Intelligent heat supply decision-making system based on artificial intelligence large model

Abstract

The invention discloses a heat supply intelligent decision-making system based on an artificial intelligence large model, and the system comprises a data layer which is used for storing multi-type data of a heat supply system source network load; the knowledge base establishment and large model training layer is used for establishing a heat supply intelligent decision diversified task knowledge base and establishing each task large model; the man-machine enhancement optimization large model layer is used for providing improvement suggestions for a large model, optimizing generated answers, evaluating the answers generated by the large model and performing large model parameter optimization training; the agent decision-making layer is used for integrating each task large model into each agent to carry out environment perception, collaborative interaction and autonomous adaptive decision-making; the analogue simulation layer is used for providing a simulation operation environment, a user interaction interface and scene dynamic parameter simulation; and the agent evaluation optimization layer is used for constructing performance indexes for evaluating the agents, and evaluating, optimizing and adjusting behaviors and decisions of the agents.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart heating systems, and specifically relates to a heating intelligent decision-making system based on an artificial intelligence large model. Background Art

[0002] Intelligent decision-making of heating systems refers to the process of using advanced information technology and intelligent algorithms to make autonomous decisions on operation monitoring, fault diagnosis and optimal scheduling of heating systems, so as to achieve safe, efficient, energy-saving and reliable operation of heating systems.

[0003] The large language model is a large-scale natural language processing model based on deep learning technology. It can understand and generate natural language text and has strong language expression and reasoning capabilities. The large language model is applied to the intelligent agent, and the large language model serves as the brain of the intelligent agent, providing language understanding, reasoning capabilities and knowledge base. Through continuous training and optimization, the large language model can generate more accurate and fluent texts, providing more abundant information and support for the intelligent agent. The intelligent agent uses prompt words to command the large model to complete specific tasks, and continuously optimizes its behavior through learning to adapt to different environments and tasks, with higher autonomy and intelligence. However, due to the complexity of the heating system, the current intelligent decision-making of the heating system has great difficulty in data integration, limited data analysis capabilities, poor model adaptability and lack of autonomous learning ability. Therefore, how to apply large models and intelligent agent technology to intelligent decision-making in heating, improve the level of intelligent decision-making, and realize environmental perception, collaborative interaction and autonomous adaptive decision-making of the heating system is an urgent problem to be solved.

[0004] Based on the above technical problems, it is necessary to design a new intelligent heating decision-making system based on a large artificial intelligence model. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a heating intelligent decision-making system based on an artificial intelligence big model. By utilizing the knowledge base, big model and multi-agent technology, a multi-level heating intelligent decision-making system is constructed to improve the intelligence level of the heating system decision-making, so that the intelligent agent integrating the big model and the knowledge base has the ability to perceive the heating system environment, collaboratively interact and make autonomous adaptive decisions.

[0006] In order to solve the above technical problems, the technical solution of the present invention is: The present invention provides a heating intelligent decision-making system based on an artificial intelligence large model, which comprises: Data layer: used to collect, store and manage various types of data on the source, grid, load and storage of the heating system, and as a large-scale data set for knowledge base establishment and large model training layer; Knowledge base establishment and large model training layer: used to establish a knowledge base for diversified tasks of intelligent heating decision-making after preprocessing, labeling, knowledge extraction and representation of large-scale data sets of the heating system; also used to train large models based on the knowledge base for diversified tasks of intelligent heating decision-making and the prompt learning strategy for intelligent heating decision-making, capture the knowledge and behavioral characteristics of each task of intelligent heating decision-making through the large model, and establish the intrinsic connection between each prompt word and high-quality answer, and fine-tune the large model according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, so as to adapt to specific business scenarios, decision-making tasks and various intelligent agents; Human-machine enhanced optimization big model layer: used for the heating dispatching and control personnel to make improvement suggestions to the big model during the dialogue process of each task of heating intelligent decision-making, and the big model optimizes the generated answers based on the suggestions; it is also used for the heating dispatching and control personnel to evaluate the multiple answers generated by the big model based on the prompt words, and train and optimize the big model parameters based on the evaluation results; Intelligent agent decision layer: used to integrate the large models of each task of intelligent heating decision-making after training and optimization into each intelligent agent, use the large model to drive the intelligent agent, and guide each intelligent agent to perform environmental perception, collaborative interaction and autonomous adaptive decision-making in the simulation environment; Simulation layer: used to provide the simulation operation environment, user interaction interface and scenario dynamic parameter simulation for heating intelligent decision-making interaction. After the heating dispatching control personnel raise questions, each intelligent agent generates decision simulation data and supports displaying answers to users in the form of text, pictures and videos. Agent evaluation and optimization layer: used to construct performance indicators for evaluating each agent, and evaluate and optimize the behavior and decision-making effects of each agent based on the performance indicators and decision simulation results.

[0007] Furthermore, the heating system source, grid, load and storage multi-type data includes: historical operation data of the heating system source, grid, load and storage, heating operation and maintenance and scheduling management manual, dynamic change parameters, and physical property data of each device in the system; Among them, the historical operation data include: the operating parameters of the source-side heat source equipment, the heat metering data of each pipeline in the heat network, the indoor temperature and heat consumption data, the operating status and storage and release heat of the heat storage equipment, the historical heat load data, and the on-demand scheduling decision strategy, the historical energy consumption analysis data, and the abnormal diagnosis data; the heating operation and maintenance and scheduling management manual includes the heating inspection content, equipment maintenance strategy, fault handling process, scheduling process, and emergency scheduling response processing rules; the dynamically changing parameters include outdoor environmental parameters and sudden abnormal fault parameters; the physical property data of each device in the system include the fixed parameters, operating characteristics, and service life of each source network, load storage equipment, and pipeline network.

[0008] Furthermore, after data preprocessing, data annotation, knowledge extraction and representation of the large-scale data set of the heating system, a knowledge base of diversified tasks for intelligent decision-making in heating is established, including: Clarify the diverse tasks of intelligent heating decision-making, including operation perception, parameter prediction, fault diagnosis, scheduling planning and energy consumption evaluation, and determine the knowledge scope and knowledge types covered by the knowledge base based on the diverse tasks; the knowledge types include factual knowledge, rule knowledge and process knowledge; According to the diverse tasks, knowledge scope and knowledge type of intelligent heating decision-making, data preprocessing, labeling and classification of large-scale data sets of heating systems are carried out; Knowledge extraction: Directly formulate rules for knowledge extraction based on clear rules and experience; use machine learning algorithms to discover potential knowledge from data; use natural language processing technology to extract key information and knowledge from unstructured text data; Knowledge representation: The extracted knowledge is represented based on production rule representation, framework representation and semantic network representation; The extracted and represented knowledge is stored in the corresponding database table or graph structure according to the format and requirements of the knowledge base management system, and indexes and associations between knowledge are established. A knowledge retrieval interface is provided to support keyword retrieval, semantic retrieval and association retrieval methods, and finally a knowledge base for diversified tasks of intelligent heating decision-making is established.

[0009] Furthermore, the large model training is performed based on the diversified task knowledge base of heating intelligent decision-making and the prompt learning strategy of heating intelligent decision-making, and the knowledge and behavior characteristics of each task of heating intelligent decision-making are captured through the large model, and the intrinsic connection between each prompt word and high-quality answer is established, including: In the large model training stage, the diversified knowledge base of heating intelligent decision-making is input into the base large model as part of the training corpus, and the heating professional terms, semantic relations, and logical relations between various factors in the heating system in the knowledge base are learned. The operation mode, behavioral characteristics and intelligent decision-making process of the heating system under different working conditions are analyzed. According to the characteristics of the heating intelligent decision-making task, the thinking process and rules are preset through the thinking chain method, and the decision-making task problem is decomposed into multiple sub-problems. Logically clear steps are provided in the form of prompts, and different types of prompt templates are designed to guide the large model to analyze and process the input information according to the steps and processes; The designed prompt template is combined with the content of the knowledge base and input into the large model, so that the model can learn the ability to extract relevant knowledge from the knowledge base and generate answers based on the prompt information, as well as collect user feedback on the answers and establish the intrinsic connection between the prompt words and the answers.

[0010] Furthermore, the large model is fine-tuned according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, including: For the task of sensing the operation of the heating system, the specific scenario requirement is to sense the operating status of each device in the heating system, and the task goal is to fully and accurately obtain the operating data of each device. The constraint condition is that the data collection is compatible with the data interfaces of devices of different types and manufacturers. For the task of predicting heating system parameters, the specific scenario requirement is to predict heat load and heating network parameters. The task goal is to predict and accurately predict the changes in heating system operating parameters in advance. The constraints are to consider the influencing factors of weather changes, user heating behavior, and historical data rules. For the heating system fault diagnosis task, the specific scenario requirement is to diagnose the fault type of the heating system and determine the cause of the fault. The task goal is to accurately identify the fault type and locate the fault cause. The constraint condition is the complexity of the fault scenario. For the scheduling and planning task of the heating system, the specific scenario requirement is to generate scheduling plans of different time scales for each source, grid, load and storage device according to the supply and demand relationship of the heat load. The task goal is to meet the heating demand, minimize the economic cost and reduce energy consumption. The constraints are the operating parameter constraints of each device and the thermal power balance constraints. For the task of evaluating the energy consumption of the heating system, the specific scenario requirements are to evaluate the energy consumption of the heating system, analyze the factors affecting energy consumption, and put forward energy-saving suggestions. The task goal is to reduce the energy consumption of the heating system, and the constraint condition is to consider the energy consumption differences under different operating conditions; The big model is fine-tuned according to the specific scenario requirements, objective functions, and constraints of each task, and a big model for each task is established, including an operation perception big model, a parameter prediction big model, a fault diagnosis big model, a scheduling planning big model, and an energy consumption assessment big model.

[0011] Furthermore, during the dialogue process of each task of the heating intelligent decision-making, the heating dispatch control personnel put forward improvement suggestions to the big model, and the big model generates answers based on the optimization suggestions, including: During the dialogue process of each task of heating intelligent decision-making, the heating dispatch control personnel can input improvement suggestions through text boxes or voice, and convert the input improvement suggestions into instructions that can be understood by the large model by using the natural language understanding function; The big model adjusts its internal knowledge base and reasoning logic based on the improvement suggestions to optimize the generated answers.

[0012] Furthermore, the heating dispatch control personnel evaluates the multiple answers generated by the large model according to the prompt words, and trains and optimizes the parameters of the large model according to the evaluation results, including: The heating dispatching and control personnel use the trained large model to generate multiple different answers based on the prompt words, and construct evaluation indicators according to the language comprehension index, semantic rationality index, context coherence index, heating knowledge accuracy index, and heating reasoning ability index to evaluate and assign values ​​to multiple different answers; Different questions, answers, and evaluation results are used as input, and reinforcement learning algorithms are used to update and optimize large model parameters.

[0013] Furthermore, the training optimized large model of each task of intelligent decision-making for heating is integrated into each intelligent agent, and the large model is used to drive the intelligent agent to guide each intelligent agent to perform environmental perception, collaborative interaction and autonomous adaptive decision-making in the simulation environment, including: According to the diversified tasks of intelligent heating decision-making, corresponding task agents are set up, including operation perception agent, parameter prediction agent, fault diagnosis agent, scheduling planning agent and energy consumption evaluation agent; Set up bottom-level equipment agents, including heat source agents, heat network agents, heat storage agents and heat user agents, respectively collect data of heat source equipment, heat network, heat storage devices and heat users and transmit them to each task agent, and issue and execute decision instructions transmitted by each task agent; Optimize the input and output interfaces of each task model to connect it with each task agent of heating intelligent decision-making, allocate a large model calling channel for each task agent, and integrate each task model into the corresponding task agent; Each task agent uses its perception unit to perceive the environment to obtain collected data, extract key information from the data, and transmit it to the big model of the decision-making unit. At the same time, a collaborative interaction mechanism is established between the task agent and the underlying equipment agent to conduct collaborative data interaction and optimize task execution and task resource allocation. Finally, through the big model of the decision-making unit, in-depth analysis and reasoning learning are performed, and each task agent makes autonomous decisions to generate decision instructions to adapt to the dynamically changing operating conditions of the heating system.

[0014] Further, the simulation operation environment for providing heating intelligent decision interaction includes: managing simulation time to analyze the operation status of the heating system at different times, updating the status of the simulation environment, and providing a flexible time control interface to allow real-time operation of simulation time during system operation; The user interaction interface for providing heating intelligent decision-making interaction includes: providing a graphical parameter configuration interface, setting simulation parameters and supporting the loading of parameter configuration files through sliders, drop-down menus, and text box interactive elements, and visually displaying and analyzing simulation results in the form of text, pictures, and videos; The dynamic parameter simulation of the scene when providing heating intelligent decision-making interaction includes: using a large model to build a simulation model of the actual heating system and a dynamic environment to perform system simulation.

[0015] Furthermore, the construction evaluates the performance indicators of each intelligent agent, and evaluates and optimizes the behavior and decision-making effect of each intelligent agent according to the performance indicators and decision simulation results, including: Construct performance indicators to evaluate the intelligent decision-making tasks of intelligent heating, including task completion, task completion efficiency, decision-making adaptability, and collaboration between intelligent agents; According to the behavior trajectory, decision-making process and output decision simulation results of each task agent, the performance index value of each task agent is calculated and compared with the preset standards to evaluate the performance of each task agent; Based on the evaluation results, the behavior and decision-making strategies of each task agent are optimized and adjusted, including: improving the algorithm and model structure within the agent, enhancing the agent's environmental perception and adaptability, and optimizing the collaboration mechanism and communication protocol between agents.

[0016] The beneficial effects of the present invention are: (1) The present invention collects multiple types of data on heating system sources, grids, loads and storage, and preprocesses, annotates, extracts and represents large-scale data sets, thereby establishing a knowledge base for diversified tasks of intelligent heating decision-making, integrating professional knowledge and experience in the field of heating systems, and providing a knowledge source for the training of large models, which helps the large models better understand and process decision-making tasks related to heating; (2) The present invention trains a large model based on the knowledge base of diversified tasks of intelligent heating decision-making and the prompt learning strategy, so that the large model can capture the knowledge and behavioral characteristics of each task of intelligent heating decision-making, and at the same time establish the intrinsic connection between prompt words and high-quality answers. This enables the large model to generate more accurate and practical answers and decision-making suggestions for different tasks of the heating system; (3) The present invention fine-tunes the large model according to the specific scenario requirements, task objectives and constraints of each heating task, and establishes a large model for each task that is adapted to specific business scenarios, decision-making tasks and intelligent agents. This fine-tuning mechanism improves the pertinence and adaptability of the large model, and can better meet the complex and changeable actual needs of the heating system; (4) The present invention establishes a human-machine enhanced optimization large model layer, so that the large model can continuously learn and improve, gradually improve the accuracy and reliability of decision-making, and realize dynamic optimization of the model; (5) The present invention integrates the large models of each task after training and optimization into the intelligent agent, and uses the large model to drive the intelligent agent, so that the intelligent agent can perform efficient environmental perception, collaborative interaction and autonomous adaptive decision-making in the simulation environment. The intelligent agent can make reasonable decisions quickly according to the actual operation of the heating system, thereby improving the intelligence level and response speed of the heating system; (6) The simulation layer of the present invention provides a simulation operating environment, user interaction interface and scenario dynamic parameter simulation for heating intelligent decision-making interaction, and supports displaying answers to users in multiple ways. This enables heating dispatch control personnel to more intuitively understand the decision-making process and results of the system, facilitates operation and adjustment, and improves user experience and decision-making efficiency; (7) The agent evaluation optimization layer of the present invention constructs performance indicators for evaluating each agent, and can conduct a comprehensive and objective evaluation of the agent's behavior and decision-making effect based on the performance indicators and decision-making simulation results. This helps to discover the problems and deficiencies of the agent and provide a basis for optimization and adjustment.

[0017] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 This is a block diagram of a heating intelligent decision-making system based on an artificial intelligence large model of the present invention; Figure 2 This is a schematic diagram of the heating intelligent decision-making principle based on an artificial intelligence large model of the present invention; Figure 3 The present invention is a schematic block diagram of the heating fault diagnosis principle based on the artificial intelligence large model. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment provides a heating intelligent decision-making system based on an artificial intelligence large model, which includes: Data layer: used to collect, store and manage various types of data on the source, grid, load and storage of the heating system, and as a large-scale data set for knowledge base establishment and large model training layer; Knowledge base establishment and large model training layer: used to establish a knowledge base for diversified tasks of intelligent heating decision-making after preprocessing, labeling, knowledge extraction and representation of large-scale data sets of the heating system; also used to train large models based on the knowledge base for diversified tasks of intelligent heating decision-making and the prompt learning strategy for intelligent heating decision-making, capture the knowledge and behavioral characteristics of each task of intelligent heating decision-making through the large model, and establish the intrinsic connection between each prompt word and high-quality answer, and fine-tune the large model according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, so as to adapt to specific business scenarios, decision-making tasks and various intelligent agents; Human-machine enhanced optimization big model layer: used for the heating dispatching and control personnel to make improvement suggestions to the big model during the dialogue process of each task of heating intelligent decision-making, and the big model optimizes the generated answers based on the suggestions; it is also used for the heating dispatching and control personnel to evaluate the multiple answers generated by the big model based on the prompt words, and train and optimize the big model parameters based on the evaluation results; Intelligent agent decision layer: used to integrate the large models of each task of intelligent heating decision-making after training and optimization into each intelligent agent, use the large model to drive the intelligent agent, and guide each intelligent agent to perform environmental perception, collaborative interaction and autonomous adaptive decision-making in the simulation environment; Simulation layer: used to provide the simulation operation environment, user interaction interface and scenario dynamic parameter simulation for heating intelligent decision-making interaction. After the heating dispatching control personnel raise questions, each intelligent agent generates decision simulation data and supports displaying answers to users in the form of text, pictures and videos. Agent evaluation and optimization layer: used to construct performance indicators for evaluating each agent, and evaluate and optimize the behavior and decision-making effects of each agent based on the performance indicators and decision simulation results.

[0023] like Figure 2The overall technical principle of intelligent decision-making for heating is shown in the figure. The heating system consists of a heat source plant, a primary network, a thermal power station, a heat storage device and a heat user. These components generate large-scale data sources and provide raw data for subsequent intelligent decision-making. The base model is a basic model that provides general artificial intelligence capabilities and learning frameworks, and provides underlying technical support for the large model of intelligent decision-making for heating. The large model of intelligent decision-making for heating: Based on the base model, combined with the large-scale data source generated by the heating system and the knowledge sources of each task in the knowledge base of the diversified task of intelligent decision-making for the heating system (factual, regular and procedural knowledge of each task of operation perception, parameter prediction, fault diagnosis, scheduling planning, and energy consumption evaluation), it is trained and optimized to have the ability of intelligent decision-making in the field of heating. The large model of intelligent decision-making for heating is integrated into the intelligent agent of heating, and the intelligent agent is driven by the large model, so that the intelligent agent has the ability of environmental perception, collaborative interaction and autonomous adaptive decision-making, and realizes the intelligent control and management of the heating system to improve the operating efficiency, reliability and energy saving effect of the heating system.

[0024] like Figure 3 The heating fault diagnosis principle based on the artificial intelligence big model is shown in the figure. Taking the heating fault diagnosis task as an example, the technical principles of big model training and big model application mainly include: 1) Base model and data input: The base model is the cornerstone of the entire technical system, and it provides general artificial intelligence capabilities. Various data sources such as fault diagnosis teaching materials data, regulations data, scientific research data, and heating system historical data are input. These data contain various information during the operation of the heating system, such as equipment status, fault records, etc. The heating natural language model is built based on the base model. Through learning the input data, it has the ability of semantic understanding, knowledge reasoning, content generation and code generation, and can process and analyze natural language information related to the heating system.

[0025] 2) Fault diagnosis sample data and model fine-tuning: Source-grid-load-side fault diagnosis reports, fault diagnosis regulations and operating manuals constitute the fault diagnosis sample data. Based on the heating natural language big model and fault diagnosis sample data, the heating system fault diagnosis big model is trained by using model fine-tuning, prompt word engineering, generative retrieval and other methods. This process enables the model to be optimized for the heating system fault diagnosis task and better understand and handle specific problems in this field.

[0026] 3) Role of knowledge base: The knowledge base includes equipment operation instructions knowledge base, equipment failure problem knowledge base, equipment failure handling knowledge base, etc. These knowledge bases provide rich professional knowledge for the large model of heating system fault diagnosis. When the model performs fault diagnosis, it can obtain relevant information from the knowledge base to assist in making accurate judgments.

[0027] 4) Heating fault diagnosis agent and application interface: The heating fault diagnosis agent is built based on the large model of heating system fault diagnosis, and has fault diagnosis decision-making capabilities, perception and insight capabilities, learning and evolution capabilities, and control and execution capabilities. Interface service API, service encapsulation SDK, service plug-in and other components provide the agent with a way to interact with external systems, so that it can play a role in the actual heating system and realize the fault diagnosis function.

[0028] The heating fault diagnosis task has the following advantages through large model training and large model application: 1) Accurate fault diagnosis: Through the training of a large amount of historical data and professional knowledge, the large model of heating system fault diagnosis can accurately identify various faults in the heating system. Combined with the information in the knowledge base, it improves the accuracy and reliability of fault diagnosis and reduces misjudgment and missed judgment. 2) Intelligent decision support: The heating fault diagnosis agent has decision-making capabilities and can provide reasonable processing suggestions and decision support based on the diagnosis results, helping operation and maintenance personnel to quickly and effectively solve fault problems and improve the stability and reliability of the heating system; 3) Self-evolution and learning: The intelligent agent has the ability to learn and evolve. With the continuous input of new data and the operation of the system, it can continuously optimize and improve the fault diagnosis model, adapt to the changes and development of the heating system, and continuously improve the ability and level of fault diagnosis; 4) Convenient system integration: Rich interface components enable the heating fault diagnosis agent to be easily integrated with other heating systems or management platforms to achieve data sharing and collaborative work, thereby improving the intelligent management level and operating efficiency of the entire heating system; 5) Knowledge accumulation and inheritance: The existence of the knowledge base not only provides support for model training and fault diagnosis, but also plays a role in knowledge accumulation and inheritance, making it easier for new operation and maintenance personnel to learn and understand the relevant knowledge and experience of heating system fault diagnosis.

[0029] In this embodiment, the heating system source grid load storage multi-type data includes: heating system source grid load storage historical operation data, heating operation and maintenance and scheduling management manual, dynamic change parameters, and physical property data of each device in the system; Among them, the historical operation data include: the operating parameters of the source-side heat source equipment, the heat metering data of each pipeline in the heat network, the indoor temperature and heat consumption data, the operating status and storage and release heat of the heat storage equipment, the historical heat load data, and the on-demand scheduling decision strategy, the historical energy consumption analysis data, and the abnormal diagnosis data; the heating operation and maintenance and scheduling management manual includes the heating inspection content, equipment maintenance strategy, fault handling process, scheduling process, and emergency scheduling response processing rules; the dynamically changing parameters include outdoor environmental parameters and sudden abnormal fault parameters; the physical property data of each device in the system include the fixed parameters, operating characteristics, and service life of each source network, load storage equipment, and pipeline network.

[0030] In this embodiment, after data preprocessing, data annotation, knowledge extraction and representation of the large-scale data set of the heating system, a knowledge base of diversified tasks for intelligent decision-making in heating is established, including: Clarify the diverse tasks of intelligent heating decision-making, including operation perception, parameter prediction, fault diagnosis, scheduling planning and energy consumption evaluation, and determine the knowledge scope and knowledge types covered by the knowledge base based on the diverse tasks; the knowledge types include factual knowledge, rule knowledge and process knowledge; According to the diverse tasks, knowledge scope and knowledge type of intelligent heating decision-making, data preprocessing, labeling and classification of large-scale data sets of heating systems are carried out; Knowledge extraction: Directly formulate rules for knowledge extraction based on clear rules and experience; use machine learning algorithms to discover potential knowledge from data; use natural language processing technology to extract key information and knowledge from unstructured text data; Knowledge representation: The extracted knowledge is represented based on production rule representation, framework representation and semantic network representation; The extracted and represented knowledge is stored in the corresponding database table or graph structure according to the format and requirements of the knowledge base management system, and indexes and associations between knowledge are established. A knowledge retrieval interface is provided to support keyword retrieval, semantic retrieval and association retrieval methods, and finally a knowledge base for diversified tasks of intelligent heating decision-making is established.

[0031] In actual applications, the specific process of establishing a task knowledge base includes: 1) Clarify the diverse tasks, knowledge scope and types of intelligent heating decision-making For example, for parameter prediction tasks: predict heating load and heating network parameters (such as temperature, flow, and pressure change trends); knowledge scope: historical heating data, meteorological data, building characteristics data, user heating behavior data, etc.; factual knowledge: the maximum heating load in a certain heating network area reaches 100MW; rule-based knowledge: when the outdoor temperature drops by 1°C, the heating load in the area is expected to increase by 5MW; process knowledge: parameter prediction process, including data preprocessing, model selection, training and evaluation steps.

[0032] For fault diagnosis tasks: quickly and accurately identify heating system equipment failures (such as pipeline leakage, equipment damage, etc.) and the causes of failures; knowledge scope: equipment failure phenomena, causes of failures, maintenance methods, etc.; factual knowledge: a certain heating network pipeline experienced a sudden drop in pressure and an abnormal increase in flow; regular knowledge: if the heating network pressure drops sharply and the flow increases abnormally, it may be a pipeline leak; procedural knowledge: fault diagnosis process, a series of steps from fault detection to location, cause analysis and then to maintenance.

[0033] For scheduling planning tasks: optimize the scheduling strategies of heat source, heat network and user side according to information such as heating demand, energy supply and equipment status; knowledge scope: heating capacity of heat source, hydraulic characteristics of heating network, user heating demand curve, energy price, etc.; factual knowledge: the maximum heating capacity of a thermal power plant is 200MW; rule-based knowledge: when energy prices rise by 10%, priority will be given to scheduling heat source equipment with low energy consumption; process knowledge: heating scheduling process, the process of formulating and executing scheduling plans based on load forecast and equipment status assessment.

[0034] 2) Data preprocessing, labeling and classification Data preprocessing: Clean the collected raw data to remove noise and outliers. For example, for temperature data collected by the sensor, if there is a value that is obviously beyond the reasonable range (such as the temperature is -100℃), it is judged as an outlier and corrected or deleted; Data labeling: Label the data according to different tasks. For example, in the fault diagnosis task, the data of sudden drop in heat network pressure and abnormal increase in flow rate are labeled as "suspected pipeline leakage data"; in the parameter prediction task, the historical heating load data are labeled as "actual heating load", and the corresponding meteorological conditions and other related information are also labeled.

[0035] Data classification: Data is classified and stored according to task type. For example, operation perception data is stored in the "operation perception data" table, parameter prediction related data is stored in the "parameter prediction data" table, and so on.

[0036] 3) Knowledge Extraction Clarify rules and extract experience: For example, in fault diagnosis, based on long-term accumulated experience, the rule is clarified as "when the boiler water level continues to be lower than the minimum water level line for more than 5 minutes, it is judged as a boiler water shortage fault", and this rule is directly applied to knowledge extraction.

[0037] Knowledge extraction from machine learning algorithms: For example, in parameter prediction tasks, time series analysis algorithms (such as the ARIMA model) are used to discover potential change patterns from historical heating load data and extract knowledge such as "heating load exhibits different fluctuation patterns on weekdays and weekends".

[0038] Knowledge extraction using natural language processing: When processing unstructured text data such as equipment maintenance records, named entity recognition and relationship extraction methods in natural language processing technology are used. For example, key information and knowledge such as "Boiler No. 1", "Burner Parts", and "Incomplete Combustion" can be extracted from "Boiler No. 1 was repaired today and the burner parts were replaced due to incomplete combustion".

[0039] 4) Knowledge Representation Production rule representation: For example, in scheduling planning, the rule "If the outdoor temperature is lower than -10℃ and the heating load exceeds 80% of the rated load, then start the backup heat source" can be expressed as: IF (outdoor temperature <-10℃) AND (heating load > 0.8*rated load) THEN start the backup heat source.

[0040] Framework representation: For the heat source equipment boiler, the following framework can be constructed: Frame Name: Boiler; Slot name 1: Model, value: [specific model]; Slot name 2: Rated power, value: [X] MW; Slot name 3: Operation status, value: [Normal / Fault]; Slot name 4: Fault description, value: [If fault occurs, fill in the specific fault information].

[0041] Semantic network representation: Build a semantic network about heat network failure. The "heat network pipeline" node is connected with nodes such as "pressure drop", "abnormal flow increase", and "pipeline leakage" through relationship edges to clearly show the semantic relationship between them.

[0042] 5) Knowledge storage and retrieval Storage: The extracted and represented knowledge is stored in the corresponding database table or graph structure. For example, the factual knowledge of operation perception is stored in the "operation perception information table" of the relational database; the knowledge of fault diagnosis is stored in the graph database in a graph structure to facilitate the display of the association relationship between knowledge.

[0043] Establish indexes and associations: Establish indexes for knowledge, such as using equipment number and monitoring time as indexes in operation perception data to facilitate quick queries. At the same time, establish associations between knowledge, such as establishing associations between fault phenomena, fault causes, and maintenance methods in fault diagnosis knowledge.

[0044] Knowledge retrieval: Provides a knowledge retrieval interface and supports multiple retrieval methods. For example, if a user enters the keyword "pipeline leakage", the user can quickly obtain the fault phenomenon, cause and maintenance knowledge related to pipeline leakage through keyword retrieval; through semantic retrieval, if the user enters "abnormal situation in the heating network", the system can understand the semantics and return relevant fault diagnosis knowledge; and the associated retrieval can retrieve the related dispatching strategy, energy consumption evaluation and other knowledge according to the operating status of a certain device.

[0045] In this embodiment, the large model training is performed based on the diversified task knowledge base of heating intelligent decision-making and the prompt learning strategy of heating intelligent decision-making. The large model captures the knowledge and behavior characteristics of each task of heating intelligent decision-making and establishes the intrinsic connection between each prompt word and high-quality answer, including: In the large model training stage, the diversified knowledge base of heating intelligent decision-making is input into the base large model as part of the training corpus, and the heating professional terms, semantic relations, and logical relations between various factors in the heating system in the knowledge base are learned. The operation mode, behavioral characteristics and intelligent decision-making process of the heating system under different working conditions are analyzed. According to the characteristics of the heating intelligent decision-making task, the thinking process and rules are preset through the thinking chain method, and the decision-making task problem is decomposed into multiple sub-problems. Logically clear steps are provided in the form of prompts, and different types of prompt templates are designed to guide the large model to analyze and process the input information according to the steps and processes; The designed prompt template is combined with the content of the knowledge base and input into the large model, so that the model can learn the ability to extract relevant knowledge from the knowledge base and generate answers based on the prompt information, as well as collect user feedback on the answers and establish the intrinsic connection between the prompt words and the answers.

[0046] In actual applications, we learn professional terms and semantic relationships for heating: for example, in the diversified knowledge base of intelligent heating decision-making, there are professional terms such as "primary network", "secondary network", and "heat exchange station". When training, the large base model learns the knowledge base corpus to understand that the "primary network" is the high-temperature hot water transmission network from the heat source to the heat exchange station, and the "secondary network" is the network from the heat exchange station to the user end. The two exchange heat and adjust parameters through the "heat exchange station", so as to master the semantic relationship between these terms.

[0047] For the term "heat medium", the model learns that it is a medium for transferring heat. Common heat media in heating systems are hot water or steam. It also understands the impact of changes in parameters such as temperature and pressure of the heat medium on the heating effect, and establishes a semantic connection between the term and the actual physical quantity of heating.

[0048] Learning the logical relationship of heating system factors: For example, when there is a logical relationship description such as "the outdoor temperature decreases, the heating load increases" in the knowledge base, after the large model learns this knowledge, when it encounters input information about the change of outdoor temperature, it can infer the possible change trend of the heating load based on the learned logical relationship, thereby providing a basis for intelligent heating decision-making.

[0049] For example, by learning the logical chain that "the output of the heat source equipment determines the heating amount of the heating network, and the flow and temperature distribution of the heating network affect the heating effect at the user end", the model can understand the interaction between the key factors in the heating system and conduct a comprehensive analysis when dealing with related decision-making problems.

[0050] Learning about operation modes and behavior characteristics: The operation modes of the heating system are different under different working conditions. For example, during the severe cold period in winter, the heat source equipment needs to maintain a high output, and the water supply temperature and flow of the heating network must be maintained at a high level to meet the heating needs of users. The large model learns the relevant cases and data in the knowledge base to master the operation mode characteristics of this severe cold period, including the operating parameter range of the equipment, adjustment strategy, etc.

[0051] For example, for the behavioral characteristics of the heating system under equipment failure conditions, such as pipeline leakage, the heat network pressure will drop and the flow rate will change abnormally. After learning these characteristics, the large model can determine the possible fault conditions when encountering similar parameter change inputs, providing support for fault diagnosis and emergency decision-making.

[0052] Learning about intelligent decision-making processes: Taking heat dispatching decisions as an example, the decision-making process includes formulating a heat supply plan for heat sources based on load forecast results, adjusting the flow distribution of the heat network, and making real-time adjustments based on user feedback. After the big model learns this decision-making process, when faced with dispatching decision-making tasks, it can analyze and process according to this process, starting from the load forecast data, and gradually generate a reasonable dispatching strategy.

[0053] It should be noted that the thinking chain presets and problem decomposition: For example, when dealing with the decision-making task of "how to ensure the quality of heating in extremely low temperature weather", the thinking process is preset through the thinking chain method. First, the problem is decomposed into several sub-questions: How will the heating load change in extremely low temperature weather (sub-question 1)? Does the heating capacity of the current heat source meet the demand (sub-question 2)? Does the transmission capacity of the heat network need to be adjusted (sub-question 3)? Do the heating facilities at the user end require special maintenance (sub-question 4)? For each sub-question, design corresponding prompt rules. For example, for sub-question 1, the prompt rule can be "Refer to the heating load data under historical extreme low temperature weather, combined with the current building conditions and population distribution in the area, and analyze the changing trend of the heating load."

[0054] Design and application of prompt templates: For example, design a prompt template for fault diagnosis: "When [specific fault phenomenon, such as a sudden drop in heat network pressure] occurs, first check [related equipment or systems, such as heat network pipeline connections], then check [other possible influencing factors, such as the status of the pressure sensor], and finally, based on the inspection results, determine that the cause of the fault may be [list possible causes of the fault, such as pipeline leakage, sensor failure]".

[0055] Such prompt templates are combined with the content about heating network faults in the knowledge base and input into the large model. When the model encounters the input information of a sudden drop in heating network pressure, it will follow the steps of the prompt template, extract relevant knowledge about heating network pipe connections, pressure sensors, etc. from the knowledge base, analyze and judge, and generate a fault diagnosis answer.

[0056] In this embodiment, the large model is fine-tuned according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, including: For the task of sensing the operation of the heating system, the specific scenario requirement is to sense the operating status of each device in the heating system, and the task goal is to fully and accurately obtain the operating data of each device. The constraint condition is that the data collection is compatible with the data interfaces of devices of different types and manufacturers. For the task of predicting heating system parameters, the specific scenario requirement is to predict heat load and heating network parameters. The task goal is to predict and accurately predict the changes in heating system operating parameters in advance. The constraints are to consider the influencing factors of weather changes, user heating behavior, and historical data rules. For the heating system fault diagnosis task, the specific scenario requirement is to diagnose the fault type of the heating system and determine the cause of the fault. The task goal is to accurately identify the fault type and locate the fault cause. The constraint condition is the complexity of the fault scenario. For the scheduling and planning task of the heating system, the specific scenario requirement is to generate scheduling plans of different time scales for each source, grid, load and storage device according to the supply and demand relationship of the heat load. The task goal is to meet the heating demand, minimize the economic cost and reduce energy consumption. The constraints are the operating parameter constraints of each device and the thermal power balance constraints. For the task of evaluating the energy consumption of the heating system, the specific scenario requirements are to evaluate the energy consumption of the heating system, analyze the factors affecting energy consumption, and put forward energy-saving suggestions. The task goal is to reduce the energy consumption of the heating system, and the constraint condition is to consider the energy consumption differences under different operating conditions; The big model is fine-tuned according to the specific scenario requirements, objective functions, and constraints of each task, and a big model for each task is established, including an operation perception big model, a parameter prediction big model, a fault diagnosis big model, a scheduling planning big model, and an energy consumption assessment big model.

[0057] In this embodiment, during the dialogue process of each task of the heating intelligent decision-making, the heating scheduling control personnel put forward improvement suggestions to the big model, and the big model generates answers based on the optimization of the suggestions, including: During the dialogue process of each task of heating intelligent decision-making, the heating dispatch control personnel can input improvement suggestions through text boxes or voice, and convert the input improvement suggestions into instructions that can be understood by the large model by using the natural language understanding function; The big model adjusts its internal knowledge base and reasoning logic based on the improvement suggestions to optimize the generated answers.

[0058] In actual application, an example is given to illustrate the optimization of heating scheduling in response to extreme weather: during an extremely cold weather period, the heating scheduling control center detected a sharp increase in heating load, and the existing heating strategy could not meet the demand. At this time, the dispatcher input improvement suggestions through voice input: "Now it is extremely cold weather, the load is increasing too fast, start all backup heat sources, and give priority to heating in residential areas, increase the water supply temperature in residential areas, and pay close attention to the pressure of the pipe network to avoid overpressure." The system uses natural language understanding to convert this speech into instructions that the big model can understand: 1) Start all backup heat sources; 2) Set heating priorities, with residential areas given priority; 3) Increase the set point of water supply temperature in residential areas; 4) Monitor the pipeline pressure in real time, issue an early warning and adjust the heating strategy when the pressure approaches the upper limit.

[0059] Based on these instructions, the big model adjusts its internal knowledge base and reasoning logic. It retrieves knowledge such as the startup process of the backup heat source and the heating parameter standards for the residential area from the knowledge base, adjusts the reasoning logic, and replans the heating scheduling plan. In the original plan, the startup conditions of the backup heat source were relatively conservative. After the big model was optimized, the backup heat source was immediately started to increase the heating output. At the same time, the heat network flow was redistributed to give priority to heating the residential area, and the water supply temperature in the residential area was increased to ensure that the residents were warm indoors. In this process, the big model also monitors the pipe network pressure in real time, and by adjusting the flow distribution in each area, it avoids the risk of excessive pipe network pressure, and finally optimizes and generates a scheduling plan that is more in line with the heating demand in extreme weather.

[0060] Take an example to illustrate the adjustment of heating strategy after equipment failure: a key circulation pump in the heating system suddenly fails, affecting the heating of some areas. After discovering the failure, the dispatcher enters improvement suggestions in the text box: "If the circulation pump fails, switch to the backup pump immediately, calculate the heating gap in the affected area, allocate heat from other heat sources to make up for it, and estimate the maintenance time, and adjust the subsequent heating strategy according to the maintenance progress." The natural language understanding feature parses the suggestions into instructions that the big model can recognize: 1) Perform standby pump switching operation; 2) Calculate the heating load gap in the affected area; 3) Develop a plan to allocate heat from other heat sources based on the heat supply capacity of the heat source and the heat network transmission capacity; 4) Dynamically adjust subsequent heating strategies based on the maintenance time estimate provided by the maintenance department.

[0061] After receiving the instruction, the large model searches the knowledge base for knowledge such as the switching process of the standby pump, the calculation method of the heating load, and the heat source allocation rules. Adjust the reasoning logic and quickly issue the standby pump start-up instruction to ensure that the heating network cycle is not interrupted. Then, accurately calculate the heating gap in the affected area, combine the real-time heating capacity of other heat sources and the hydraulic conditions of the heating network, and formulate a plan to allocate heat from adjacent heat sources to ensure that the heating in the affected area is basically stable. At the same time, the large model connects with the information system of the maintenance department to obtain the maintenance progress in real time, and plans the subsequent heating strategy in advance according to the maintenance time estimate. For example, before the maintenance is completed, continuously optimize the heat allocation plan, and after the maintenance is completed, smoothly restore the normal heating strategy, thereby optimizing and generating a comprehensive heating strategy to deal with equipment failures.

[0062] In this embodiment, the heating scheduling control personnel evaluates multiple answers generated by the large model according to the prompt words, and trains and optimizes the parameters of the large model according to the evaluation results, including: The heating dispatching and control personnel use the trained large model to generate multiple different answers based on the prompt words, and construct evaluation indicators according to the language comprehension index, semantic rationality index, context coherence index, heating knowledge accuracy index, and heating reasoning ability index to evaluate and assign values ​​to multiple different answers; Different questions, answers, and evaluation results are used as input, and reinforcement learning algorithms are used to update and optimize large model parameters.

[0063] In actual applications, taking the evaluation and model optimization of the winter peak dispatching strategy of the heating system as an example, during the winter heating peak period, the heating dispatching and control personnel are faced with the problem of how to balance the heating demand of each region and optimize energy distribution. At this time, the dispatcher enters the prompt word: "In the current extremely cold weather, the heating load continues to peak, and the heating effect of some old communities is not good. How to formulate a reasonable heating dispatching strategy?" The big model generates the following different answers based on the knowledge and logic learned during training: Answer 1: First, increase the heating output of all heat sources to ensure the overall heating output. Then, conduct pressure testing on the heating network of old residential areas. If the pressure is insufficient, appropriately increase the pressure of the branch network of old residential areas. At the same time, arrange staff to inspect the heating equipment of old residential areas and deal with possible faults in a timely manner.

[0064] Answer 2: Prioritize increasing the heating power of heat sources near old residential areas and accurately supply heat to old residential areas. At the same time, install intelligent regulating valves in the heat network to adjust the flow distribution in real time according to the temperature feedback of each area to ensure balanced heating in each area. In addition, activate the emergency plan and prepare backup heat sources to deal with possible emergencies.

[0065] Answer 3: Check the hydraulic balance of the heating network, recalculate the flow demand of each area, and adjust the opening of each branch valve according to the calculation results to ensure even heat distribution. For old residential areas, increase the circulation pump to increase the flow rate of heat medium and enhance the heating effect. In addition, keep in close contact with the meteorological department and adjust the heating strategy in advance according to future weather changes.

[0066] The heating dispatch control personnel evaluate and assign values ​​to these answers according to the constructed evaluation indicators: Language comprehension index: The three answers are clearly expressed in language, allowing dispatchers to understand their core content, and are all given high scores.

[0067] Semantic rationality index: Answer 1 directly increases the output of all heat sources without considering energy supply and equipment carrying capacity, and its semantic rationality is slightly lower; Answer 2’s idea of ​​precise heating and intelligent regulating valve adjustment is reasonable, but the description of the timing of starting the backup heat source is vague; Answer 3’s content of comprehensively checking the hydraulic balance and adjusting the strategy according to the weather is semantically reasonable, and Answer 3 has the highest score in the comprehensive evaluation.

[0068] Context coherence index: All three answers revolve around solving the current heating problem, and the context is coherent. However, answer 2 is not closely connected with the previous text when mentioning the intelligent regulating valve and backup heat source. Answers 1 and 3 have better coherence.

[0069] Accuracy index of heating knowledge: The practice of increasing the pressure of branch pipelines in old residential areas in answer one may lead to safety hazards, and there are problems with the accuracy of heating knowledge; answers two and three are accurate in terms of knowledge on heating equipment operation and scheduling strategies, and answers two and three scored higher.

[0070] Heating reasoning ability index: Answer three makes a comprehensive strategy based on reasoning based on the characteristics of the heating network and the problems of old residential areas, and has the strongest reasoning ability; Answer two has some reasoning, but it is not comprehensive enough; Answer one has simple and direct reasoning and lacks consideration of complex situations.

[0071] After comprehensively evaluating and assigning values ​​to various indicators, different questions (How to formulate a reasonable heating scheduling strategy in the current extremely cold weather?), answers (the three answers mentioned above), and evaluation results are used as inputs, and the reinforcement learning algorithm is used to update and optimize the parameters of the large model. In this process, the large model learns that in complex situations such as extremely cold weather and poor heating in some areas, priority should be given to hydraulic balance analysis and flow optimization distribution of the heating network, rational use of smart devices and adjustment of strategies according to weather changes, rather than simply increasing the output of heat sources. By continuously accumulating such evaluation and optimization processes, the large model can generate more reasonable and accurate answers when facing problems related to heating scheduling, thereby improving the scientificity and effectiveness of heating scheduling decisions.

[0072] In this embodiment, the large model of each task of heating intelligent decision-making after training and optimization is integrated into each intelligent agent set, and the large model is used to drive the intelligent agent to guide each intelligent agent to perform environmental perception, collaborative interaction and autonomous adaptive decision-making in a simulation environment, including: According to the diversified tasks of intelligent heating decision-making, corresponding task agents are set up, including operation perception agent, parameter prediction agent, fault diagnosis agent, scheduling planning agent and energy consumption evaluation agent; Set up bottom-level equipment agents, including heat source agents, heat network agents, heat storage agents and heat user agents, respectively collect data of heat source equipment, heat network, heat storage devices and heat users and transmit them to each task agent, and issue and execute decision instructions transmitted by each task agent; Optimize the input and output interfaces of each task model to connect it with each task agent of heating intelligent decision-making, allocate a large model calling channel for each task agent, and integrate each task model into the corresponding task agent; Each task agent uses its perception unit to perceive the environment to obtain collected data, extract key information from the data, and transmit it to the big model of the decision-making unit. At the same time, a collaborative interaction mechanism is established between the task agent and the underlying equipment agent to conduct collaborative data interaction and optimize task execution and task resource allocation. Finally, through the big model of the decision-making unit, in-depth analysis and reasoning learning are performed, and each task agent makes autonomous decisions to generate decision instructions to adapt to the dynamically changing operating conditions of the heating system.

[0073] In actual applications, taking the case of a heating system encountering extreme weather as an example, the workflow of each agent includes: 1) Data collection of underlying device agents Heat source agent: quickly collects operating data of heat source equipment such as thermal power plants and boiler rooms, including boiler combustion efficiency, steam production, heating power, etc. For example, the heat source agent of a thermal power plant detects that the fuel consumption of the boiler has increased significantly in order to maintain a high heating output.

[0074] Heating network intelligent agent: Real-time monitoring of the temperature, pressure and flow distribution of the heating network. It was found that the water supply temperature of the heating network in some areas dropped and the flow also fluctuated. This was because the increase in heating load led to uneven heat distribution.

[0075] Thermal storage agent: reports the thermal storage status and releasable heat of the thermal storage device, providing a reference for subsequent heating scheduling.

[0076] Thermal User Agent: Collects feedback information from the user side. For example, some users report that the indoor temperature is low and fails to meet the comfort temperature standard.

[0077] 2) Task agent environmental perception and decision-making Operation perception agent: The perception unit obtains data transmitted by the underlying equipment agent and extracts key information, such as the operating status change trend of each device, the real-time working condition of the heating network, etc. This information is transmitted to the big model of the decision-making unit. After analysis, the big model determines that the heating system as a whole is in a high-load operation state, and some links show signs of instability.

[0078] Parameter prediction agent: Combining historical data and current meteorological information, the large model of its decision-making unit predicts that the heating load will continue to rise for a period of time, and the outdoor temperature is expected to drop further in the next few hours.

[0079] Fault diagnosis agent: Analyzes the collected data and determines that although there is no serious fault in the current heating system, some equipment has potential risks due to high-load operation. For example, the motor current of a circulation pump is close to the rated value, which may cause an overload fault.

[0080] Dispatching planning agent: Based on the information provided by the operation perception, parameter prediction and fault diagnosis agents, the dispatching strategy is formulated with the assistance of the large model of its decision-making unit. It is decided to increase the heating output of the thermal power plant and start some standby boiler rooms; adjust the valve opening of the heating network, optimize the flow distribution, and give priority to heating for residential users and key areas; at the same time, control the heat storage agent to release some heat to make up for the heating gap.

[0081] Energy consumption assessment agent: Real-time assessment of energy consumption, found that as the heating load increases, energy consumption increases significantly. Through large-scale model analysis, it proposes suggestions for optimizing the operating parameters of heat source equipment to reduce energy consumption while ensuring the quality of heating, such as adjusting the combustion air ratio of the boiler to improve combustion efficiency.

[0082] 3) Collaborative interaction and instruction execution A collaborative interaction mechanism is established between the task agents to share information and coordinate actions. For example, when formulating strategies, the scheduling agent refers to the energy-saving suggestions of the energy consumption assessment agent to avoid excessive energy consumption; the fault diagnosis agent informs the scheduling agent of the potential risks of the equipment so that the safe operation of the equipment can be considered in the scheduling strategy.

[0083] The scheduling and planning agent sends the decision instructions to the underlying equipment agent for execution. After receiving the instruction to increase the heating output, the heat source agent adjusts the equipment operating parameters of the thermal power plant and boiler room; the heat network agent adjusts the valve opening according to the instruction to achieve the optimal distribution of the heat network flow; the heat user agent feedbacks the heating adjustment situation to the user and guides the user to use the heating facilities reasonably.

[0084] In this embodiment, the simulation operation environment for providing heating intelligent decision interaction includes: managing simulation time to analyze the operation status of the heating system at different times, updating the status of the simulation environment, and providing a flexible time control interface to allow real-time operation of simulation time during system operation; The user interaction interface for providing heating intelligent decision-making interaction includes: providing a graphical parameter configuration interface, setting simulation parameters and supporting the loading of parameter configuration files through sliders, drop-down menus, and text box interactive elements, and visually displaying and analyzing simulation results in the form of text, pictures, and videos; The dynamic parameter simulation of the scene when providing heating intelligent decision-making interaction includes: using a large model to build a simulation model of the actual heating system and a dynamic environment to perform system simulation.

[0085] In this embodiment, the construction evaluates the performance index of each agent, and evaluates and optimizes the behavior and decision-making effect of each agent according to the performance index and the decision simulation result, including: Construct performance indicators to evaluate the intelligent decision-making tasks of intelligent heating, including task completion, task completion efficiency, decision-making adaptability, and collaboration between intelligent agents; According to the behavior trajectory, decision-making process and output decision simulation results of each task agent, the performance index value of each task agent is calculated and compared with the preset standards to evaluate the performance of each task agent; Based on the evaluation results, the behavior and decision-making strategies of each task agent are optimized and adjusted, including: improving the algorithm and model structure within the agent, enhancing the agent's environmental perception and adaptability, and optimizing the collaboration mechanism and communication protocol between agents.

[0086] In several embodiments provided in the present application, it should be understood that the disclosed system and method can also be implemented in other ways. The system embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0087] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part. If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device to perform all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

[0088] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A heating intelligent decision-making system based on an artificial intelligence large model, characterized in that: It includes: Data layer: used to collect, store and manage various types of data on the source, grid, load and storage of the heating system, and as a large-scale data set for knowledge base establishment and large model training layer; Knowledge base establishment and large model training layer: used to perform data preprocessing, data annotation, knowledge extraction and representation on large-scale data sets of the heating system, and then establish a knowledge base for diversified tasks of intelligent decision-making in heating; It is also used to train a large model based on the knowledge base of diversified tasks of intelligent heating decision-making and the prompt learning strategy of intelligent heating decision-making. The large model captures the knowledge and behavioral characteristics of each task of intelligent heating decision-making, and establishes the intrinsic connection between each prompt word and high-quality answer. It also fine-tunes the large model according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, so as to adapt to specific business scenarios, decision-making tasks and various intelligent agents. Human-machine enhanced optimization big model layer: used for the heating dispatching and control personnel to make improvement suggestions to the big model during the dialogue process of each task of heating intelligent decision-making, and the big model optimizes the generated answers based on the suggestions; it is also used for the heating dispatching and control personnel to evaluate the multiple answers generated by the big model based on the prompt words, and train and optimize the big model parameters based on the evaluation results; Intelligent agent decision layer: used to integrate the large models of each task of intelligent heating decision-making after training and optimization into each intelligent agent, use the large model to drive the intelligent agent, and guide each intelligent agent to perform environmental perception, collaborative interaction and autonomous adaptive decision-making in the simulation environment; Simulation layer: used to provide the simulation operation environment, user interaction interface and scenario dynamic parameter simulation for heating intelligent decision-making interaction. After the heating dispatching control personnel raise questions, each intelligent agent generates decision simulation data and supports displaying answers to users in the form of text, pictures and videos. Agent evaluation and optimization layer: used to construct performance indicators for evaluating each agent, and evaluate and optimize the behavior and decision-making effects of each agent based on the performance indicators and decision simulation results.

2. The intelligent heating decision-making system according to claim 1 is characterized in that: The heating system source, grid, load and storage multi-type data includes: historical operation data of the heating system source, grid, load and storage, heating operation and maintenance and scheduling management manual, dynamic change parameters, and physical property data of each device in the system; Among them, the historical operation data include: the operating parameters of the source-side heat source equipment, the heat metering data of each pipeline in the heat network, the indoor temperature and heat consumption data, the operating status and storage and release heat of the heat storage equipment, the historical heat load data, and the on-demand scheduling decision strategy, the historical energy consumption analysis data, and the abnormal diagnosis data; the heating operation and maintenance and scheduling management manual includes the heating inspection content, equipment maintenance strategy, fault handling process, scheduling process, and emergency scheduling response processing rules; the dynamically changing parameters include outdoor environmental parameters and sudden abnormal fault parameters; the physical property data of each device in the system include the fixed parameters, operating characteristics, and service life of each source network, load storage equipment, and pipeline network.

3. The intelligent heating decision-making system according to claim 1 is characterized in that: After data preprocessing, data annotation, knowledge extraction and representation of the large-scale data set of the heating system, a knowledge base of diversified tasks of intelligent decision-making for heating is established, including: Clarify the diverse tasks of intelligent heating decision-making, including operation perception, parameter prediction, fault diagnosis, scheduling planning and energy consumption evaluation, and determine the knowledge scope and knowledge types covered by the knowledge base based on the diverse tasks; the knowledge types include factual knowledge, rule knowledge and process knowledge; According to the diverse tasks, knowledge scope and knowledge type of intelligent heating decision-making, data preprocessing, labeling and classification of large-scale data sets of heating systems are carried out; Knowledge extraction: Directly formulate rules for knowledge extraction based on clear rules and experience; use machine learning algorithms to discover potential knowledge from data; use natural language processing technology to extract key information and knowledge from unstructured text data; Knowledge representation: The extracted knowledge is represented based on production rule representation, framework representation and semantic network representation; The extracted and represented knowledge is stored in the corresponding database table or graph structure according to the format and requirements of the knowledge base management system, and indexes and associations between knowledge are established. A knowledge retrieval interface is provided to support keyword retrieval, semantic retrieval and association retrieval methods, and finally a knowledge base for diversified tasks of intelligent heating decision-making is established.

4. The intelligent heating decision-making system according to claim 1, characterized in that: The large model training is performed based on the diversified task knowledge base of heating intelligent decision-making and the prompt learning strategy of heating intelligent decision-making. The large model captures the knowledge and behavior characteristics of each task of heating intelligent decision-making and establishes the intrinsic connection between each prompt word and high-quality answer, including: In the large model training stage, the diversified knowledge base of heating intelligent decision-making is input into the base large model as part of the training corpus, and the heating professional terms, semantic relations, and logical relations between various factors in the heating system in the knowledge base are learned. The operation mode, behavioral characteristics and intelligent decision-making process of the heating system under different working conditions are analyzed. According to the characteristics of the heating intelligent decision-making task, the thinking process and rules are preset through the thinking chain method, and the decision-making task problem is decomposed into multiple sub-problems. Logically clear steps are provided in the form of prompts, and different types of prompt templates are designed to guide the large model to analyze and process the input information according to the steps and processes; The designed prompt template is combined with the content of the knowledge base and input into the large model, so that the model can learn the ability to extract relevant knowledge from the knowledge base and generate answers based on the prompt information, as well as collect user feedback on the answers and establish the intrinsic connection between the prompt words and the answers.

5. The intelligent heating decision-making system according to claim 1 is characterized in that: The large model is fine-tuned according to the specific scenario requirements, task objectives and constraints of each heating task to establish a large model for each task, including: For the task of sensing the operation of the heating system, the specific scenario requirement is to sense the operating status of each device in the heating system, and the task goal is to fully and accurately obtain the operating data of each device. The constraint condition is that the data collection is compatible with the data interfaces of devices of different types and manufacturers. For the task of predicting heating system parameters, the specific scenario requirement is to predict heat load and heating network parameters. The task goal is to predict and accurately predict the changes in heating system operating parameters in advance. The constraints are to consider the influencing factors of weather changes, user heating behavior, and historical data rules. For the heating system fault diagnosis task, the specific scenario requirement is to diagnose the fault type of the heating system and determine the cause of the fault. The task goal is to accurately identify the fault type and locate the fault cause. The constraint condition is the complexity of the fault scenario. For the scheduling and planning task of the heating system, the specific scenario requirement is to generate scheduling plans of different time scales for each source, grid, load and storage device according to the supply and demand relationship of the heat load. The task goal is to meet the heating demand, minimize the economic cost and reduce energy consumption. The constraints are the operating parameter constraints of each device and the thermal power balance constraints. For the task of evaluating the energy consumption of the heating system, the specific scenario requirements are to evaluate the energy consumption of the heating system, analyze the factors affecting energy consumption, and put forward energy-saving suggestions. The task goal is to reduce the energy consumption of the heating system, and the constraint condition is to consider the energy consumption differences under different operating conditions; The big model is fine-tuned according to the specific scenario requirements, objective functions, and constraints of each task, and a big model for each task is established, including an operation perception big model, a parameter prediction big model, a fault diagnosis big model, a scheduling planning big model, and an energy consumption assessment big model.

6. The intelligent heating decision-making system according to claim 1, characterized in that: During the dialogue process of each task of heating intelligent decision-making, the heating dispatch control personnel put forward improvement suggestions to the big model, and the big model generates answers based on the optimization suggestions, including: During the dialogue process of each task of heating intelligent decision-making, the heating dispatch control personnel can input improvement suggestions through text boxes or voice, and convert the input improvement suggestions into instructions that can be understood by the large model by using the natural language understanding function; The big model adjusts its internal knowledge base and reasoning logic based on the improvement suggestions to optimize the generated answers.

7. The intelligent heating decision-making system according to claim 1, characterized in that: The heating dispatch control personnel evaluate the multiple answers generated by the large model according to the prompt words, and train and optimize the parameters of the large model according to the evaluation results, including: The heating dispatching and control personnel use the trained large model to generate multiple different answers based on the prompt words, and construct evaluation indicators according to the language comprehension index, semantic rationality index, context coherence index, heating knowledge accuracy index, and heating reasoning ability index to evaluate and assign values ​​to multiple different answers; Different questions, answers, and evaluation results are used as input, and reinforcement learning algorithms are used to update and optimize large model parameters.

8. The intelligent heating decision-making system according to claim 1, characterized in that: The aforementioned method integrates the large models of each task of intelligent heating decision-making after training and optimization into each intelligent agent set, uses the large model to drive the intelligent agent, and guides each intelligent agent to perform environmental perception, collaborative interaction, and autonomous adaptive decision-making in a simulation environment, including: According to the diversified tasks of intelligent heating decision-making, corresponding task agents are set up, including operation perception agent, parameter prediction agent, fault diagnosis agent, scheduling planning agent and energy consumption evaluation agent; Set up bottom-level equipment agents, including heat source agents, heat network agents, heat storage agents and heat user agents, respectively collect data of heat source equipment, heat network, heat storage devices and heat users and transmit them to each task agent, and issue and execute decision instructions transmitted by each task agent; Optimize the input and output interfaces of each task model to connect it with each task agent of heating intelligent decision-making, allocate a large model calling channel for each task agent, and integrate each task model into the corresponding task agent; Each task agent uses its perception unit to perceive the environment to obtain collected data, extract key information from the data, and transmit it to the big model of the decision-making unit. At the same time, a collaborative interaction mechanism is established between the task agent and the underlying equipment agent to conduct collaborative data interaction and optimize task execution and task resource allocation. Finally, through the big model of the decision-making unit, in-depth analysis and reasoning learning are performed, and each task agent makes autonomous decisions to generate decision instructions to adapt to the dynamically changing operating conditions of the heating system.

9. The intelligent heating decision-making system according to claim 1, characterized in that: The simulation operation environment for providing heating intelligent decision interaction includes: managing simulation time to analyze the operation status of the heating system at different times, updating the status of the simulation environment, and providing a flexible time control interface to allow real-time operation of simulation time during system operation; The user interaction interface for providing heating intelligent decision-making interaction includes: providing a graphical parameter configuration interface, setting simulation parameters and supporting the loading of parameter configuration files through sliders, drop-down menus, and text box interactive elements, and visually displaying and analyzing simulation results in the form of text, pictures, and videos; The dynamic parameter simulation of the scene when providing heating intelligent decision-making interaction includes: using a large model to build a simulation model of the actual heating system and a dynamic environment to perform system simulation.

10. The intelligent heating decision-making system according to claim 1, characterized in that: The construction evaluates the performance indicators of each intelligent agent, and evaluates and optimizes the behavior and decision-making effect of each intelligent agent based on the performance indicators and decision simulation results, including: Construct performance indicators to evaluate the intelligent decision-making tasks of intelligent heating, including task completion, task completion efficiency, decision-making adaptability, and collaboration between intelligent agents; According to the behavior trajectory, decision-making process and output decision simulation results of each task agent, the performance index value of each task agent is calculated and compared with the preset standards to evaluate the performance of each task agent; Based on the evaluation results, the behavior and decision-making strategies of each task agent are optimized and adjusted, including: improving the algorithm and model structure within the agent, enhancing the agent's environmental perception and adaptability, and optimizing the collaboration mechanism and communication protocol between agents.

Images