Digital intelligent agent construction method and system for shortening power supply index decision period

Abstract

The invention discloses a digital intelligent agent construction method and system for shortening a power supply index decision period. The method comprises the following steps: cleaning, aligning and fusing power supply service data; receiving a request input by a user in a natural language, and performing semantic analysis on the natural language request to identify a user intention; retrieving and fusing multi-source data related to the intention from the processed data according to the identified user intention; based on the intelligent diagnosis model, performing quantitative evaluation and anomaly diagnosis on the retrieved and fused multi-source data to generate a diagnosis result; according to the diagnosis result, matching a recommended governance strategy from a preset solution knowledge base, and performing quantitative pre-evaluation on the implementation effect of the recommended governance strategy; and outputting the multi-mode response content, and tracking the actual index change to complete the management closed loop. According to the method, the decision-making efficiency of the power supply indexes is improved by constructing a full-process automatic decision-making system of data intelligent fusion, intention driving, intelligent diagnosis and closed-loop feedback.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and specifically to a method and system for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators. Background Technology

[0002] With the continuous advancement of digital transformation in the power industry, power supply stations and urban distribution teams, as core business units, need to handle multiple tasks, including indicator monitoring, transformer area anomaly diagnosis, work quality assessment, and execution effect tracking. Currently, business data is scattered across multiple independent platforms such as the electricity consumption information collection system, equipment operation and maintenance management system, and work order system, lacking a unified integration channel. For example, when calculating the monthly data collection success rate of a power supply station, it is necessary to export raw data from the electricity consumption information collection system, retrieve fault records from the equipment operation and maintenance system, and then manually summarize and analyze the data, with each such operation taking more than 2 hours. In terms of transformer area anomaly diagnosis, such as troubleshooting high loss or low voltage issues, it currently relies heavily on the personal experience of operation and maintenance personnel. It is usually necessary to first check the lines and then verify the load data, resulting in a significant trial-and-error process. For example, when a transformer area experienced high loss due to three-phase load imbalance, line inspection was prioritized on-site, resulting in a waste of human resources.

[0003] Traditional business processes lack automated collaboration mechanisms. From identifying anomalies to formulating optimization strategies, multiple steps such as data collection, analysis, and report generation require manual intervention. Data collection takes 2.5 hours, analysis 3 hours, and report generation another 2.5 hours, resulting in a total decision-making cycle of up to 8 hours. Even after anomaly handling, effectiveness evaluation still relies on manual comparison with historical data, taking approximately 1 hour, which cannot meet the high timeliness requirements of emergency business scenarios such as peak summer demand. Furthermore, existing technologies do not fully cover all power supply stations and distribution teams and lack stable operation and maintenance mechanisms. Some remote power supply stations still use a purely manual ledger management model, lacking rapid response channels when system failures occur, resulting in slow overall progress in digital transformation and difficulty in effectively supporting the marketing department's strategic transformation from a traditional operation and maintenance model to a digital operation model.

[0004] The main drawbacks of existing technologies can be summarized as follows:

[0005] The failure to build a unified data integration system resulted in business data being scattered across multiple systems. This reliance on manual cross-platform data retrieval and aggregation led to data collection and aggregation taking more than 2.5 hours and was prone to data bias due to manual operation, failing to provide effective support for rapid decision-making.

[0006] Current technology relies on human experience for the diagnosis and evaluation of abnormalities in transformer areas. The average time for a single abnormality diagnosis is 10 minutes, while the evaluation takes 1 hour, which is inefficient. Furthermore, the diagnostic results are greatly affected by the subjective experience of the personnel, resulting in a high deviation rate.

[0007] Existing methods lack a complete closed-loop management mechanism that covers the entire process from indicator monitoring, anomaly diagnosis, effect evaluation to strategy optimization. Data from each business link needs to be manually entered twice, resulting in serious information gaps and making it impossible to intuitively and dynamically present the actual effectiveness of governance measures. Summary of the Invention

[0008] The technical problem to be solved by this invention is to overcome the problems of long decision-making cycle, low efficiency and poor accuracy of power supply indicators caused by data silos, reliance on human experience and lack of closed-loop mechanism in the prior art. The purpose is to provide a digital intelligent agent construction method to shorten the decision-making cycle of power supply indicators, and solve the key problem of how to realize the full-process automation, intelligence and closed-loop management of power supply indicator decision-making.

[0009] This invention is achieved through the following technical solution:

[0010] Firstly, this application provides a method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators, comprising the following steps:

[0011] Cleaning, alignment, and fusion processing are performed on power supply service data from multiple heterogeneous business systems;

[0012] Receive requests from users in natural language and perform semantic parsing on the requests to identify user intent;

[0013] Based on the identified user intent, retrieve and fuse multi-source data related to the intent from the fused data;

[0014] Based on the intelligent diagnostic model, the retrieved and fused multi-source data is quantitatively evaluated and anomaly diagnosed to generate diagnostic results including anomaly type, cause and confidence level.

[0015] Based on the diagnostic results, recommended governance strategies are matched from a pre-set solution knowledge base, and the implementation effect of the recommended governance strategies is quantitatively pre-evaluated using a power distribution network simulation model.

[0016] The output includes a multimodal response containing diagnostic results, recommended governance strategies, and pre-assessment results, and tracks the actual indicator changes after the implementation of the recommended governance strategies to complete the management closed loop from monitoring to assessment feedback.

[0017] A further optimization scheme involves cleaning, aligning, and fusing power supply service data from multiple heterogeneous business systems, specifically including the following steps:

[0018] By using both automated interface synchronization and manual template import, power supply service data from the electricity information collection system, equipment operation and maintenance system and work order system are accessed to obtain the raw data after access.

[0019] A data cleaning pipeline is executed on the raw data after access to obtain cleaned data;

[0020] The cleaned data from different systems with different timestamp formats are uniformly converted into a standard format, and a unique mapping index is established for key data to obtain standardized data with unified timestamps.

[0021] Based on the transformer substation number and timestamp dimension, the time-series monitoring data in the unified timestamp standard data and the basic attribute data in the relational database are subjected to correlation verification and fusion to obtain the fused data.

[0022] A further optimized solution involves receiving user requests input in natural language and performing semantic parsing on the requests to identify user intent, specifically including the following steps:

[0023] Receive user requests via natural language text or structured form input, and obtain natural language requests;

[0024] Using a large language model fine-tuned based on corpus instructions from the power supply business domain, deep semantic parsing is performed on the natural language request to identify user intent;

[0025] Based on the identified user intent, generate corresponding data processing strategy instructions.

[0026] A further optimized solution involves generating corresponding data processing strategy instructions based on the identified user intent, specifically including the following steps:

[0027] If the identified user intent is an indicator query, a data retrieval instruction is generated. The data retrieval instruction is used to obtain indicator data within a specified time and spatial range from the fused data.

[0028] If the identified user intent is an anomaly diagnosis, a diagnostic analysis instruction is generated. The diagnostic analysis instruction is used to trigger the rule engine and / or machine learning anomaly diagnosis model to perform quantitative evaluation and root cause analysis on the data of the relevant transformer area.

[0029] If the identified user intent is a solution recommendation, a strategy matching instruction is generated. The strategy matching instruction is used to match the recommended governance strategy from the solution knowledge base based on the diagnostic results and trigger the distribution network simulation model to perform a pre-evaluation of the effect.

[0030] A further optimized solution involves retrieving and fusing multi-source data related to the identified user intent from the fused data, specifically including the following steps:

[0031] Based on the key entity information contained in the user intent, a standardized data retrieval instruction is generated; the key entity information includes spatial entities, temporal entities, and indicator entities.

[0032] Based on the data retrieval instructions, data related to the key entity information is retrieved in parallel from time-series databases, relational databases, and vector databases to obtain an initial retrieval result set from different databases;

[0033] The multi-source data in the initial search result set are subjected to correlation verification and fusion, and the data is aligned based on the station area number and timestamp to generate complete multi-source data corresponding to the user's intent.

[0034] A further optimized solution involves using an intelligent diagnostic model to quantitatively evaluate and diagnose anomalies in the retrieved and fused multi-source data, generating diagnostic results that include anomaly type, cause, and confidence level. This specifically includes the following steps:

[0035] Based on the multi-source data, key business indicator values ​​are automatically calculated, and the indicator values ​​are compared with preset thresholds to generate indicator health assessment results.

[0036] Based on the rule engine, predefined business rules are executed to perform preliminary diagnosis of abnormal indicators and generate rule diagnosis results that include the first diagnostic cause and the first confidence level.

[0037] When the confidence level of the rule-based diagnostic result is lower than a preset threshold or the abnormal pattern of the indicator does not conform to the predefined rule, the machine learning anomaly diagnostic model is triggered to analyze the historical load curve and operating data, and generate a model diagnostic result containing a second diagnostic cause and a second confidence level.

[0038] By combining the rule-based diagnostic results with the model-based diagnostic results, and performing correlation analysis with transformer topology data, load data, and historical work orders, a structured diagnostic result is generated that includes the anomaly type, main cause, scope of impact, and overall confidence level.

[0039] A further optimized solution involves matching recommended governance strategies from a pre-defined solution knowledge base, specifically including the following steps:

[0040] Based on the causes of the anomalies in the diagnostic results, and combined with the current transformer capacity, remaining space, and renovation budget of the transformer area, the optimal solution is recommended from a solution knowledge base containing hardware renovation and flexible regulation solutions using a weighted scoring algorithm.

[0041] A further optimization is that the multimodal response content includes at least one or more combinations of text analysis reports, structured data tables, and trend charts;

[0042] The actual changes in the tracking strategy after its implementation are specifically included: comparing the changes in line loss rate, voltage qualification rate and data acquisition success rate before and after the implementation of the strategy on a weekly, monthly or quarterly basis, and presenting the governance effect in the form of charts.

[0043] Secondly, this application provides a digital intelligent agent system for shortening the decision-making cycle of power supply indicators, used to implement the method described above, the system comprising:

[0044] The data governance module is connected to the electricity information collection system, equipment operation and maintenance system and work order system, and is used to clean, align and merge power supply service data from the multiple heterogeneous business systems.

[0045] The intent recognition module is used to receive a request input by the user in natural language and to perform semantic parsing on the request to identify the user's intent.

[0046] The data retrieval and fusion module is communicatively connected to the data governance module and the intent recognition module, and is used to retrieve and fuse relevant multi-source data from the fused data according to the identified user intent.

[0047] The intelligent diagnostic module is communicatively connected to the data retrieval and fusion module, and is used to perform quantitative evaluation and anomaly diagnosis on the retrieved and fused data based on the intelligent diagnostic model, and generate diagnostic results.

[0048] The strategy recommendation and pre-evaluation module is communicatively connected to the intelligent diagnosis module, and is used to match recommended governance strategies based on the diagnosis results and perform effect pre-evaluation using a power distribution network simulation model.

[0049] The closed-loop management module is communicatively connected to the strategy recommendation and pre-evaluation module. It is used to output multimodal response content and track the actual indicator changes after the strategy is implemented in order to complete the management closed loop.

[0050] Thirdly, this application provides a computer-readable storage medium storing a digital intelligent agent construction program for shortening the decision cycle of power supply indicators, wherein when the digital intelligent agent construction program for shortening the decision cycle of power supply indicators is executed by a processor, the steps of the digital intelligent agent construction method for shortening the decision cycle of power supply indicators as described above are implemented.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] By using automated data processing technology to replace traditional manual operations, the problem of decision delay caused by the difficulty of integrating multi-source heterogeneous data is effectively solved.

[0053] Through natural language interaction and intelligent parsing, the barrier to entry for using professional systems has been lowered, enabling a shift from passive querying to proactive insight.

[0054] The collaborative diagnostic mechanism combining rule engines and machine learning models not only significantly improves the efficiency of anomaly identification and root cause analysis, but also overcomes the subjective limitations of human experience.

[0055] Through strategy pre-assessment and dynamic effect tracking, a complete "monitoring-diagnosis-decision-verification" management closed loop has been formed, which significantly enhances the agility, accuracy and sustainable optimization capabilities of power supply service management. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0057] Figure 1 A flowchart of a digital intelligent agent construction method for shortening the decision-making cycle of power supply indicators provided in an embodiment of this application;

[0058] Figure 2 This is a functional block diagram of the digital intelligent agent construction system for shortening the decision-making cycle of power supply indicators provided in the embodiments of this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0060] Firstly, such as Figure 1As shown, this application provides a method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators, including the following steps:

[0061] Step S1: Clean, align, and merge power supply service data from multiple heterogeneous business systems, such as the electricity information collection system, equipment operation and maintenance system, and work order system, to provide a high-quality data foundation for subsequent analysis;

[0062] Step S2: Receive a request input by the user in natural language, and perform semantic parsing on the request to identify the user's intent;

[0063] Step S3: Based on the identified user intent, retrieve and fuse multi-source data related to the intent from the fused data;

[0064] Step S4: Based on the intelligent diagnostic model, perform quantitative evaluation and anomaly diagnosis on the retrieved and fused multi-source data to generate diagnostic results including anomaly type, cause and confidence level;

[0065] Step S5: Based on the diagnostic results, a recommended governance strategy is matched from the preset solution knowledge base, and the implementation effect of the recommended governance strategy is quantitatively pre-evaluated using a power distribution network simulation model;

[0066] Step S6: Output a multimodal response containing diagnostic results, recommended governance strategies, and pre-assessment results, and track the actual indicator changes after the implementation of the recommended governance strategies to complete the management closed loop from monitoring to assessment feedback.

[0067] In one embodiment, this application provides a method for constructing a digital intelligent agent to shorten the decision-making cycle for power supply indicators. This method constructs a digital intelligent agent system for power supply services, achieving fully automated closed-loop management based on a three-layer architecture of "interaction unit - intelligent agent engine unit - data source unit." Its core lies in the unified integration and intelligent reasoning of multi-source heterogeneous data through the intelligent agent engine unit, and in achieving natural and efficient human-machine collaboration through the interaction unit.

[0068] The three-layer architecture forms an organic whole, realizing an automated pipeline from user intent perception to final policy generation, specifically including the following steps:

[0069] Step S1 is supported by the data source unit, which cleans, aligns and merges power supply service data from multiple heterogeneous business systems such as the power consumption information collection system, equipment operation and maintenance system and work order system, to provide a high-quality data foundation for subsequent analysis.

[0070] In step S2, the interaction unit receives the user request and the intelligent agent engine unit parses it. The interaction unit receives the user's request input in natural language and performs semantic parsing on the request to identify the user's intent.

[0071] Step S3 is driven by the intelligent agent engine unit, which retrieves and merges multi-source data related to the identified user intent from the data fusion processing of the data source unit.

[0072] Step S4 is executed by the intelligent agent engine unit. Based on the intelligent diagnostic model, the retrieved and fused multi-source data is quantitatively evaluated and anomaly diagnosed to generate a diagnostic result that includes anomaly type, cause and confidence level.

[0073] Step S5 is executed by the intelligent agent engine unit. Based on the diagnostic results, a recommended governance strategy is matched from the preset solution knowledge base, and the implementation effect of the recommended governance strategy is quantitatively pre-evaluated using the power distribution network simulation model.

[0074] Step S6 outputs the results from the interaction unit and is tracked by the intelligent agent engine unit. It outputs multimodal response content including diagnostic results, recommended governance strategies and pre-evaluation results, and tracks the actual indicator changes after the implementation of the recommended governance strategies, so as to complete the management closed loop from monitoring to evaluation feedback.

[0075] In one embodiment, this application constructs and utilizes a data source unit storage architecture to achieve the cleaning, alignment, and fusion processing of multi-source heterogeneous power supply service data in step S1, providing data support for intelligent decision-making.

[0076] The data source unit adopts a hybrid storage architecture to achieve unified governance and efficient access to multi-source data, specifically including:

[0077] In terms of monitoring data management, time-series databases are used to efficiently store and query monitoring data such as transformer voltage, current and load power that are strongly correlated with time. For example, time-series databases such as InfluxDB can be used to achieve fast writing and aggregated querying of such data.

[0078] In terms of structured data management, relational databases are used to store various types of structured business information, including basic attributes of the transformer area such as number, capacity and topology, work order data such as work order number, user information and submission time, and the indicator threshold library required by the system. This type of data is suitable for management through relational databases such as MySQL.

[0079] In terms of knowledge retrieval support, vector databases are used to store vector embeddings generated from unstructured knowledge resources such as documents, reports and cases after transformation, in order to support efficient retrieval based on semantic similarity. Dedicated vector databases such as Chroma can be used to implement this function.

[0080] In terms of performance optimization, deploying a caching database to cache frequently accessed metric data, user session states, and hot knowledge content can significantly reduce the access pressure on the backend database and improve the overall system response performance. For example, Redis can be introduced as a caching solution.

[0081] In one embodiment, step S1 involves cleaning, aligning, and fusing power supply service data from multiple heterogeneous service systems, specifically including the following sub-steps:

[0082] Step S11: Through a dual-channel approach of automated interface synchronization and manual template import, power supply service data from the electricity information collection system, equipment operation and maintenance management system, and work order system is accessed to obtain raw data. The automated interface connects directly to the database via a pre-defined RESTful API interface, enabling connections to the electricity information collection system, equipment operation and maintenance system, GIS system, etc., and utilizes the Quartz scheduled task framework to configure an automatic data synchronization task to be triggered every morning, extracting incremental data to the data platform. Manual import addresses legacy system data or temporary reports not covered by the interface. The system provides Excel templates with pre-defined fields and validation rules. After users download, fill in, and upload, the system's backend service automatically performs format compliance checks and basic logic validations (e.g., checking if the transformer area number exists). Once validation passes, the data is stored in the corresponding database.

[0083] Step S12: Perform data cleaning pipeline processing on the raw data after access to obtain cleaned data; the specific cleaning and treatment process includes: missing value processing, that is, filling in missing points in the monitoring data by using time series interpolation or referring to data of the same type of transformer area; outlier removal, that is, identifying and removing abnormal peak data caused by sensor failure or other reasons based on the 3σ principle or box plot method.

[0084] Step S13: Convert the cleaned data from different systems with varying timestamp formats into a standard time format, and establish a unique mapping index for key data fields to obtain standardized data with unified timestamps; specifically, perform data alignment, that is, convert timestamps from heterogeneous systems into the ISO 8601 standard format, and establish a globally unique mapping index for key data; wherein, the key data is the station area number;

[0085] Step S14: Using the transformer substation number and timestamp as the core dimensions, the time-series monitoring data in the standardized data after timestamp unification is correlated and deeply fused with the basic attribute data stored in the relational database to finally obtain the fused data; specifically, through data fusion and correlation technology, based on the transformer substation number and timestamp dimensions, the monitoring data in the time-series database and the basic attribute data in the relational database are correlated and fused to form a complete data slice that can be used for subsequent analysis.

[0086] In one embodiment, step S2 involves receiving a request input by the user in natural language and performing semantic parsing on the request to identify the user's intent. This specifically includes the following sub-steps:

[0087] Step S21: Receive user input. The user inputs a request through the web or mobile interface provided by the interaction unit, either in natural language text or a structured form. For example, the user can query by entering a natural language statement such as "Query the low voltage distribution areas of XX power supply station last month", or submit the request using a structured form provided by the system.

[0088] Step S22: Deep Semantic Parsing and Intent Recognition. Using a large language model fine-tuned based on power supply service corpus instructions, deep semantic parsing is performed on the natural language request to identify its core intent. Typical intents include categories such as indicator query, anomaly diagnosis, solution recommendation, or knowledge question answering.

[0089] Step S23: Generate data processing strategy instructions. Based on the identified user intent, generate corresponding data processing strategy instructions.

[0090] Steps S21 and S22 described above are mainly completed collaboratively by the intent recognition module in the interaction unit and the intelligent agent engine unit. The intent recognition module is built upon a finely tuned domain-specific large language model, such as a model based on Baidu Wenxin Qianfan for domain adaptation. This module is responsible for core tasks such as semantic parsing and intent recognition.

[0091] In one embodiment, step S23: generating a corresponding data processing strategy instruction based on the identified user intent, specifically includes the following steps:

[0092] Step S231: If the identified user intent is an indicator query, a data retrieval instruction is generated. The data retrieval instruction is used to obtain indicator data within a specified time and spatial range from the fused data.

[0093] Step S232: If the identified user intent is an anomaly diagnosis, a diagnostic analysis instruction is generated. The diagnostic analysis instruction is used to trigger the rule engine and / or machine learning anomaly diagnosis model to perform quantitative evaluation and root cause analysis on the data of the relevant substation.

[0094] Step S233: If the identified user intent is a solution recommendation, a strategy matching instruction is generated. This instruction is used to match a recommended governance strategy from the solution knowledge base based on the diagnostic results and trigger the distribution network simulation model to perform a pre-evaluation of the effect. The strategy decision module invokes the corresponding processing strategy based on the intent recognition results.

[0095] In one embodiment, step S3: retrieving and fusing multi-source data related to the identified user intent from the fused data, specifically includes the following steps:

[0096] Step S31: Generate standardized data retrieval instructions based on the key entity information contained in the user intent; the key entity information includes spatial entities, temporal entities, and indicator entities.

[0097] Step S32: Based on the data retrieval instructions, retrieve data related to the key entity information from time-series databases, relational databases, and vector databases in parallel to obtain an initial retrieval result set from different databases.

[0098] Step S33: Perform correlation verification and fusion on the multi-source data in the initial retrieval result set, align the data based on the station area number and timestamp, and generate complete multi-source data corresponding to the user's intent.

[0099] In one embodiment, step S4: This involves quantitatively evaluating and diagnosing anomalies in the retrieved and fused multi-source data based on an intelligent diagnostic model, and generating diagnostic results including anomaly type, cause, and confidence level. Specifically, this includes the following sub-steps:

[0100] Step S41: Quantitative Assessment and Indicator Calculation. Based on the multi-source data, the system automatically calculates key business indicator values ​​and compares them with preset thresholds to generate indicator health assessment results. The system automatically calculates key business indicators such as data collection success rate and line loss rate. Data collection success rate = (number of successfully collected users / total number of users to be collected) × 100%, and line loss rate = (power supply - electricity sales) / power supply × 100%. The system calculates the deviation rate between indicator values ​​and preset thresholds in real time and automatically generates an indicator health dashboard.

[0101] Step S42: Rule Engine Diagnosis. Based on the rule engine, predefined business rules are executed to perform a preliminary diagnosis of abnormal indicators, generating a rule diagnosis result that includes the first diagnostic cause and the first confidence level. Specifically, the Drools rule engine is used to solidify expert experience into executable business rules. For example, when the three-phase imbalance of a transformer area is greater than 15%, it is set to be due to three-phase load imbalance; or when the transformer load rate of a transformer area is greater than 80% and the duration exceeds 2 hours, it is set to be due to insufficient transformer capacity.

[0102] Step S43: Model-Assisted Diagnosis. When the confidence level of the rule-based diagnostic result is lower than a preset threshold or the abnormal indicator pattern does not conform to the predefined rules, the machine learning anomaly diagnostic model is triggered to analyze historical load curves and operational data, generating a model diagnostic result that includes a second diagnostic cause and a second confidence level. For complex and hidden fault modes, a deep learning model built on the TensorFlow or PyTorch framework is used for assisted diagnosis, such as an LSTM anomaly detection model or a graph neural network model trained on historical data, to identify anomaly patterns that are difficult to cover by rules.

[0103] Step S44: Generate structured diagnostic results. The diagnostic results from both the rule-based approach and the model are combined with the transformer area topology data, load data, and historical work orders for correlation analysis to generate structured diagnostic results that include the anomaly type, main cause, scope of impact, and overall confidence level. The diagnostic module automatically outputs complete diagnostic conclusions by correlating the multi-source analysis results with the actual operating data of the transformer area.

[0104] In one embodiment, step S5: Based on the diagnostic results, a recommended governance strategy is matched from a preset solution knowledge base, and the implementation effect of the recommended governance strategy is quantitatively pre-evaluated using a power distribution network simulation model. This specifically includes the following steps:

[0105] Step S51: Based on the causes of the anomalies in the diagnostic results, and combined with the actual operating scenario of the transformer substation, including the current transformer capacity, remaining physical space, and budget constraints for upgrades, the system uses a weighted scoring algorithm to match and recommend the optimal solution from a pre-set solution knowledge base. This knowledge base includes hardware upgrade solutions such as transformer capacity expansion and line upgrades, as well as flexible adjustment solutions such as reactive power compensation and three-phase load balancing adjustments. Based on the diagnostic results and the specific conditions of the transformer substation, the system performs a multi-dimensional evaluation using weighted scoring to recommend the most suitable mitigation solution.

[0106] Step S52: Input the key parameters of the recommended governance strategy into the distribution network simulation model to quantitatively pre-evaluate the effect of the strategy implementation. For example, input the new capacity parameters after the distribution transformer capacity increase into the simulation engine based on the forward-backward substitution method to simulate the voltage distribution and line loss changes of key nodes in the back-end area after the scheme is implemented, and quantitatively predict the improvement effect of the scheme, such as outputting specific predicted values ​​such as an average voltage increase of 5 volts at the user end.

[0107] In one embodiment, step S6 involves outputting multimodal response content and tracking the policy effect to complete the management loop, specifically including the following sub-steps:

[0108] Step S61: Output multimodal response content containing diagnostic results, recommended governance strategies, and pre-assessment results. This content includes at least one or more combinations of text analysis reports, structured data tables, and trend charts. The interaction unit dynamically generates content in multiple formats from the processing results of the agent engine, such as text analysis reports, structured data tables, trend line charts, and geographic distribution maps of transformer substations, to adapt to the cognitive preferences of different users.

[0109] Step S62: Track the actual changes in indicators after the implementation of the recommended governance strategy, specifically by week, month, or quarter. Compare the changes in core indicators such as line loss rate, voltage qualification rate, and data acquisition success rate before and after the strategy implementation, and present the governance effect intuitively in chart form. After the solution is implemented, the system automatically tracks the subsequent indicator changes of the target area, providing data feedback for strategy optimization through multi-dimensional comparisons.

[0110] In addition, the system also ensures the integrity and intelligence of closed-loop management through multiple functional modules;

[0111] Specifically, the observation and auditing module records all user operations, system decision-making processes, data flow, and model inference logs throughout the entire process, forming an unalterable audit trail chain;

[0112] The user profile module dynamically records and updates user role attributes, historical query preferences, and frequently used function modules to form a personalized user profile.

[0113] When processing knowledge-based question answering, the retrieval enhancement generation and orchestration module first retrieves relevant document fragments from the vectorized knowledge base, and then submits these fragments along with the original question to the large model to generate accurate and verifiable answers.

[0114] The data and index building module is responsible for cleaning, feature extraction, and vectorization of the incoming structured and unstructured data, and for building efficient inverted and vector indexes to support fast retrieval.

[0115] To enable those skilled in the art to better understand the technical solution of the present invention, the following describes in detail the complete application process of the method described in the present invention, taking the diagnosis and treatment of high-loss anomalies in transformer substations as an example, in conjunction with a typical application scenario.

[0116] User A (an operations and maintenance personnel at a power supply station) logs into the system through the interactive unit and enters a natural language question in the dialog box: "Analyze the causes of transformer substations in XX power supply station with a line loss rate exceeding 10% in June 2024, and recommend a remediation plan." This request is sent to the intelligent agent engine unit, where its internal intent recognition module parses it. Based on a finely tuned large language model, the module accurately identifies the user's core intent as "anomaly diagnosis" and "solution recommendation," and extracts key entities, including the spatial entity (XX power supply station), the temporal entity (June 2024), and the indicator entity (line loss rate > 10%).

[0117] Based on the identified intent, the strategy decision-making module generates standardized data retrieval instructions. The data source unit responds to these instructions, retrieving in parallel the basic information and topological relationships of all transformer substations under the target power supply from the relational database, obtaining daily electricity consumption data for these substations in a specified month from the time-series database to calculate the average daily line loss rate, and retrieving relevant fault or maintenance records from the historical work order database for the same period. The system then performs correlation verification and fusion on the initial retrieval results from different databases, ultimately generating a fused dataset containing a complete list of transformer substations, line loss data, basic attributes, and related work orders.

[0118] Upon entering the diagnostic phase, the system first identifies target transformer areas with persistently high line loss rates based on fused data. The rule engine (Drools) then activates, executing predefined business rules for preliminary diagnosis. For example, if a transformer area has a high line loss rate but a low load rate, the rule engine indicates a risk of "metering anomaly or electricity theft"; if three-phase imbalance also exceeds the standard, "three-phase load imbalance" is added as a suspected cause. Simultaneously, for complex patterns that rules cannot cover, deep learning models (such as LSTM) are triggered to perform in-depth analysis of the target transformer area's historical load curves, further identifying abnormal electricity consumption characteristics highly correlated with "electricity theft." The system integrates the outputs of the rule engine and the deep learning model, and performs correlation analysis with the transformer area topology, load data, and historical work orders to generate a structured diagnostic result concluding that "the primary cause of high line loss in this transformer area is suspected electricity theft, and the secondary cause is three-phase load imbalance."

[0119] In the strategy recommendation phase, the system matches two candidate solutions, "on-site electricity inspection" and "adjusting three-phase load," from its built-in solution knowledge base based on the diagnostic results. Considering the actual scenario of this transformer area being located in a suburban area with a high risk of electricity theft, the system prioritizes "on-site electricity inspection" as the primary solution using a weighted scoring algorithm. To further evaluate the mitigation potential, the system inputs the parameters of the "three-phase load balancing adjustment" solution into the distribution network simulation model for pre-evaluation. The simulation results show that implementing this auxiliary solution is expected to further reduce the line loss rate in the affected area by approximately 2%, and this result is output as a supplementary recommendation.

[0120] Finally, based on the combined intent of "anomaly diagnosis + solution recommendation," the system uses a dedicated response template to generate multimodal response content. This content includes a text report clearly stating the diagnostic conclusions, recommended solutions, and pre-assessment results; a data table listing high-loss transformer areas and their key indicators; and a chart showing the trend of line loss rates between the target transformer area and normal transformer areas. After compliance verification, this response content is finally presented to user A through the interactive unit, completing the fully intelligent processing from problem identification to solution output.

[0121] This invention achieves seamless automation of the entire process from data integration, intent recognition, intelligent reasoning to response generation by constructing a three-layer intelligent agent architecture consisting of an interaction unit, an intelligent agent engine unit, and a data source unit. Based on a hybrid storage design of the data source unit, this architecture combines time-series databases, relational databases, and vector databases, effectively supporting unified governance and efficient access to multi-source heterogeneous data. Simultaneously, through a dual-channel mechanism for data access and processing—combining automatic API synchronization and manual import—it fundamentally solves the data silo problem existing in traditional business systems.

[0122] At the intelligent analysis level, the system employs an intelligent assessment and anomaly diagnosis mechanism that combines a rule engine and a quantitative model. By solidifying expert experience into executable rules and introducing machine learning models to handle complex fault modes, it significantly improves the accuracy and efficiency of diagnosis, effectively replacing the traditional subjective judgment method that relies on human experience. At the interaction level, leveraging the RAG retrieval enhancement generation technology and NL2SQL natural language to query technology of the intelligent question-answering assistant, it achieves intelligent conversion from user natural language input to structured data queries, greatly reducing the system's usage threshold and expanding its application scenarios.

[0123] In addition, the system has established a comprehensive security and compliance guarantee mechanism. Through data encryption, identity authentication, and full-process audit log recording, it ensures that all operations are traceable and auditable, meeting the industry's high standards for data security and system stability, and providing a solid guarantee for the reliable operation and compliance management of the system.

[0124] Secondly, such as Figure 2As shown, this application provides a digital intelligent agent system for shortening the decision-making cycle of power supply indicators, used to implement the method described above. The system includes a data governance module 100, an intent recognition module 200, a data retrieval and fusion module 300, an intelligent diagnosis module 400, a strategy recommendation and pre-evaluation module 500, and a closed-loop management module 600.

[0125] The data governance module 100 is communicatively connected to the electricity information collection system, equipment operation and maintenance system and work order system, and is used to clean, align and merge power supply service data from the multiple heterogeneous business systems.

[0126] The intent recognition module 200 is used to receive a request input by a user in natural language and to perform semantic parsing on the request to recognize the user's intent;

[0127] The data retrieval and fusion module 300 is communicatively connected to the data governance module 100 and the intent recognition module 200, and is used to retrieve and fuse relevant multi-source data from the fused data according to the identified user intent.

[0128] The intelligent diagnosis module 400 is communicatively connected to the data retrieval and fusion module 300, and is used to perform quantitative evaluation and anomaly diagnosis on the retrieved and fused data based on the intelligent diagnosis model, and generate diagnostic results.

[0129] The strategy recommendation and pre-evaluation module 500 is communicatively connected to the intelligent diagnosis module 400, and is used to match recommended governance strategies based on the diagnosis results and perform effect pre-evaluation using a distribution network simulation model;

[0130] The closed-loop management module 600 is communicatively connected to the strategy recommendation and pre-evaluation module 500, and is used to output multimodal response content and track the actual indicator changes after the strategy is implemented to complete the management closed loop.

[0131] The functions of each module in the above-mentioned digital intelligent agent construction system for shortening the decision cycle of power supply indicators correspond to the steps in the above-mentioned digital intelligent agent construction method embodiment for shortening the decision cycle of power supply indicators. Their functions and implementation processes will not be described in detail here.

[0132] Thirdly, this application provides a digital intelligent agent construction device for shortening the decision-making cycle of power supply indicators. The digital intelligent agent construction device for shortening the decision-making cycle of power supply indicators can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0133] In this embodiment of the application, the digital intelligent agent construction device for shortening the decision cycle of power supply indicators may include a processor, a memory, a communication interface, and a communication bus.

[0134] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0135] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces. These interfaces are used for interconnecting devices within the intelligent digital agent building equipment that shortens the power supply indicator decision-making cycle, as well as interfaces for interconnecting the intelligent digital agent building equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0136] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0137] The processor can be a general-purpose processor, which can call the intelligent agent construction program for shortening the power supply indicator decision cycle stored in the memory and execute the intelligent agent construction method for shortening the power supply indicator decision cycle provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the intelligent agent construction program for shortening the power supply indicator decision cycle is called can refer to the various embodiments of the intelligent agent construction method for shortening the power supply indicator decision cycle of this application, and will not be repeated here.

[0138] Fourthly, embodiments of this application also provide a readable storage medium.

[0139] The present application stores a digital intelligent agent construction program for shortening the decision cycle of power supply indicators on a readable storage medium. When the digital intelligent agent construction program for shortening the decision cycle of power supply indicators is executed by a processor, it implements the steps of the digital intelligent agent construction method for shortening the decision cycle of power supply indicators as described above.

[0140] The method implemented when the digital intelligent agent construction program for shortening the power supply index decision cycle is executed can refer to the various embodiments of the digital intelligent agent construction method for shortening the power supply index decision cycle of this application, and will not be repeated here.

[0141] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators, characterized in that, Includes the following steps: Cleaning, alignment, and fusion processing are performed on power supply service data from multiple heterogeneous business systems; Receive requests from users in natural language and perform semantic parsing on the requests to identify user intent; Based on the identified user intent, retrieve and fuse multi-source data related to the intent from the fused data; Based on the intelligent diagnostic model, the retrieved and fused multi-source data is quantitatively evaluated and anomaly diagnosed to generate diagnostic results including anomaly type, cause and confidence level. Based on the diagnostic results, recommended governance strategies are matched from a pre-set solution knowledge base, and the implementation effect of the recommended governance strategies is quantitatively pre-evaluated using a power distribution network simulation model. The output includes a multimodal response containing diagnostic results, recommended governance strategies, and pre-assessment results, and tracks the actual indicator changes after the implementation of the recommended governance strategies to complete the management closed loop from monitoring to assessment feedback.

2. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The process of cleaning, aligning, and fusing power supply service data from multiple heterogeneous business systems specifically includes the following steps: By using both automated interface synchronization and manual template import, power supply service data from the electricity information collection system, equipment operation and maintenance system and work order system are accessed to obtain the raw data after access. A data cleaning pipeline is executed on the raw data after access to obtain cleaned data; The cleaned data from different systems with different timestamp formats are uniformly converted into a standard format, and a unique mapping index is established for key data to obtain standardized data with unified timestamps. Based on the transformer substation number and timestamp dimension, the time-series monitoring data in the unified timestamp standard data and the basic attribute data in the relational database are subjected to correlation verification and fusion to obtain the fused data.

3. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The process of receiving a user's request input in natural language and performing semantic parsing on the request to identify the user's intent specifically includes the following steps: Receive user requests via natural language text or structured form input, and obtain natural language requests; Using a large language model fine-tuned based on corpus instructions from the power supply business domain, deep semantic parsing is performed on the natural language request to identify user intent; Based on the identified user intent, generate corresponding data processing strategy instructions.

4. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 3, characterized in that, The step of generating corresponding data processing strategy instructions based on the identified user intent specifically includes the following steps: If the identified user intent is an indicator query, a data retrieval instruction is generated. The data retrieval instruction is used to obtain indicator data within a specified time and spatial range from the fused data. If the identified user intent is an anomaly diagnosis, a diagnostic analysis instruction is generated. The diagnostic analysis instruction is used to trigger the rule engine and / or machine learning anomaly diagnosis model to perform quantitative evaluation and root cause analysis on the data of the relevant transformer area. If the identified user intent is a solution recommendation, a strategy matching instruction is generated. The strategy matching instruction is used to match the recommended governance strategy from the solution knowledge base based on the diagnostic results and trigger the distribution network simulation model to perform a pre-evaluation of the effect.

5. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The step of retrieving and fusing multi-source data related to the identified user intent from the fused data specifically includes the following steps: Based on the key entity information contained in the user intent, a standardized data retrieval instruction is generated; the key entity information includes spatial entities, temporal entities, and indicator entities. Based on the data retrieval instructions, data related to the key entity information is retrieved in parallel from time-series databases, relational databases, and vector databases to obtain an initial retrieval result set from different databases; The multi-source data in the initial search result set are subjected to correlation verification and fusion, and the data is aligned based on the station area number and timestamp to generate complete multi-source data corresponding to the user's intent.

6. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The intelligent diagnostic model is used to quantitatively evaluate and diagnose anomalies in the retrieved and fused multi-source data, generating diagnostic results that include anomaly type, cause, and confidence level. Specifically, this includes the following steps: Based on the multi-source data, key business indicator values ​​are automatically calculated, and the indicator values ​​are compared with preset thresholds to generate indicator health assessment results. Based on the rule engine, predefined business rules are executed to perform preliminary diagnosis of abnormal indicators and generate rule diagnosis results that include the first diagnostic cause and the first confidence level. When the confidence level of the rule-based diagnostic result is lower than a preset threshold or the abnormal pattern of the indicator does not conform to the predefined rule, the machine learning anomaly diagnostic model is triggered to analyze the historical load curve and operating data, and generate a model diagnostic result containing a second diagnostic cause and a second confidence level. By combining the rule-based diagnostic results with the model-based diagnostic results, and performing correlation analysis with transformer topology data, load data, and historical work orders, a structured diagnostic result is generated that includes the anomaly type, main cause, scope of impact, and overall confidence level.

7. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The step of matching and recommending governance strategies from a preset solution knowledge base specifically includes the following steps: Based on the causes of the anomalies in the diagnostic results, and combined with the current transformer capacity, remaining space, and renovation budget of the transformer area, the optimal solution is recommended from a solution knowledge base containing hardware renovation and flexible regulation solutions using a weighted scoring algorithm.

8. The method for constructing a digital intelligent agent to shorten the decision-making cycle of power supply indicators according to claim 1, characterized in that, The multimodal response content includes at least one or more combinations of text analysis reports, structured data tables, and trend charts; The actual changes in the tracking strategy after its implementation are specifically included: comparing the changes in line loss rate, voltage qualification rate and data acquisition success rate before and after the implementation of the strategy on a weekly, monthly or quarterly basis, and presenting the governance effect in the form of charts.

9. A digital intelligent agent system for shortening the decision-making cycle of power supply indicators, used to implement the method as described in any one of claims 1 to 8, characterized in that, The system includes: The data governance module is connected to the electricity information collection system, equipment operation and maintenance system and work order system, and is used to clean, align and merge power supply service data from the multiple heterogeneous business systems. The intent recognition module is used to receive a request input by the user in natural language and to perform semantic parsing on the request to identify the user's intent. The data retrieval and fusion module is communicatively connected to the data governance module and the intent recognition module, and is used to retrieve and fuse relevant multi-source data from the fused data according to the identified user intent. The intelligent diagnostic module is communicatively connected to the data retrieval and fusion module, and is used to perform quantitative evaluation and anomaly diagnosis on the retrieved and fused data based on the intelligent diagnostic model, and generate diagnostic results. The strategy recommendation and pre-evaluation module is communicatively connected to the intelligent diagnosis module, and is used to match recommended governance strategies based on the diagnosis results and perform effect pre-evaluation using a power distribution network simulation model. The closed-loop management module is communicatively connected to the strategy recommendation and pre-evaluation module. It is used to output multimodal response content and track the actual indicator changes after the strategy is implemented in order to complete the management closed loop.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a digital intelligent agent construction program for shortening the power supply index decision cycle, wherein when the digital intelligent agent construction program for shortening the power supply index decision cycle is executed by a processor, the steps of the digital intelligent agent construction method for shortening the power supply index decision cycle as described in any one of claims 1 to 8 are implemented.

Images