and-answer result fusion and traceability device, with reference to... Figure 4 As shown, it includes:

[0214] The question structuring module is used to perform structured parsing on the received user power natural language question and answer requests to obtain structured task information;

[0215] A multi-source result acquisition module is used to acquire multi-source results based on the structured task information;

[0216] The results standardization module is used to convert the multi-source results into a unified result unit.

[0217] A consistency verification module is used to perform consistency verification on the unified result unit and obtain the verification result.

[0218] The fusion weight calculation module is used to perform fusion processing on the unified result unit according to the verification result and the preset weight rules to obtain the fusion result;

[0219] A fusion interpretation generation module is used to generate a fusion interpretation result based on the fusion result;

[0220] The trusted traceability module is used to generate trusted traceability information based on the structured task information, multi-source results, unified result units, verification results, and fusion results.

[0221] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0222] Example 6

[0223] This embodiment discloses a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the power question-and-answer result fusion and tracing method as described in Embodiment 1.

[0224] Example 7:

[0225] This embodiment discloses a computer device, including:

[0226] Memory, used to store computer instructions;

[0227] A processor is used to execute the computer instructions to implement the steps of the power question-and-answer result fusion and tracing method described in Embodiment 1.

[0228] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0229] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0230] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0231] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0232] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An electric power question and answer query result fusion tracing method, characterized in that, include: The received user electricity natural language question-and-answer request is parsed in a structured manner to obtain structured task information; Obtain multi-source results based on the structured task information; The multi-source results are converted into a unified result unit; The consistency of the unified result unit is checked to obtain the check result; The unified result unit is fused according to the verification result and the preset weighting rules to obtain the fusion result; A fusion interpretation result is generated based on the fusion result; Trustworthy traceability information is generated based on the structured task information, multi-source results, unified result units, verification results, and fusion results.

2. The method of claim 1, wherein, The structured task information includes at least one of the following: business intent, business object, indicator conditions, time range, query conditions, statistical method, business scenario, and expected output type. The business intent is used to characterize the type of user's power natural language question and answer request, which is either knowledge-based question and answer, indicator query, status judgment, cause analysis, alarm handling, trend analysis, or report generation. The business object is used to represent the power plant, line, transformer, protection device, energy storage station, battery cluster, communication link or telemetry point corresponding to the user's power natural language question and answer request; The indicator conditions are used to characterize the indicator name, indicator alias, indicator definition, or threshold rule that needs to be queried or judged. The time range is used to characterize the real-time, current day, past week, past month, or custom time period involved in the user's electricity natural language question and answer request; The query criteria are used to characterize the device type, voltage level, region, alarm type, statistical period, or filtering conditions.

3. The method for fusing and tracing the results of power-related question-and-answer sessions according to claim 1, characterized in that, The multi-source results include at least two of the following: knowledge retrieval results, indicator query results, data service results, tool call results, and rule judgment results. The step of obtaining multi-source results based on the structured task information includes: If the structured task information includes knowledge question-and-answer intent, obtain knowledge retrieval results from the domain knowledge base, procedure document library, equipment manual library, or fault case library; If the structured task information includes an indicator query or indicator judgment intent, obtain the indicator query results from the indicator library, indicator caliber library, or threshold rule library; If the structured task information includes the intent to query running data, obtain the data service results from the real-time database, historical database, time-series data service, or unified model query service; If the structured task information includes diagnostic, analytical, or report generation intent, invoke alarm analysis tools, status diagnostic tools, report generation tools, or algorithm models, and obtain the tool invocation results; If the structured task information includes threshold judgment, handling suggestions, or compliance verification intent, call the business rules, threshold rules, verification rules, or handling rules to obtain the rule judgment result.

4. The method for fusing and tracing the results of power-related question-and-answer sessions according to claim 1, characterized in that, The unified result unit includes result source type, result source identifier, business object, indicator name, time range, data value, conclusion text, source basis, confidence level, and traceability information; The result source type is used to identify whether the result comes from knowledge retrieval, indicator query, data service, tool call, or rule judgment; The result source identifier is used to identify the knowledge base number, document number, indicator library number, data service number, tool number, or rule number; The business object is used to identify the plant, line, equipment, component, protection device, energy storage station, battery cluster, communication link or telemetry point corresponding to the result; The indicator name is used to identify the indicator, indicator alias, or indicator code corresponding to the result; The time range is used to identify the start and end time, statistical period, data timestamp, or generation time of the result. The data values ​​are used to record numerical values, tables, time-series data, or statistical results obtained from data services, indicator queries, or tool analysis. The conclusion text is used to record the conclusive content formed by knowledge retrieval, tool analysis, or rule judgment; The source basis is used to record knowledge fragments, document locations, indicator definitions, rule clauses, data table names, or tool return bases; The confidence level is used to indicate the degree of credibility of the result unit in the current task; The traceability information is used to record the result generation chain, source path, call record, and generation time; The unified result unit also includes at least one of the following: unit, scope of application, generation time, calling parameters, and exception information; The unit is used to record the unit of measurement corresponding to the indicator value or data value; The scope of application is used to indicate the business scenarios, equipment types, voltage levels, or data ranges to which the result applies. The calling parameters are used to record data query conditions, tool calling parameters, rule execution parameters, or model input parameters; The exception information is used to record data missing, call failure, insufficient permissions, low confidence or conflict markers; The step of converting the multi-source results into a unified result unit includes: Convert knowledge retrieval results into knowledge-based result units; And / or, Convert the indicator query results into indicator-type result units; And / or, Transform data service results into data-type result units; And / or, Convert tool call results into tool-type result units; And / or, Convert the rule judgment result into a rule-type result unit.

5. The method for fusing and tracing the results of power-related question-and-answer sessions according to claim 1, characterized in that, The consistency verification includes business object consistency verification, time range consistency verification, indicator caliber consistency verification, data source consistency verification, and conclusion consistency verification. The process of performing consistency verification on the unified result unit to obtain the verification result includes: Determine whether the business object corresponding to the unified result unit is consistent with the business object in the structured task information: if they are inconsistent, generate an object conflict flag and the verification fails; if they are consistent, the verification passes. Determine whether the time range corresponding to the unified result unit is consistent with the time range in the structured task information: if they are inconsistent, generate a time conflict flag and the verification fails; if they are consistent, the verification passes. Determine whether the indicator name, calculation method, unit, threshold, or statistical period corresponding to the unified result unit are consistent: if inconsistent, generate an indicator caliber conflict flag and the verification fails; if consistent, the verification passes. Determine whether the data collection source, statistical source, or update time corresponding to different data sources in the unified result unit are consistent: if they are inconsistent, a data source conflict flag is generated and the verification fails; if they are consistent, the verification passes. Determine whether the conclusions among the knowledge basis, data results, tool analysis results, or rule judgment results in the unified result unit are consistent: if inconsistent, generate a conclusion conflict flag and the verification fails; if consistent, the verification passes. The consistency verification also includes at least one of data integrity verification, permission availability verification, and call status verification; The verification result includes at least one of the following: verification passed, verification failed, conflict type, conflict field, conflict source, conflict description, and handling suggestion; The conflict types include at least one of the following: object conflict, time conflict, indicator conflict, data source conflict, conclusion conflict, missing data, low confidence, unavailable source, and abnormal tool invocation.

6. The method for fusing and tracing the results of power-related question-and-answer sessions according to claim 1, characterized in that, The preset weighting rules include at least one of the following: source credibility, object matching degree, time matching degree, indicator caliber matching degree, data integrity, time freshness, rule priority, tool call success rate, historical feedback quality, and manual confirmation status. The source credibility is used to indicate the reliability of the source of the result; The object matching degree is used to represent the degree of matching between business objects in the unified result unit and business objects in the structured task information. The time matching degree is used to represent the degree of matching between the time range in the unified result unit and the time range requested by the user; The indicator caliber matching degree is used to indicate the degree of consistency between the indicator name, calculation method, unit, threshold, and statistical period in the unified result unit and the caliber of the indicator library; The data integrity refers to whether the key fields, data values, source basis, and calling parameters in the unified result unit are complete. The time freshness is used to represent the time interval between the result generation time or data update time and the current task; The rule priority is used to indicate the priority order among different business rules or threshold rules; The tool call success rate is used to indicate the availability of the tool service in the current or historical tasks; The historical feedback quality is used to represent the evaluation of similar results sources by historical user feedback or manual confirmation; The manual confirmation status is used to indicate whether the result has been confirmed by business personnel; The fusion results include at least one of the following: main conclusion basis, auxiliary conclusion basis, conflict explanation, low confidence explanation, and explanation of not using basis; When multiple unified result units have consistent conclusions and the credibility of their sources meets the requirements, they are used together as the basis for the main conclusion. When a unified result unit is consistent with the direction of the main conclusion but the credibility of its source does not meet the requirements, it shall be used as a basis for auxiliary conclusions. When there is a conflict between the unified result unit and the main conclusion, record the explanation of the conflict; When the source credibility of a unified result unit does not meet the requirements, record a low confidence level description; When a unified result unit has missing data, an unavailable source, or an abnormal tool call, the record will not be used based on the explanation. And / or, The integrated interpretation results include direct conclusions, data basis, explanation of indicator definitions, knowledge basis, cause analysis, handling suggestions, and risk warnings; The direct conclusions are used to answer user questions; The data cited is used to illustrate the data values, statistical results, or trend changes corresponding to the conclusions. The description of the indicator scope is used to explain the name of the indicator, calculation method, unit, threshold or statistical period used; The knowledge base mentioned is used to explain the referenced procedural documents, equipment manuals, or case knowledge; The causal analysis is used to explain the possible reasons for reaching the conclusion; The proposed actions are used to provide suggestions for inspection, review, processing, or follow-up operations. The risk warning is used to explain the confidence level, missing data, conflict markers, or matters requiring manual confirmation.

7. The method for fusing and tracing the results of power-related question-and-answer sessions according to claim 1, characterized in that, The credible traceability information includes at least one of the following: original question, structured task information, result source information, query conditions, query statement, data source, indicator definition, knowledge source, tool call information, rule basis, unified result unit information, verification result, conflict information, fusion weight, and conclusion generation record. And / or, It also includes: optimizing at least one of the following based on user feedback, manual confirmation results, and anomaly records: entity recognition rules, indicator matching rules, indicator alias library, caliber mapping rules, conflict verification rules, weighting rules, interpretation templates, traceability fields, and tool invocation strategies; and writing at least one of the following as feedback records into the trusted traceability information: user feedback, manual confirmation results, anomaly records, and optimization actions.

8. A power-related question-and-answer result fusion and traceability device, characterized in that, include: The question structuring module is used to perform structured parsing on the received user power natural language question and answer requests to obtain structured task information; A multi-source result acquisition module is used to acquire multi-source results based on the structured task information; The results standardization module is used to convert the multi-source results into a unified result unit. A consistency verification module is used to perform consistency verification on the unified result unit and obtain the verification result. The fusion weight calculation module is used to perform fusion processing on the unified result unit according to the verification result and the preset weight rules to obtain the fusion result; A fusion interpretation generation module is used to generate a fusion interpretation result based on the fusion result; The trusted traceability module is used to generate trusted traceability information based on the structured task information, multi-source results, unified result units, verification results, and fusion results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the power question-and-answer result fusion and tracing method as described in any one of claims 1-7.

10. A computer device, characterized in that, include: Memory, used to store instructions; A processor is configured to execute the instructions, causing the device to perform operations that implement the power question-answering result fusion and tracing method as described in any one of claims 1-7.

Images