Construction method of power customer address data governance agent

Abstract

The embodiment of the invention provides a construction method of a power customer address data governance agent, and belongs to the technical field of big data. The method comprises the following steps: constructing a power customer address data knowledge base through three-layer construction logic of multi-source data fusion, large model semantic enhancement and knowledge graph relationship modeling; core agents are constructed, the core agents comprise a rule generation agent, an intelligent analysis agent and a quality evaluation agent, all the agents operate based on the knowledge base, and modular design is adopted; and establishing a multi-agent collaborative mechanism, and realizing collaborative operation and closed-loop optimization of the core agent through data circulation, instruction interaction and result feedback. According to the construction method, intelligent and full-process management of the power customer address data is realized, the abnormal address identification accuracy is improved, the transaction data processing time is shortened, and the basic-level labor cost and the business handling error rate are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, in particular, to a construction method of an electric power customer address data governance intelligent agent. BACKGROUND

[0002] In the process of electric power system operation, customer address data as core basic data is the key support to guarantee the efficient development of various businesses. Currently, electric power customer address data governance faces four core pain points: First, data specification is insufficient. Influenced by factors such as lack of unified address standard, non-standard input process, and "one place with multiple names", address data generally has problems such as missing levels and chaotic superior-inferior relationship, which makes it difficult to support precise business landing; Second, governance efficiency is low. The traditional governance mode relies on manual preparation of verification rules, and the grassroots operation and maintenance need to handle data through account comparison, on-site verification, telephone communication and other ways, which is greatly affected by subjective judgment, and the processing time of a single data is more than 7 days, which cannot meet the timeliness requirements of new power system data governance; Third, the rules are static and lagging. The existing verification rules are manually preset static rules, which cannot adapt to the dynamic changes of address data such as new community and road renaming, resulting in less than 60% coverage of abnormal address identification, and potential data problems are easily missed; Fourth, the governance closed loop is missing. The existing technology lacks a full-link collaborative mechanism of "problem discovery-rule generation-governance execution-quality evaluation", and the data governance effect cannot be fed back and optimized in real time, forming a governance breakpoint.

[0003] It can be seen that the existing technology cannot meet the requirements of new power system on data governance timeliness and accuracy. SUMMARY

[0004] The purpose of the embodiment of the present application is to provide a construction method of an electric power customer address data governance intelligent agent, which realizes intelligent and full-process governance of electric power customer address data, improves the accuracy of abnormal address identification, shortens the processing time of abnormal data, and reduces the labor cost of grassroots and the error rate of business handling.

[0005] In order to achieve the above purpose, the embodiment of the present application provides a construction method of an electric power customer address data governance intelligent agent, which comprises: An electric power customer address data knowledge base is constructed through multi-source data fusion, large model semantic enhancement and knowledge graph relationship modeling three-layer construction logic; A core intelligent agent is constructed, including a rule generation intelligent agent, an intelligent analysis intelligent agent and a quality evaluation intelligent agent, each intelligent agent is based on the knowledge base and adopts modular design; A multi-agent collaboration mechanism is established to realize the collaborative operation and closed-loop optimization of core agents through data flow, instruction interaction and result feedback.

[0006] Preferably, the multi-source data fusion comprises: Collecting internal data, including power marketing system address data, GIS map data and historical management data; Collecting external data, including standard address data of the Ministry of Civil Affairs, regional division data of the Statistics Bureau and map address data; Cleaning and standardizing the collected data to form a standardized data set.

[0007] Preferably, the large model semantic enhancement comprises: Fine-tuning the standardized data set using a large language model to optimize the model's understanding of power address terminology; The precision P, recall R and F1 score are used to evaluate the power terminology parsing ability of the fine-tuned model, and the F1 score is ensured to reach a preset threshold.

[0008] Preferably, the precision P is calculated according to formula (1), , (1) Where N corr is the number of correctly parsed power terms, and N pred is the total number of power terms parsed by the model; The recall R is calculated according to formula (2), , (2) Where N true is the total number of power terms actually contained in the test set; The F1 score is calculated according to formula (3), , (3) The preset threshold is not less than 0.95.

[0009] Preferably, the knowledge graph relationship modeling comprises: Based on the fine-tuned large model, address entities and attributes are extracted; Constructing an address entity relationship graph to achieve deep semantic understanding of complex addresses.

[0010] Preferably, the rule generation agent based on the power customer address data knowledge base has the functions of data access, large model learning, rule generation and rule output; wherein, The data access function is used to obtain historical management data and real-time address data in the knowledge base in real time; The large model learning function is used to learn the data in multiple dimensions by using the fine-tuned Guangming Power large model, and identify address data problem patterns. The rule generation function is used to dynamically generate four types of verification rules, including format checking, anomaly detection, semantic correction, and membership detection, based on the learning results. The rule output function is used to automatically convert the generated rules into executable scripts and synchronize them to the intelligent analysis agent.

[0011] Preferably, according to formula (4), the generated rules are screened by introducing rule confidence, , (4) where C(R) is the confidence of rule R, and rules with a confidence lower than 80% need to be relearned; N hit is the number of real abnormal addresses hit by the rule in the test set; N test is the total number of addresses in the rule test set; and α=1, β=2 are smoothing coefficients. And according to formula (5), a rule weight update algorithm is used to dynamically and iteratively optimize the rules, , (5) where W t (R) is the weight of rule R after the tth iteration, and the initial W_0=1; C t (R) is the actual confidence of rule R in the tth round; C avg (t) is the average confidence of all rules in the tth round; rules with a weight ≥1.2 are preferentially executed, and rules with a weight <0.8 are put into the optimization queue.

[0012] Preferably, the intelligent analysis agent is constructed as follows: Adopting a dual-driven rule generation mode driven by autonomous learning and business demand; Having full life cycle management functions of problem discovery, task assignment, rectification verification, and effect feedback; For complex problems, it has the function of automatically analyzing problems and matching special sub-agents.

[0013] Preferably, the quality evaluation agent is constructed as follows: Having a multi-dimensional data collection function of collecting governance process data, data quality data, and business impact data; Using a large model combined with a preset evaluation system to quantitatively evaluate the collected data, and the evaluation system includes resolution accuracy, rule effectiveness, governance efficiency, and business support dimensions; Having the functions of generating visual reports and pushing warning information.

[0014] Preferably, the multi-agent collaboration mechanism includes: The data flow path is: knowledge base→rule generation agent→intelligent analysis agent→quality evaluation agent→knowledge base; The instruction interaction includes that the rule generation agent sends a rule update instruction to the intelligent analysis agent, the intelligent analysis agent sends a governance result instruction to the quality evaluation agent, and the quality evaluation agent sends a rule optimization instruction to the rule generation agent; The evaluation result is fed back to the rule generation agent and the knowledge base through the quality evaluation agent, and full-link closed-loop optimization is realized.

[0015] According to the above technical solution, through the fusion of "large model fine-tuning + knowledge graph", complex address deep semantic understanding is realized, the accuracy of address normalization analysis and the accuracy of abnormal address identification are greatly improved, far exceeding traditional manual analysis; through the dynamic rule generation mechanism, the problem of incomplete coverage and lagging update of traditional static rules is solved, the rule update cycle is greatly shortened, and the address data dynamic change is adapted. At the same time, the business support capability is strengthened, so that the response time of the data supporting emergency repair after governance is effectively shortened, the error rate of marketing business handling is effectively reduced, and the high requirements of the new power system on data quality are met. In addition, the intelligent agent constructed by the method can adapt to the address data characteristics of different regional power companies, and through the fusion of multi-source data of the knowledge base and the fine-tuning of the large model, cross-regional reuse is realized.

[0016] Other features and advantages of the embodiments of the present application will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the embodiments of the present application together with the following specific embodiments, but do not constitute a limitation on the embodiments of the present application. In the drawings: Figure 1 is a flow schematic diagram of the construction method of the power customer address data governance intelligent agent provided by the present application; Figure 2 is a rule generation agent module structure schematic diagram provided by the present application; Figure 3 is a smart analysis agent module structure schematic diagram provided by the present application; Figure 4 is a quality evaluation agent module structure schematic diagram provided by the present application; Figure 5 is a multi-agent collaborative operation timing diagram provided by the present application. DETAILED DESCRIPTION

[0018] The specific implementation of the embodiments of the present application is described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiments of the present application, and is not used to limit the embodiments of the present application.

[0019] It should be noted that the acquisition, transmission, storage, use, processing, etc. of data in the technical solutions of the present application comply with the relevant provisions of laws and regulations. In the embodiments of the present application, some industry existing solutions, components, models, etc. may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0020] Currently, the field of artificial intelligence is experiencing a paradigm shift from a single model to an agent architecture. Its core advantage lies in giving AI the ability to autonomously plan, call tools, and close the task loop, which drives the evolution of application scenarios from dialogue interaction to productivity tools. Multi-agent collaborative architecture has become a focus of technical research.

[0021] Based on this, referring to Figures 1 to 5 , the present application provides a construction method of a power customer address data governance agent, which comprises: A power customer address data knowledge base is constructed through a three-layer construction logic of multi-source data fusion, large model semantic enhancement, and knowledge graph relationship modeling; A core agent is constructed, including a rule generation agent, an intelligent analysis agent, and a quality evaluation agent, each agent is run based on the knowledge base, and a modular design is adopted; A multi-agent collaborative mechanism is established, and through data flow, instruction interaction and result feedback, the collaborative operation and closed-loop optimization of the core agent are realized.

[0022] Specifically, the above-mentioned multi-source data fusion includes: collecting internal data, the internal data including power marketing system address data, GIS map data and historical governance data; collecting external data, the external data including standard address data of the Ministry of Civil Affairs, regional division data of the Statistics Bureau and map address data; cleaning and standardizing the collected data to form a standardized data set. The data cleaning includes removing duplicate / abnormal / missing data and correcting inconsistency, and the standardization annotation is annotated according to the "province-city-district-street-community-doorplate" level.

[0023] In the present embodiment, the above-mentioned large model semantic enhancement includes: fine-tuning training the standardized data set by using a large language model (Guangming Power Large Model, Deepseek) to optimize the understanding ability of the model to power address terminology, and ensure that the address standardization analysis accuracy is improved to more than 95%.

[0024] In order to avoid the problem of no quantization standard of the model effect after fine-tuning, the precision P, recall R and F1 score are introduced to evaluate the power term resolution capability, wherein, The precision formula for the accuracy of the model to resolve terms is: , Wherein, N corr is the number of correctly resolved power terms, N pred is the total number of power terms resolved by the model; The recall formula for the completeness of the model to cover power terms is: , Wherein, N true is the total number of power terms actually contained in the test set; The formula for F1 score obtained by combining precision and recall is: , The target threshold F1 is calculated to be greater than or equal to 0.95 (greater than or equal to 0.98 in emergency repair scene), and if it does not meet the requirement, power term samples need to be supplemented for further fine-tuning. For example, the test set contains 100 addresses with power terms, if the model resolves 98 terms, of which 95 are correct, then P=95 / 98≈96.9%, R=95 / 100=95%, F1=2×96.9%×95% / (96.9%+95%)≈95.9%, which meets the requirement.

[0025] In the embodiment, the above knowledge graph relationship modeling includes: Based on the fine-tuned large model, address entities and attributes are extracted; wherein, the address entities include "XX city XX district", "XX community" and the like, the attributes include "belonging to the area", "power type" and the like, and an address entity superior-inferior relationship graph is constructed, such as "XX community→XX street→XX district", to realize deep semantic understanding of complex addresses such as "XX road attached to No. 2 building of XX community".

[0026] In the present application, the rule generation agent constructed based on the power customer address data knowledge base has the functions of data access, large model learning, rule generation and rule output; wherein, The data access function is used to obtain historical governance data (such as abnormal address records in the past 3 years) and real-time address data (such as newly added customer addresses) in the knowledge base in real time; The large model learning function is used to learn the data in multiple dimensions by using the fine-tuned Guangming Power large model, and to identify address data problem patterns (such as level missing, logical conflict, and non-standard terms); The rule generation function is used to dynamically generate four types of verification rules of format checking, abnormality detection, semantic correction and membership relationship detection based on the learning results; wherein, Format verification rules such as "Address must contain four levels of province, city, district, and street" and "Door number must be a combination of numbers and letters"; Abnormality detection rules such as "Taiwan district number does not match the belonging street in the address" and "Logical conflict between energy consumption address and property address"; Semantic correction rules such as "XX village" unified to "XX community" and "Appendix No. 1" standardized to "Appendix No. 1"; Subordination detection rules such as checking whether "XX small community" belongs to "XX street" based on knowledge graph relationship.

[0027] The rule output function is used to automatically convert the generated rules into executable scripts (such as SQL scripts, Python scripts) and synchronize them to the intelligent analysis agent.

[0028] In the rule screening stage, to avoid the problem of invalid rules occupying computing resources, a rule confidence formula is introduced: , Where C(R) is the confidence of rule R, and rules with a confidence lower than 80% need to be relearned; N hit is the number of real abnormal addresses hit by the rule in the test set; N test is the total number of addresses in the rule test set; and α=1, β=2 are smoothing coefficients (to avoid confidence distortion when it is too small).

[0029] For example, a "Taiwan district number-street matching rule" test set contains 500 addresses, and hits 450 real abnormalities, so C(R)=(450+1) / (500+2)×100%≈89.8%, close to the threshold, and needs to supplement 20 abnormal samples to optimize to ≥90%.

[0030] In the rule iteration stage, to realize "effective rules are executed first", a dynamic weight update formula is introduced to adjust the priority based on the rule application effect: , Where W t (R) is the weight of rule R after the tth iteration, and the initial W_0=1; C t (R) is the actual confidence of rule R in the tth round; C avg (t) is the average confidence of all rules in the tth round; rules with a weight ≥1.2 are executed first, and rules with a weight <0.8 are put into the optimization queue.

[0031] For example, rule A has an initial weight of 1.0, a first round confidence of 95%, and an average confidence of all rules of 90%, so W1(A) = 1.0 x 95% / 90% ≈ 1.06, the priority is raised; rule B has a confidence of 80%, so W1(B) = 1.0 x 80% / 90% ≈ 0.89, which needs to be optimized.

[0032] In the present embodiment, the above-mentioned intelligent analysis intelligent agent structure comprises: a double-driven rule generation mode driven by autonomous learning and business demand; a full life cycle management function with problem discovery, task assignment, rectification verification and effect feedback; and for complex problems, a function of automatically analyzing problems and matching special sub-intelligent agents (for complex problems such as "cross-regional address name duplication", automatically analyzing problem list and matching special sub-intelligent agents such as "cross-regional address comparison sub-intelligent agent", realizing accurate problem management). Among them, The autonomous learning drive mines data line rules such as "the same address prefix in the same area" and data relationship rules such as "there is a corresponding relationship between the meter box number and the door number", and iteratively improves the rule library; The business demand drive receives the proposed rule compilation demand, for example, for emergency repair scene, the demand of priority checking community + door number integrity, etc., generates special checking rules after being analyzed by the large model; As for the problem full life cycle management function, first, discover the problem, execute the checking rule script, automatically identify the address data problem (such as "level missing" and "affiliation error"), and generate the problem list; then, assign the task, based on the problem type (such as assigning to the operation and maintenance team for the area attribution error, and assigning to the data entry team for the non-standard terminology), automatically push the task to the corresponding person in charge; then, verify the rectification, receive the rectification result feedback from the person in charge, compare the standard data in the knowledge base through the large model, and verify the effectiveness of the rectification; finally, feedback the effect, synchronize the rectification result to the rule generation intelligent agent, and use it for rule iteration optimization.

[0033] In the construction method of the power customer address data governance intelligent agent provided by the present application, the above-mentioned quality evaluation intelligent agent structure comprises: It has a multi-dimensional data collection function of collecting governance process data (problem identification quantity of intelligent analysis intelligent agent, rectification completion rate, etc.), data quality data (address resolution accuracy, rule coverage, abnormal address residual rate, etc.) and business impact data (emergency repair response time based on the data after governance, marketing business handling time, etc.); The collected data is quantitatively evaluated by using a large model (Guangming large model, Tongyi Qianwen, etc.) combined with a preset evaluation system, and the evaluation system includes resolution accuracy, rule effectiveness, governance efficiency and business support dimensions; as shown in Table 1: Table 1

[0034] Further, based on the evaluation results, a visual report containing percentage data quality scores, problem distribution heat maps, optimization suggestions, etc. is generated, supporting PDF / HTML format output. And when the indicators are below the threshold (such as abnormal recognition coverage <95%), automatically push the warning information to the data governance responsibility department (such as the operation and maintenance department, marketing department), and attach optimization suggestions (such as "supplement XX area map address data to improve rule coverage").

[0035] In the present embodiment, the multi-agent collaborative mechanism includes: The data flow path is: knowledge base → rule generation agent (data input) → intelligent analysis agent (rule input) → quality evaluation agent (governance data input) → knowledge base (evaluation result feedback for data update); The instruction interaction includes the rule generation agent sending a "rule update instruction" to the intelligent analysis agent; the intelligent analysis agent sending a "governance result instruction" to the quality evaluation agent; and the quality evaluation agent sending a "rule optimization instruction" to the rule generation agent, realizing precise instruction interaction; At the same time, the quality evaluation agent feeds back the evaluation results (such as "certain type of rule coverage is insufficient") to the rule generation agent to drive rule iteration; and synchronously feeds back to the knowledge base to update the standard address data, forming a "data-rule-governance-evaluation-optimization" full-link closed loop.

[0036] The following provides a specific embodiment to illustrate the construction method of the power customer address data governance intelligent agent provided by the present application: S1, preparation: Data collection: collect address data, civil affairs department five-level address data (2025 version), map address data; Large model selection: use the Guangming Power large model as the base model, input the above data for fine-tuning (training period 15 days, training sample size 5 million); Knowledge graph construction: based on the fine-tuned large model, extract 1.2 million address entities, and construct a "province-city-district-street-community-community-gate number" seven-level relationship graph.

[0037] S2, intelligent agent construction and operation: S21, rule generation agent operation: Input data: abnormal address records, 2025 January-June new customer addresses; Generate rules: automatically generate 8 format verification rules, 12 abnormal detection rules, 15 semantic correction rules, and 10 affiliation detection rules, and the rule script is synchronized to the intelligent analysis agent; S22, intelligent analysis agent operation: Dual-drive rule generation: 3 rules for "correspondence between transformer area number and street" are independently learned and discovered; 5 special rules are generated from the business requirements for "rural power grid area address management". Governance Execution: Address data was verified, abnormal addresses were identified, and rectification tasks were automatically assigned to the maintenance teams. 98% of the rectification work was completed. S23. Operation of the quality assessment agent: Data collection: Collect data on the governance process, data quality, and business data; Evaluation output: The generated address data quality score is 89 points, and a warning of "insufficient suburban rule coverage (88%)" is pushed out, suggesting that suburban map provider address data be supplemented.

[0038] S3. Implementation Results: Abnormal address identification accuracy: 98.7% (0.2 percentage points higher than expected); Processing time for abnormal data: 2.5 days (0.5 days lower than expected); Labor costs at the grassroots level: reduced by 65% ​​(from 50 people / day to 17 people / day); Business support: Emergency repair response time has been reduced from an average of 45 minutes to 35 minutes, and the error rate in marketing business processing has decreased from 3.2% to 0.8%.

[0039] Therefore, this invention achieves deep semantic understanding of complex addresses through the fusion of "large-scale model fine-tuning + knowledge graph," significantly improving the accuracy of address normalization parsing and abnormal address identification, far exceeding traditional manual parsing. A dynamic rule generation mechanism solves the problems of incomplete coverage and delayed updates in traditional static rules, greatly shortening the rule update cycle and adapting to dynamic changes in address data. Simultaneously, it strengthens business support capabilities, effectively shortening the response time for emergency repairs and reducing the error rate in marketing business processing, meeting the high data quality requirements of the new power system. Furthermore, the intelligent agent constructed by this method can adapt to the address data characteristics of power companies in different regions, achieving cross-regional reuse through multi-source data fusion of the knowledge base and large-scale model fine-tuning.

[0040] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0041] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0042] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0043] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0044] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0045] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, for storing instructions and data used and / or generated by the computing device. The memory can also include non-volatile memory, such as read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or nonvolatile random access memory (RAM) (e.g., fault tolerant RAM), for storing instructions and data used and / or generated by the computing device. The memory is an example of computer readable media.

[0046] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0047] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0048] The above only is an embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A method for constructing an intelligent agent for governing electricity customer address data, characterized in that, The method includes: A knowledge base for electricity customer address data is constructed through a three-layer construction logic of multi-source data fusion, large-model semantic enhancement, and knowledge graph relationship modeling. Construct core intelligent agents, including rule generation intelligent agents, intelligent analysis intelligent agents, and quality assessment intelligent agents. Each intelligent agent operates based on the knowledge base and adopts a modular design. A multi-agent collaborative mechanism is established to achieve the collaborative operation and closed-loop optimization of the core agents through data flow, command interaction, and result feedback.

2. The method for constructing an intelligent agent for power customer address data governance according to claim 1, characterized in that, The multi-source data fusion includes: Collect internal data, including electricity marketing system address data, GIS map data, and historical governance data; Collect external data, including standard address data from the Ministry of Civil Affairs, regional division data from the National Bureau of Statistics, and address data from map providers; The collected data is cleaned and standardized to form a standardized dataset.

3. The method for constructing an intelligent agent for power customer address data governance according to claim 2, characterized in that, The semantic enhancement of the large model includes: The standardized dataset was fine-tuned using a large language model to optimize the model's ability to understand electrical address terms. The power terminology parsing capability of the fine-tuned model was evaluated using precision (P), recall (R), and F1 score, ensuring that the F1 score reached a preset threshold.

4. The method for constructing an intelligent agent for power customer address data governance according to claim 3, characterized in that, Calculate the accuracy P according to formula (1). ，（1） Where, N corr The number of correctly parsed electrical terms, N pred The total number of electrical terms analyzed by the model; Calculate the recall rate R according to formula (2). ，（2） Where, N true The total number of electrical terms actually contained in the test set; Calculate the F1 score according to formula (3). ，（3） The preset threshold Not less than 0.

95.

5. The method for constructing an intelligent agent for power customer address data governance according to claim 3, characterized in that, The knowledge graph relationship modeling includes: Based on the finely tuned large model, extract the address entities and attributes; Construct a hierarchy graph of address entities to achieve deep semantic understanding of complex addresses.

6. The method for constructing an intelligent agent for power customer address data governance according to claim 1, characterized in that, The rule-generating intelligent agent constructed based on a knowledge base of electricity customer address data possesses functions such as data access, large-scale model learning, rule generation, and rule output; among which, The data access function is used to acquire historical governance data and real-time address data from the knowledge base in real time. The large model learning function is used to learn the data from multiple dimensions using the fine-tuned Guangming Power large model, and to identify address data problem patterns. The rule generation function is used to dynamically generate four types of verification rules based on the learning results: format verification, anomaly detection, semantic error correction, and membership detection. The rule output function is used to automatically convert the generated rules into executable scripts and synchronize them to the intelligent analysis agent.

7. The method for constructing an intelligent agent for power customer address data governance according to claim 6, characterized in that, According to formula (4), rule confidence is introduced to filter the generated rules. ，（4） Where C(R) is the confidence level of rule R, and rules with a confidence level below 80% need to be relearned; N hit N represents the number of real anomaly addresses that were hit by the rule in the test set. test The total number of addresses in the rule test set; α=1, β=2 are smoothing coefficients; Based on formula (5), the rule weight update algorithm is used to dynamically iterate and optimize the rule. ，（5） Among them, W t (R) represents the weight of rule R after the t-th iteration, initially W_0=1; C t (R) represents the actual confidence level of rule R in round t; C avg (t) represents the average confidence level of all rules in round t; rules with a weight ≥ 1.2 are executed first, and rules with a weight < 0.8 are entered into the optimization queue.

8. The method for constructing an intelligent agent for power customer address data governance according to claim 1, characterized in that, The intelligent analytical agent is constructed as follows: It adopts a dual-drive rule generation mode driven by self-learning and business needs; It has full lifecycle management functions, including problem discovery, task assignment, rectification verification, and effect feedback; For complex problems, it has the function of automatically analyzing the problem and matching it with a specific sub-agent.

9. The method for constructing an intelligent agent for power customer address data governance according to claim 1, characterized in that, The quality assessment agent is constructed as follows: It has multi-dimensional data collection capabilities, including data on the governance process, data quality, and business impact. A large model is used in conjunction with a pre-set evaluation system to quantitatively evaluate the collected data. The evaluation system includes dimensions such as parsing accuracy, rule effectiveness, governance efficiency, and business support. It has the function of generating visual reports and pushing early warning information.

10. The method for constructing an intelligent agent for power customer address data governance according to claim 1, characterized in that, The multi-agent cooperative mechanism includes: The data flow path is: Knowledge Base → Rule Generation Agent → Intelligent Analysis Agent → Quality Assessment Agent → Knowledge Base; The instruction interaction includes the rule generation agent sending rule update instructions to the intelligent analysis agent, the intelligent analysis agent sending governance result instructions to the quality assessment agent, and the quality assessment agent sending rule optimization instructions to the rule generation agent. The quality assessment agent feeds back the assessment results to the rule generation agent and the knowledge base, achieving closed-loop optimization across the entire chain.

Images