Automatic heat supply customer service work order filling method and system based on large model

Abstract

The invention is suitable for the technical field of heat supply management, and provides a heat supply customer service work order automatic filling system based on a large model, which comprises a collaborative agent cluster used for processing user requests in parallel, a dynamic knowledge network module, a digital twin access and simulation module and a work order decision and active intervention engine, the invention discloses an automatic heat supply customer service work order filling method based on a large model. The method comprises the following steps: accessing a user multi-modal service request through a full channel; cooperating with the agent cluster to work in parallel; the dynamic knowledge network performs multi-hop reasoning according to the output to generate a preliminary diagnosis reasoning chain; the work order decision engine packages the user information, the diagnosis chain and the situation data into an initial work order, and calls a digital twinborn body to carry out simulation deduction on the fault influence; the work order processing whole process data is captured by the self-evolution closed-loop management module; the active service of handling before complaint is realized, the perception and satisfaction of the user are greatly improved, and derivative complaints are reduced from the source.

Description

Technical Field

[0001] This invention relates to the field of heating management technology, and more specifically, to a method and system for automatically filling in heating customer service work orders based on a large model. Background Technology

[0002] With the maturation of large language model technology, its application in customer service is becoming increasingly widespread. In the heating industry, some companies have already launched large language models for answering hotlines and automatically generating work orders, significantly improving service efficiency. Related patents have also disclosed customer service systems that integrate large models and knowledge graphs, as well as general voice form filling technology. Meanwhile, leading heating companies are using technologies such as ERP and digital twins to reduce work order generation at the source, pursuing the goal of "near-zero work orders."

[0003] However, existing technical solutions have significant limitations: First, they mostly rely on large, single-unit models for end-to-end processing, which is insufficient in terms of professionalism, interpretability, and decision reliability when faced with the complex physical characteristics of heating systems, which are characterized by strong coupling and strong interference. Second, existing knowledge bases are mostly static or passively updated, unable to learn and evolve rapidly from real-time interaction and data streams. Most importantly, existing solutions are essentially still "passive response"—user reports a problem—the system dispatches a work order—failing to form a deep closed loop of "perception-simulation-decision-intervention" with the heating physical system, and thus failing to achieve a paradigm shift from "solving existing problems" to "preventing future problems."

[0004] Therefore, there is an urgent need for a new generation of intelligent customer service work order system that can deeply integrate the information space (customer service interaction, knowledge) and the physical space (heating system operation), and has the capabilities of collaborative reasoning, simulation prediction and proactive intervention. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies: there is an urgent need for a new generation of intelligent customer service work order system that can deeply integrate information space and physical space and has collaborative reasoning, simulation prediction and proactive intervention capabilities. Therefore, this invention proposes an automatic form filling method and system for heating customer service work orders based on a large model.

[0006] The specific technical solution is: a heating customer service work order auto-filling system based on a large model, including:

[0007] The collaborative intelligent agent cluster includes an intent parsing intelligent agent, a professional knowledge retrieval intelligent agent, and a context-aware intelligent agent: the intent parsing intelligent agent is used to identify the deep intent and emotion of the user's original request; the professional knowledge retrieval intelligent agent is used to retrieve accurate fault databases, maintenance plans, and policy provisions from a dynamic knowledge network; and the context-aware intelligent agent is used to access and analyze the status data, meteorological data, and work order outbreak trends in the same spatiotemporal dimension of the real-time heating digital twin.

[0008] The dynamic knowledge network module consists of interconnected and coupled topology maps of the heating system, fault evolution maps, and maintenance case maps, and is designed with an adaptive adjustment mechanism for the edge weights of the maps based on real-time work order feedback.

[0009] The digital twin access and simulation module connects to the Internet of Things data of the physical heating system, builds and runs a multi-layer digital twin model of "station-network-household", and is used to simulate the impact of faults and evaluate intervention strategies.

[0010] The work order decision-making and proactive intervention engine, connecting the collaborative intelligent agent cluster, dynamic knowledge network module, and digital twin access and simulation module, is used for:

[0011] By integrating the outputs of multiple agents, an enhanced structured chemical document containing a root cause reasoning chain is generated;

[0012] Digital twins are used to simulate intervention strategies and predict the impact of different maintenance plans on the overall thermal balance of the network.

[0013] While confirming passive work orders, proactive warning work orders for potentially affected users or adjustment instructions for the system are generated based on context awareness.

[0014] The self-evolutionary closed-loop management module is used to transform the full-process data of each completed work order, including the difference between the twin simulation prediction and the actual result, into training samples and knowledge graph update materials, driving the continuous evolution of the system.

[0015] In the technical solution of the present invention, the collaborative intelligent agent cluster is constructed using an intelligent agent framework based on a large language model. Each intelligent agent performs specialized division of labor through dedicated prompt word engineering and fine-tuning, and an arbitration intelligent agent evaluates the confidence of each output and resolves conflicts, ultimately forming a unified decision.

[0016] In the technical solution of this invention, the collaborative intelligent agent cluster comprises an intent parsing intelligent agent, a professional knowledge retrieval intelligent agent, and a context-aware intelligent agent. The intent parsing intelligent agent receives the speech-to-text transcription of the user's original request, identifies and outputs the deep business intent and user emotional state of the request through a large language model fine-tuned by the history of heating customer service dialogue. The professional knowledge retrieval intelligent agent, based on the business intent output by the intent parsing intelligent agent, retrieves in parallel from the dynamic knowledge network fault diagnosis library, maintenance solution library, and relevant policy provisions matching the current request, and outputs a structured set of candidate solutions. The context-aware intelligent agent, in parallel, accesses and analyzes system status data, external meteorological data, and work order outbreak situation data based on the same spatiotemporal dimension of the geographic information system provided by the real-time heating digital twin, and outputs the environment and system-level context of the request.

[0017] The collaborative intelligent agent cluster also includes an arbitration intelligent agent, which is used to receive the parallel output results of the intent parsing intelligent agent, the professional knowledge retrieval intelligent agent and the context-aware intelligent agent, evaluate the confidence of each result and resolve conflicts, and integrate them to form an analysis report pointing to unified decision-making.

[0018] The intent parsing agent, professional knowledge retrieval agent, and context-aware agent are functionally specialized through a dedicated prompt word engineering process. These dedicated prompt words define the roles, task objectives, output formats, and analysis dimensions of each agent. The context-aware agent accesses real-time heating digital twin status data, including real-time pressure, temperature, and flow operating parameters and their historical curves for the heating station, pipeline network, and in-home equipment associated with the user's address. The analysis report output by the arbitration agent includes at least a core fault diagnosis confirmed through reasoning, a multi-source evidence chain supporting the diagnosis, a comprehensive assessment of the work order's urgency, and preliminary processing suggestions.

[0019] In the technical solution of the present invention, in the dynamic knowledge network module, the fault evolution graph represents the chain of "symptom-suspected component-confirmed cause-solution-required spare parts and skills" through a graph structure. The weight of the edge is dynamically strengthened or weakened according to the resolution verification results of historical work orders, so as to realize the self-evaluation and ranking of knowledge utility.

[0020] In the technical solution of the present invention, the digital twin access and simulation module is used to quickly locate and inject fault parameters in the digital twin according to the fault point described in the work order, simulate and calculate the potential impact range and degree of the fault on the room temperature of upstream and downstream users, and output the list of affected users as a related work order recommendation.

[0021] In the technical solution of this invention, the proactive intervention function of the work order decision and proactive intervention engine includes:

[0022] Horizontal proactive early warning: When a work order for a blockage in a unit's riser is detected, a preventative work order is automatically sent to all residents supplied by that riser to inform them that the room temperature may be too low.

[0023] Vertical proactive regulation: When a sudden surge of non-heating work orders in a certain area occurs within a short period of time, and the digital twin simulation confirms that the output of the heating station is insufficient, an adjustment suggestion work order is automatically generated and sent to the heat energy production scheduling system.

[0024] Another objective of this invention is to provide an automatic form filling method for heating customer service work orders based on a large model, applicable to any of the systems described above, comprising the following steps:

[0025] Step 1: Access user multimodal service requests through all channels;

[0026] Step 2: Collaborative intelligent agents work in parallel: the intent parsing agent extracts requests and emotions; the professional knowledge retrieval agent obtains candidate solutions from the dynamic knowledge network; the context-aware agent obtains real-time data from user tables, weather, and surrounding work order status.

[0027] Step 3: The dynamic knowledge network performs multi-hop reasoning based on the output of Step 2 to generate a preliminary diagnostic reasoning chain;

[0028] Step 4: The work order decision engine encapsulates user information, diagnostic chain, and contextual data into an initial work order, and calls the digital twin to simulate and extrapolate the impact of the fault.

[0029] Step 5: Based on the simulation results, the decision engine performs at least one of the following operations:

[0030] a. Improve and confirm the dispatch of repair work orders to the original users;

[0031] b. Generate and dispatch proactive warning work orders for potentially affected users identified in the simulation;

[0032] c. Generate and send equipment parameter adjustment suggestion work orders to the production scheduling system;

[0033] Step Six: The entire process data of work order processing is captured by the self-evolution closed-loop management module, which is used to optimize the intelligent agent, update the knowledge network, and calibrate the digital twin model.

[0034] In the technical solution of the present invention, the simulation and deduction in step four involves adjusting the valve opening or pipe resistance coefficient of the corresponding household to a fault state in the digital twin, running the hydraulic and thermal model, and calculating the predicted room temperature drop and the time to reach a new equilibrium for other households in the building, especially those on the upper and lower floors and those in the same location.

[0035] In the technical solution of the present invention, the calibration of the digital twin model in step six involves comparing the fault point finally resolved by the maintenance work order and the meter data retested after maintenance with the fault point and room temperature change predicted by simulation, calculating the simulation error, and using the error data to automatically fine-tune the relevant model parameters of the digital twin.

[0036] In the technical solution of this invention, the final generated work order includes an interpretable view, which visually displays to customer service personnel or maintenance personnel: the user's original description, the analysis conclusions of each agent, the reasoning path of the knowledge graph, and the simulation influence range of the digital twin, thereby improving the credibility and efficiency of work order processing.

[0037] Compared with the prior art, the present invention can achieve the following:

[0038] 1. By combining knowledge reasoning with physical simulation, work order diagnosis is upgraded from experience-based guessing to simulation verification, resulting in more accurate root cause identification; proactive service before complaints are received significantly improves user perception and satisfaction, reducing derivative complaints at the source; multi-agent parallel processing and automated proactive work order dispatch transform the customer service center from a cost center into a value creation and risk warning center; dynamic knowledge networks and self-evolution mechanisms enable the automatic and efficient accumulation and optimization of enterprise knowledge assets, reducing reliance on expert experience; and an efficient bridge is built between the customer service side and the production side, enabling customer demands to be directly and scientifically fed back to guide the operation and control of the heating network.

[0039] 2. Unlike existing technologies that use a single model, this invention proposes a cluster of intelligent agents with clearly defined functions such as parsing, retrieval, and perception; it imitates the collaborative model of human expert teams, improves the accuracy and efficiency of handling complex professional problems through specialized division of labor, and ensures decision consistency through an arbitration mechanism, thus achieving better interpretability and robustness in terms of technology.

[0040] 3. Unlike traditional knowledge graphs, the dynamic knowledge network constructed in this invention emphasizes the adaptive adjustment of graph edge weights; the verification results of each work order are fed back, strengthening effective reasoning paths and weakening ineffective paths, enabling the knowledge system to have the ability to evolve through use and disuse, becoming a continuously learning knowledge organism.

[0041] 4. The digital twin access and simulation module is the fundamental difference between this invention and all existing customer service work order systems. Existing systems process information flow, while this invention deeply integrates a heating digital twin. When a fault description is received, the system not only reasons in the information space, but also injects faults and simulates their impact in the virtual mirror of the physical system, the digital twin, thereby quantitatively predicting the scope of the fault's impact and providing a scientific basis for proactive intervention.

[0042] Based on simulation prediction, the system not only generates work orders to resolve reported repair issues, but also proactively generates two types of new work orders: one is an early warning service work order sent to potentially affected users, transforming passive complaint handling into proactive notification; the other is an adjustment instruction work order sent to the production scheduling system, attempting to eliminate potential problems at the source. This achieves reverse driving from the customer service end to the production end, forming a complete control closed loop. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the internal collaboration and decision-making process of a collaborative intelligent agent cluster. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0045] In this embodiment of the invention, the heating customer service work order automatic filling system based on a large model includes:

[0046] A collaborative intelligent agent cluster for parallel processing of user requests includes an intent-interpreting agent, a professional knowledge retrieval agent, and a context-aware agent: the intent-interpreting agent identifies the deep intent and sentiment of the user's original request; the professional knowledge retrieval agent retrieves accurate fault databases, maintenance plans, and policy provisions from a dynamic knowledge network; and the context-aware agent accesses and analyzes the status data, meteorological data, and work order outbreak trends in the same spatiotemporal dimension of the real-time heating digital twin (see details). Figure 1 );

[0047] Unlike existing technologies that use a single model, this invention proposes a cluster of intelligent agents with clearly defined functions such as parsing, retrieval, and perception. It mimics the collaborative model of human expert teams, improves the accuracy and efficiency of handling complex professional problems through specialized division of labor, and ensures decision consistency through an arbitration mechanism, thus achieving better interpretability and robustness in terms of technology.

[0048] The dynamic knowledge network module consists of interconnected and coupled topology maps of the heating system, fault evolution maps, and maintenance case maps, and is designed with an adaptive adjustment mechanism for the edge weights of the maps based on real-time work order feedback.

[0049] Unlike traditional knowledge graphs, the dynamic knowledge network constructed in this invention emphasizes the adaptive adjustment of graph edge weights; the verification results of each work order are fed back, strengthening effective reasoning paths and weakening ineffective paths, enabling the knowledge system to have the evolutionary ability of use and disuse, and becoming a continuously learning knowledge organism;

[0050] The digital twin access and simulation module connects to the Internet of Things data of the physical heating system, builds and runs a multi-layer digital twin model of "station-network-household", and is used to simulate the impact of faults and evaluate intervention strategies.

[0051] The digital twin access and simulation module is the fundamental difference between this invention and all existing customer service ticket systems. Existing systems process information flow, while this invention deeply integrates a heating digital twin. When a fault description is received, the system not only reasons in the information space, but also injects faults and simulates their impact in the virtual mirror of the physical system, the digital twin, thereby quantitatively predicting the scope of the fault's impact and providing a scientific basis for proactive intervention.

[0052] Based on simulation prediction, the system not only generates work orders to resolve reported repair issues, but also proactively generates two new types of work orders: one is an early warning service work order sent to potentially affected users, transforming passive complaint handling into proactive notification; the other is an adjustment instruction work order sent to the production scheduling system, attempting to eliminate potential problems at the source. This achieves reverse driving from the customer service end to the production end, forming a complete control closed loop.

[0053] The work order decision-making and proactive intervention engine, connecting the collaborative intelligent agent cluster, dynamic knowledge network module, and digital twin access and simulation module, is used for:

[0054] By integrating the outputs of multiple agents, an enhanced structured chemical document containing a root cause reasoning chain is generated;

[0055] Digital twins are used to simulate intervention strategies and predict the impact of different maintenance plans on the overall thermal balance of the network.

[0056] While confirming passive work orders, proactive warning work orders for potentially affected users or adjustment instructions for the system are generated based on context awareness.

[0057] The self-evolutionary closed-loop management module is used to transform the full-process data of each completed work order, including the difference between the twin simulation prediction and the actual result, into training samples and knowledge graph update materials, driving the system to continuously evolve.

[0058] This invention designs a rigorous data feedback and model calibration mechanism; the difference between the work order execution results and the simulation predictions is used to calibrate the digital twin model, making it infinitely close to the real physical system; the data throughout the process is used to train the agent and optimize the knowledge network; thus forming a powerful self-evolutionary closed loop, and the system efficiency continuously increases with the time of use.

[0059] By combining knowledge reasoning with physical simulation, work order diagnosis is upgraded from experience-based guessing to simulation verification, resulting in more accurate root cause identification. Proactive services, addressing issues before they arise, significantly improve user perception and satisfaction, reducing derivative complaints at the source. Multi-agent parallel processing and automated proactive work order dispatch transform the customer service center from a cost center into a value creation and risk warning center. Dynamic knowledge networks and self-evolution mechanisms enable the automatic and efficient accumulation and optimization of enterprise knowledge assets, reducing reliance on expert experience. An efficient bridge is built between the customer service and production sides, allowing customer feedback to directly and scientifically guide the operation and control of the heating network.

[0060] In this embodiment of the invention, the collaborative intelligent agent cluster is constructed using an intelligent agent framework based on a large language model. Each intelligent agent performs specialized tasks through dedicated prompt word engineering and fine-tuning, and an arbitration intelligent agent evaluates the confidence of each agent's output and resolves conflicts, ultimately forming a unified decision.

[0061] In this embodiment of the invention, the collaborative intelligent agent cluster comprises an intent parsing intelligent agent, a professional knowledge retrieval intelligent agent, and a context-aware intelligent agent. The intent parsing intelligent agent receives the speech-to-text transcription of the user's original request, identifies and outputs the deep business intent and user emotional state of the request through a large language model fine-tuned by the history of heating customer service dialogue. The professional knowledge retrieval intelligent agent, based on the business intent output by the intent parsing intelligent agent, retrieves in parallel from the dynamic knowledge network fault diagnosis library, maintenance solution library, and relevant policy provisions matching the current request, and outputs a structured set of candidate solutions. The context-aware intelligent agent, in parallel, accesses and analyzes system status data, external meteorological data, and work order outbreak situation data based on the same spatiotemporal dimension of the geographic information system provided by the real-time heating digital twin, and outputs the environment and system-level context of the request.

[0062] The collaborative intelligent agent cluster also includes an arbitration intelligent agent, which is used to receive the parallel output results of the intent parsing intelligent agent, the professional knowledge retrieval intelligent agent and the context-aware intelligent agent, evaluate the confidence of each result and resolve conflicts, and integrate them to form an analysis report pointing to unified decision-making.

[0063] The intent parsing agent, professional knowledge retrieval agent, and context-aware agent are functionally specialized through a dedicated prompt word engineering process. These dedicated prompt words define the roles, task objectives, output formats, and analysis dimensions of each agent. The context-aware agent accesses real-time heating digital twin status data, including real-time pressure, temperature, and flow operating parameters and their historical curves for the heating station, pipeline network, and in-home equipment associated with the user's address. The analysis report output by the arbitration agent includes at least a core fault diagnosis confirmed through reasoning, a multi-source evidence chain supporting the diagnosis, a comprehensive assessment of the work order's urgency, and preliminary processing suggestions.

[0064] In this embodiment of the invention, the fault evolution graph in the dynamic knowledge network module represents the chain of "symptom-suspected component-confirmed cause-solution-required spare parts and skills" through a graph structure. The weight of the edges is dynamically strengthened or weakened according to the resolution verification results of historical work orders, which is used to realize the self-evaluation and ranking of knowledge utility.

[0065] In this embodiment of the invention, the digital twin access and simulation module is used to quickly locate and inject fault parameters in the digital twin based on the fault point described in the work order, simulate and calculate the potential impact range and degree of the fault on the room temperature of upstream and downstream users, and output the list of affected users as a recommended related work order.

[0066] In this embodiment of the invention, the proactive intervention function of the work order decision and proactive intervention engine includes:

[0067] Horizontal proactive early warning: When a work order for a blockage in a unit's riser is detected, a preventative work order is automatically sent to all residents supplied by that riser to inform them that the room temperature may be too low.

[0068] Vertical proactive regulation: When a sudden surge of non-heating work orders in a certain area occurs within a short period of time, and the digital twin simulation confirms that the output of the heating station is insufficient, an adjustment suggestion work order is automatically generated and sent to the heat energy production scheduling system.

[0069] In this embodiment of the invention, a method for automatically filling out heating customer service work orders based on a large model, applied to any of the systems described above, includes the following steps:

[0070] Step 1: Access user multimodal service requests through all channels;

[0071] Step 2: Collaborative intelligent agents work in parallel: the intent parsing agent extracts requests and emotions; the professional knowledge retrieval agent obtains candidate solutions from the dynamic knowledge network; the context-aware agent obtains real-time data from user tables, weather, and surrounding work order status.

[0072] Step 3: The dynamic knowledge network performs multi-hop reasoning based on the output of Step 2 to generate a preliminary diagnostic reasoning chain;

[0073] Step 4: The work order decision engine encapsulates user information, diagnostic chain, and contextual data into an initial work order, and calls the digital twin to simulate and extrapolate the impact of the fault.

[0074] Step 5: Based on the simulation results, the decision engine performs at least one of the following operations:

[0075] a. Improve and confirm the dispatch of repair work orders to the original users;

[0076] b. Generate and dispatch proactive warning work orders for potentially affected users identified in the simulation;

[0077] c. Generate and send equipment parameter adjustment suggestion work orders to the production scheduling system;

[0078] Step Six: The entire process data of work order processing is captured by the self-evolution closed-loop management module, which is used to optimize the intelligent agent, update the knowledge network, and calibrate the digital twin model.

[0079] In this embodiment of the invention, the multi-hop reasoning in step three specifically includes: based on the user's description of "the upper half of the radiator is hot and the lower half is cold", combined with the household meter data of "low flow", reasoning is performed in the fault evolution graph along the path of "symptom: upper part hot and lower part cold -> suspected component: radiator / manifold -> confirmed cause: air blockage / foreign object blockage / poor circulation" and then combined with the context of "3 similar work orders in the same unit recently", the weight of "unit riser circulation problem" is significantly increased.

[0080] The simulation in step four involves adjusting the valve opening or pipe resistance coefficient of the corresponding household in the digital twin to a fault state, running the hydraulic and thermal model, and calculating the predicted room temperature drop and the time to reach a new equilibrium for other households in the building, especially those on the upper and lower floors and those in the same location.

[0081] The calibration of the digital twin model in step six involves comparing the fault points finally resolved by the maintenance work order and the meter data retested after maintenance with the fault points and room temperature changes predicted by simulation, calculating the simulation error, and using the error data to automatically fine-tune the relevant model parameters of the digital twin.

[0082] The final generated work order includes an interpretable view that visually displays to customer service or maintenance personnel: the user's original description, the analysis conclusions of each agent, the reasoning path of the knowledge graph, and the simulation impact range of the digital twin, thereby improving the credibility and efficiency of work order processing.

[0083] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A heating customer service work order automatic filling system based on a large model, characterized in that, include: A cluster of collaborative intelligent agents for parallel processing of user requests, the cluster of collaborative intelligent agents includes an intent parsing agent, a professional knowledge retrieval agent, and a context-aware agent: the intent parsing agent is used to identify the deep intent and emotion of the user's original request; The professional knowledge retrieval agent is used to retrieve accurate fault databases, maintenance plans, and policy provisions from a dynamic knowledge network; the context-aware agent is used to access and analyze the status data, meteorological data, and work order outbreak trends of the real-time heating digital twin. The dynamic knowledge network module is composed of interconnected and coupled heating system topology map, fault evolution map and maintenance case map, and is designed with a map edge weight adaptive adjustment mechanism based on real-time work order feedback. The digital twin access and simulation module connects to the Internet of Things data of the physical heating system, builds and runs a multi-layer digital twin model of "station-network-household", and is used to simulate the impact of faults and evaluate intervention strategies. The work order decision-making and proactive intervention engine, connecting the collaborative intelligent agent cluster, dynamic knowledge network module, and digital twin access and simulation module, is used for: By integrating the outputs of multiple agents, an enhanced structured chemical document containing a root cause reasoning chain is generated; Digital twins are used to simulate intervention strategies and predict the impact of different maintenance plans on the overall thermal balance of the network. While confirming passive work orders, proactive warning work orders for potentially affected users or adjustment instructions for the system are generated based on context awareness. The self-evolutionary closed-loop management module is used to transform the full-process data of each completed work order, including the difference between the twin simulation prediction and the actual result, into training samples and knowledge graph update materials, driving the continuous evolution of the system.

2. The automatic form filling system for heating customer service work orders based on a large model as described in claim 1, characterized in that, The collaborative intelligent agent cluster is constructed using an intelligent agent framework based on a large language model. Each intelligent agent performs specialized tasks through dedicated prompt word engineering and fine-tuning, and an arbitration intelligent agent evaluates the confidence of each agent's output and resolves conflicts, ultimately forming a unified decision.

3. The automatic form filling system for heating customer service work orders based on a large model as described in claim 1 or 2, characterized in that, The collaborative intelligent agent cluster comprises an intent parsing intelligent agent, a professional knowledge retrieval intelligent agent, and a context-aware intelligent agent. The intent parsing intelligent agent receives the speech-to-text transcription of the user's original request, and identifies and outputs the deep business intent and user emotional state of the request through a large language model fine-tuned by the history of heating customer service dialogue. The professional knowledge retrieval intelligent agent, based on the business intent output by the intent parsing intelligent agent, retrieves in parallel from the dynamic knowledge network fault diagnosis library, maintenance solution library, and relevant policy provisions that match the current request, and outputs a structured set of candidate solutions. The context-aware intelligent agent is used to access and analyze in parallel the system status data, external meteorological data, and work order outbreak situation data based on the same spatiotemporal dimension of the geographic information system provided by the real-time heating digital twin, and outputs the environment and system-level context of the request. The collaborative intelligent agent cluster also includes an arbitration intelligent agent, which is used to receive the parallel output results of the intent parsing intelligent agent, the professional knowledge retrieval intelligent agent and the context-aware intelligent agent, evaluate the confidence of each result and resolve conflicts, and integrate them to form an analysis report pointing to unified decision-making. The intent parsing agent, professional knowledge retrieval agent, and context-aware agent are functionally specialized through a dedicated prompt word engineering process. These dedicated prompt words define the roles, task objectives, output formats, and analysis dimensions of each agent. The context-aware agent accesses real-time heating digital twin status data, including real-time pressure, temperature, and flow operating parameters and their historical curves for the heating station, pipeline network, and in-home equipment associated with the user's address. The analysis report output by the arbitration agent includes at least a core fault diagnosis confirmed through reasoning, a multi-source evidence chain supporting the diagnosis, a comprehensive assessment of the work order's urgency, and preliminary processing suggestions.

4. The automatic form filling system for heating customer service work orders based on a large model as described in claim 1, characterized in that, In the dynamic knowledge network module, the fault evolution graph represents the chain of "symptom-suspected component-confirmed cause-solution-required spare parts and skills" through a graph structure. The weight of the edges is dynamically strengthened or weakened according to the resolution verification results of historical work orders, which is used to realize the self-evaluation and ranking of knowledge utility.

5. The automatic form filling system for heating customer service work orders based on a large model as described in claim 1, characterized in that, The digital twin access and simulation module is used to quickly locate and inject fault parameters in the digital twin based on the fault point described in the work order, simulate and calculate the potential impact range and degree of the fault on the room temperature of upstream and downstream users, and output the list of affected users as a recommended related work order.

6. The automatic form filling system for heating customer service work orders based on a large model as described in claim 1, characterized in that, The proactive intervention function of the work order decision and proactive intervention engine includes: Horizontal proactive early warning: When a work order for a blockage in a unit's riser is detected, a preventative work order is automatically sent to all residents supplied by that riser to inform them that the room temperature may be too low. Vertical proactive regulation: When a sudden surge of non-heating work orders in a certain area occurs within a short period of time, and the digital twin simulation confirms that the output of the heating station is insufficient, an adjustment suggestion work order is automatically generated and sent to the heat energy production scheduling system.

7. A method for automatically filling out heating customer service work orders based on a large model, applied to the system described in any one of claims 1-6, characterized in that, Includes the following steps: Step 1: Access user multimodal service requests through all channels; Step 2: Collaborative intelligent agents work in parallel: Intent parsing agents extract requests and emotions; The professional knowledge retrieval agent obtains candidate solutions from a dynamic knowledge network; the context-aware agent obtains real-time data from user meters, weather, and the status of surrounding work orders. Step 3: The dynamic knowledge network performs multi-hop reasoning based on the output of Step 2 to generate a preliminary diagnostic reasoning chain; Step 4: The work order decision engine encapsulates user information, diagnostic chain, and contextual data into an initial work order, and calls the digital twin to simulate and extrapolate the impact of the fault. Step 5: Based on the simulation results, the decision engine performs at least one of the following operations: a. Improve and confirm the dispatch of repair work orders to the original users; b. Generate and dispatch proactive warning work orders for potentially affected users identified in the simulation; c. Generate and send equipment parameter adjustment suggestion work orders to the production scheduling system; Step Six: The entire process data of work order processing is captured by the self-evolution closed-loop management module, which is used to optimize the intelligent agent, update the knowledge network, and calibrate the digital twin model.

8. The method for automatically filling out heating customer service work orders based on a large model according to claim 7, characterized in that, The simulation in step four involves adjusting the valve opening or pipe resistance coefficient of the corresponding household in the digital twin to a fault state, running the hydraulic and thermal model, and calculating the predicted room temperature drop and the time to reach a new equilibrium for other households in the building, especially those on the upper and lower floors and in the same location.

9. The method for automatically filling out heating customer service work orders based on a large model according to claim 7, characterized in that, The calibration of the digital twin model in step six involves comparing the fault points finally resolved by the maintenance work order and the meter data retested after maintenance with the fault points and room temperature changes predicted by the simulation, calculating the simulation error, and using the error data to automatically fine-tune the relevant model parameters of the digital twin.

10. The method for automatically filling out heating customer service work orders based on a large model according to claim 7, characterized in that, The final generated work order includes an interpretable view that visually displays to customer service or maintenance personnel: the user's original description, the analysis conclusions of each agent, the reasoning path of the knowledge graph, and the simulation impact range of the digital twin, thereby improving the credibility and efficiency of work order processing.

Images