Multi-heat-source heat supply operation method based on multiple agents and distributed gas-fired boiler

Abstract

The invention discloses a multi-heat-source heat supply operation method based on multiple agents and distributed gas boilers, which comprises the following steps: selecting a part of heating stations, respectively configuring at least one small gas boiler, forming a multi-heat-source heat supply system with a source side thermal power plant, and establishing a multi-heat-source heat supply system operation mechanism; the thermal station edge computing terminal predicts a thermal load demand value of each thermal station and feeds back the thermal load demand value to the cloud heat supply dispatching center, and the cloud heat supply dispatching center establishes a thermal load optimal distribution model by combining the heat supply capacity, the heat supply cost and the carbon emission of the thermal power plant and each small gas-fired boiler; each heating station in the system is regarded as an intelligent agent, a cloud heat supply dispatching center introduces a deep reinforcement learning algorithm to carry out offline training on the intelligent agents, the trained intelligent agents of the heating stations are issued to corresponding edge computing terminals of the heating stations, and local information of the related heating stations is observed; and generating equipment control instruction actions of the heating station and the small gas-fired boiler, and carrying out online intelligent decision making.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-heat source operation of heating systems, and particularly relates to a multi-heat source heating operation method based on multi-agent and distributed gas boilers. BACKGROUND

[0002] With the development of urbanization, the heating load is increasing year by year, and in particular, during the heating severe cold period, the heat supply of the source side heat plant is usually seriously insufficient, and it is difficult to meet the user heat load demand. Therefore, how to realize the balance between heat load supply and demand and ensure the stability of heat users has become a key problem in the heating industry.

[0003] At present, in order to solve this problem, small peak shaving heat sources are usually added to the heating station as backup heat sources of the source side heat plant. When the heat load demand of the heat plant cannot be met, the small peak shaving heat sources are started to make up for the lack of heat and improve the energy utilization rate of the heating system. However, there are many heating stations in the heating system, and the operation characteristics and basic properties of the heating stations are different. Considering the factors such as heating cost, how to select the key heating stations to add small peak shaving heat sources and perform heat load distribution, equipment operation precise control of the source side heat plant and small peak shaving heat sources, improve the heating quality, stability and system operation performance is a problem that needs to be solved at present.

[0004] Based on the above technical problems, a new multi-heat source heating operation method based on multi-agent and distributed gas boilers needs to be designed. SUMMARY

[0005] The technical problem to be solved by the application is to overcome the shortcomings of the prior art and provide a multi-heat source heating operation method based on multi-agent and distributed gas boilers. Some heating stations are selected to be configured with small gas boilers, and heat load prediction, heat load optimized distribution and heating station agent training are realized through the cooperation of the edge computing terminal of the heating station and the cloud heating dispatching center to generate equipment control instruction actions of the heating station and the small gas boiler, thereby improving the heating quality, stability and system operation performance.

[0006] In order to solve the above technical problems, the technical scheme of the application is as follows:

[0007] The application provides a multi-heat source heating operation method based on multi-agent and distributed gas boilers, which comprises the following steps:

[0008] S1, obtaining multi-dimensional factors of station inside, station outside and self operation of the heating station, selecting some heating stations to be respectively configured with at least one small gas boiler, and forming a multi-heat source heating system with the source side heat plant;

[0009] S2. Establish a multi-heat source heating system operation mechanism: use the source-side thermal power plant as an active heat source to meet the basic heat load requirements of each heating station, and use each small gas boiler as a passive heat source to provide additional heat to make up for the heat load shortfall;

[0010] S3. In the first stage, the pre-configured edge computing terminal at the thermal power station predicts the heat load demand value of each thermal power station during each time period and feeds it back to the cloud-based heating dispatch center. The cloud-based heating dispatch center then establishes a heat load optimization allocation model based on the heating capacity, heating cost, and carbon emissions of the thermal power plant and each small gas-fired boiler. It then outputs the heating load allocation results for each time period for the source-side thermal power plant and each small gas-fired boiler at each thermal power station.

[0011] In the second phase, each heating station in the multi-source heating system is considered an intelligent agent. The cloud-based heating dispatch center uses a deep reinforcement learning algorithm to perform offline training on each heating station intelligent agent. The intelligent agent gradually learns a near-optimal control strategy through continuous interaction with the heating system. The trained intelligent agent for each heating station is then sent to the corresponding edge computing terminal of the heating station.

[0012] S5. The edge computing terminal of the thermal power station observes the local information of the relevant thermal power station, generates equipment control instructions for the thermal power station and small gas boilers, and makes online intelligent decisions.

[0013] Furthermore, the S1 includes:

[0014] Obtaining data on the heating station both inside and outside the station, including geospatial data on where small gas boilers can be placed inside and outside the station, surrounding gas supply conditions, and heating network pipeline layout;

[0015] Obtain the thermal power station's own operating data, including historical fault records of the thermal power station, historical heat supply and demand data, energy consumption data of the thermal power station, heat load ranges of subordinate users, and the types and importance of heat users;

[0016] Based on the internal and external data of the thermal power station and its own operation data, a thermal power station feature library is constructed, and after feature extraction, the key features of the thermal power station are obtained;

[0017] Based on the key features of the heating station, a thermal station portrait label library is generated, including the thermal station geospatial feature label, gas supply feature label, thermal station energy consumption feature label, and thermal station static attribute label. The weight value of each label is combined to generate a portrait of each thermal station.

[0018] Taking into account the profiles of each thermal power station and the economic cost constraints, some key thermal power stations were selected, each of which was equipped with at least one small gas boiler, and formed a multi-heat source heating system with the source-side thermal power plant.

[0019] Further, the S2 includes:

[0020] The source-side thermal power plant is used as an active heat source. Based on the heat load results of each thermal power station, the thermal power plant shall give priority to meeting the heat load demand by increasing the operating load of the unit and adjusting the extraction steam volume, circulating water volume or combustion volume of the heating unit. If the thermal power plant still cannot meet the heat load demand after adjustment, indicating that the supply capacity of the thermal power plant has been exceeded, then each small gas boiler shall be used as a passive heat source. Based on the heat load shortage value, the gas combustion volume shall be gradually increased to supplement additional heat so that the water supply of the small gas boiler and the water supply of the thermal power station heat exchanger are mixed to reach the required water supply temperature, thereby collaboratively meeting the heat load demand of each thermal station.

[0021] In addition, when some equipment in the thermal power plant fails, equipment is under maintenance, or abnormalities occur in the heating network, the heating output of small gas boilers will be increased and the operating parameters of the heating network will be adjusted to optimize the distribution of heat media and coordinate heating with the thermal power plant.

[0022] Furthermore, the S3 includes:

[0023] Multiple pre-configured edge computing terminals at heating stations acquire relevant operational and meteorological data that influence the heat load of each heating station. These data are then trained using a pre-set machine learning algorithm and a heat load prediction model for each heating station. The model then obtains the heat load demand value for each heating station during each time period and feeds it back to the pre-configured cloud-based heating dispatch center.

[0024] The cloud-based heating dispatch center obtains the heat load demand value of each heating station at each time period, as well as the heating capacity and heating cost data of thermal power plants and small gas boilers at each heating station. With the goal of meeting the heat load demand of each heating station, minimizing heating costs and carbon emissions, and taking the equipment heating capacity and the heat load supply and demand balance of each heating station as constraints, a heat load optimization allocation model is established, which is expressed as:

[0025]

[0026] T is the total dispatch period; N is the number of heating stations; T hs,j,t 、T gb,j,t are the hot water temperatures output by the thermal power plant and the small gas boiler to the jth thermal power station at time t; T sv,g,j,t is the water supply temperature setting value of the jth thermal power station at time t; M is the number of equipment in the multi-heat source heating system; C i,t is the operation and maintenance cost of the i-th device at time t; hs The price of fuel required for the thermal power plant; gb The price of gas required for a small gas boiler; P hs,t P is the operating power of the thermal power plant unit at time t; gb,t is the operating power of the small gas boiler at time t; hs ,ξgb are the carbon emission factors of thermal power plants and small gas boilers, respectively;

[0027] Solve the heat load optimization distribution model and output the heat load output of the source-side thermal power plant and the heat load output distribution results of the small gas boilers in each thermal power station in each period.

[0028] Furthermore, the constraints of heat load supply and demand balance satisfied by each thermal power station are expressed as follows:

[0029]

[0030] Q hs,t is the thermal load output of the power plant at time t; Q gb,t is the heat load output of the small gas boiler at time t; Q hl,t is the heat loss when the heat network transmits heat at time t; Q j,t is the heat load demand of the jth thermal power station at time t.

[0031] Furthermore, S4 also includes: allocating the thermal load output of each unit in the thermal power plant according to the thermal load output distribution results of the source-side thermal power plant in each time period and combining the operating characteristics of each unit in the thermal power plant, and then adjusting the unit operating parameters according to the thermal load output allocated to each unit.

[0032] Further, the S4 includes:

[0033] In the second stage, each heating station in the multi-heat source heating system is considered as an intelligent agent, responsible for controlling the regulating equipment of each heating station, the number of operating small gas boilers, and related execution equipment according to the heat load output distribution results of the small gas boilers in each heating station in each time period;

[0034] Set the state space, action space, and reward of the agent, and model the control problem of each thermal power station's regulating equipment, the number of small gas boilers in operation, and related execution equipment as a Markov decision problem;

[0035] The cloud-based heating dispatch center introduces a deep reinforcement learning algorithm model. After initializing the algorithm parameters, the total number of steps for agent training is set, with each step corresponding to a moment.

[0036] At each moment, when the algorithm model is offline trained, each intelligent agent interacts with the simulation environment of the multi-heat source heating system and generates corresponding actions according to the current state and behavioral strategy. The environment then feeds back the new state and reward value based on the action of the intelligent agent and stores it in the experience replay pool as an experience tuple. A batch of experience data is then randomly sampled from the experience replay pool to maximize the cumulative reward, update the algorithm model parameters, and gradually learn a near-optimal control action strategy until the algorithm model converges or the preset number of training times is reached, completing the offline training of each thermal station intelligent agent.

[0037] The trained deep reinforcement learning algorithm models of each thermal station agent are encapsulated into executable agent programs and sent to the corresponding thermal station edge computing terminals.

[0038] Furthermore, the state space of the intelligent agent includes: operating data of the thermal power station itself and the small gas boilers, environmental data and the heat load output distribution results of the small gas boilers in each thermal power station;

[0039] The action space of the agent includes: the opening of the regulating valve associated with each thermal power station, the speed of the water pump, the start and stop status of the small gas boiler arranged in the thermal power station, the opening of the gas flow regulating valve, the operating parameters of the water pump supporting the small gas boiler, and the combustion power of the burner;

[0040] The rewards of the intelligent agent include: optimal stability of heating temperature, highest gas utilization efficiency and minimum operating cost.

[0041] Furthermore, the offline training of the algorithm model also includes: introducing an imitation learning mechanism, where the intelligent agent actively learns the experience data of heating experts to obtain an initial behavior strategy.

[0042] Further, the S5 includes:

[0043] The edge computing terminal at the thermal power station observes local information of the relevant thermal power station, including the supply and return water temperature, supply and return water pressure, circulating water flow, the heat load output distribution results of the small gas boilers in each thermal power station, outdoor temperature, gas flow, combustion efficiency, and furnace pressure of the small gas boilers;

[0044] The intelligent agent in the edge computing terminal of the thermal power station makes online intelligent decisions based on the local information of the thermal power station and issues equipment control instructions for the thermal power station and small gas boilers.

[0045] The beneficial effects of the present invention are:

[0046] (1) The present invention configures small gas boilers in some thermal power stations to form a multi-heat source heating system with the source-side thermal power plant. When a thermal power plant fails or has insufficient heating capacity, the small gas boiler can be used as a backup heat source to ensure the heating demand of the thermal power station and reduce the risk of heating interruption. In addition, by considering the multi-dimensional factors of the thermal power station's internal and external stations and its own operation, the small gas boiler can be flexibly configured according to the specific conditions of different thermal power stations, so that the heating system can better adapt to various complex working conditions and demand changes.

[0047] (2) The present application meets the basic heat load demand by taking the thermal power plant as an active heat source, utilizes its scale effect and high efficiency to ensure that the basic heat supply demand is met. At the same time, the small gas boiler is used as a passive heat source to provide additional heat, which is started only when the heat load is insufficient, avoiding unnecessary operation of the small gas boiler, improving energy utilization efficiency and reducing operating costs; and the heat supply capacity of the thermal power plant and the small gas boiler can be flexibly adjusted according to the actual heat load demand, realizing the collaborative operation of multiple heat sources, and better balancing the heat supply capacity of the entire heat supply system and the heat load demand;

[0048] (3) The present application predicts the heat load demand value of each heat station at each time period through the edge computing terminal of the heat station, and feeds back to the cloud heat supply dispatching center, which can accurately master the change of the heat load, and establish a heat load optimization distribution model combining the heat supply capacity, heat supply cost and carbon emission of the thermal power plant and each small gas boiler, so as to realize the precise dispatching of the heat source and minimize the heat supply cost and carbon emission under the premise of meeting the heat load demand;

[0049] (4) The present application regards each heat station in the multi-heat source heat supply system as an intelligent agent, which is trained offline through a deep reinforcement learning algorithm. The intelligent agent can learn the nearly optimal control strategy through continuous interaction with the heat supply system, which means that the heat station can automatically adjust the operating parameters according to different working conditions and heat load demands, improving the overall performance and operating efficiency of the heat supply system. In addition, the training process enables the intelligent agent to adapt to various complex heat supply scenarios and changes, enhancing the adaptability of the heat supply system to different environments and demands;

[0050] (5) The present application observes the local information of the relevant heat station through the edge computing terminal of the heat station, generates device control instruction actions, and makes online intelligent decisions, which enables the heat station to use the intelligent agent to adjust the equipment operating parameters of the heat station and the small gas boiler in real time according to the real-time operating data and working conditions, realize real-time tracking and optimization control of the heat load, and improve the quality and stability of heat supply;

[0051] (6) The present application reduces the computing pressure and network transmission burden of the cloud by having the edge computing terminal undertake part of the data processing and decision-making tasks, enables the regulating equipment to respond and control accurately and quickly, and in the case of cloud failure or connection interruption, the edge computing terminal can still make autonomous decisions and control to a certain extent according to the locally stored data and the preset intelligent agent; the cloud heat supply dispatching center uses powerful computing power to conduct in-depth analysis and mining from a global perspective, establishes a heat load optimization distribution model and an intelligent agent training, provides decision-making basis for the optimized operation of the heat supply system, and improves the learning efficiency of the algorithm model and the accuracy of the control strategy.

[0052] Other features and advantages will be described in the following description, and in part will 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.

[0053] 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

[0054] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is a flow chart of a multi-heat source heating operation method based on multi-agents and distributed gas boilers of the present invention;

[0056] Figure 2 This is a schematic structural diagram of the multi-heat source heating system of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] like Figure 1 、 Figure 2 As shown, this embodiment provides a multi-heat source heating operation method based on multi-agents and distributed gas boilers, which includes:

[0059] S1. Obtain multi-dimensional factors of the thermal power station's internal and external operations, and select some thermal power stations to configure at least one small gas boiler each, forming a multi-heat source heating system with the source-side thermal power plant;

[0060] S2. Establish a multi-heat source heating system operation mechanism: use the source-side thermal power plant as an active heat source to meet the basic heat load requirements of each heating station, and use each small gas boiler as a passive heat source to provide additional heat to make up for the heat load shortfall;

[0061] S3. In the first stage, the pre-configured edge computing terminal at the thermal power station predicts the heat load demand value of each thermal power station during each time period and feeds it back to the cloud-based heating dispatch center. The cloud-based heating dispatch center then establishes a heat load optimization allocation model based on the heating capacity, heating cost, and carbon emissions of the thermal power plant and each small gas-fired boiler. It then outputs the heating load allocation results for each time period for the source-side thermal power plant and each small gas-fired boiler at each thermal power station.

[0062] In the second phase, each heating station in the multi-source heating system is considered an intelligent agent. The cloud-based heating dispatch center uses a deep reinforcement learning algorithm to perform offline training on each heating station intelligent agent. The intelligent agent gradually learns a near-optimal control strategy through continuous interaction with the heating system. The trained intelligent agent for each heating station is then sent to the corresponding edge computing terminal of the heating station.

[0063] S5. The edge computing terminal of the thermal power station observes the local information of the relevant thermal power station, generates equipment control instructions for the thermal power station and small gas boilers, and makes online intelligent decisions.

[0064] In this embodiment, S1 includes:

[0065] Obtaining data on the heating station both inside and outside the station, including geospatial data on where small gas boilers can be placed inside and outside the station, surrounding gas supply conditions, and heating network pipeline layout;

[0066] Obtain the thermal power station's own operating data, including historical fault records of the thermal power station, historical heat supply and demand data, energy consumption data of the thermal power station, heat load ranges of subordinate users, and the types and importance of heat users;

[0067] Based on the internal and external data of the thermal power station and its own operation data, a thermal power station feature library is constructed, and after feature extraction, the key features of the thermal power station are obtained;

[0068] Based on the key features of the heating station, a thermal station portrait label library is generated, including the thermal station geospatial feature label, gas supply feature label, thermal station energy consumption feature label, and thermal station static attribute label. The weight value of each label is combined to generate a portrait of each thermal station.

[0069] Taking into account the profiles of each thermal power station and the economic cost constraints, some key thermal power stations were selected, each of which was equipped with at least one small gas boiler, and formed a multi-heat source heating system with the source-side thermal power plant.

[0070] In practical applications, the geographical spatial data of the inside and outside of the station where small gas-fired boilers can be arranged are obtained, including information such as topography, land use type, building distribution, etc., to provide a geographical spatial basis for subsequent small gas-fired boiler layout planning. For example, some heat stations can deploy small gas-fired boilers in basements, and some heat stations have no space inside to deploy, so they can consider deploying small gas-fired boilers outside the station.

[0071] In addition, in addition to deploying small gas-fired boilers, related pipelines also need to be laid to connect with the original heat network, so the layout information of the original heat network pipeline, such as the direction, pipe diameter, and burial depth, as well as the feasibility and complexity of laying new pipelines, also need to be considered. For example, the inlet and outlet positions, pipe diameter specifications, pipe material, and the direction and slope of the station's heat network pipeline are mastered, and the corresponding heating areas and user types (such as residential heating, commercial heating, etc.) of different pipelines are determined. The types, quantities, and operating parameters of the heat exchange equipment inside the station are understood, such as the heat exchange efficiency of the heat exchanger, the inlet and outlet water temperature, etc. These data are very important for evaluating the matching degree of small gas-fired boilers with the existing heat network and the space for improving the heating capacity. In addition, the cross-distribution of the heat network pipeline outside the station and other municipal facilities (such as water supply and drainage pipelines, power cables, etc.) is analyzed to avoid conflicts with other facilities when arranging small gas-fired boilers and connecting pipelines, and the possibility of heat network pipeline expansion and extension is considered to reserve space for future growth of heating demand.

[0072] Some heat stations have a large number of heat users and a large heating area, which is prone to supply and demand imbalance. It is necessary to arrange small gas-fired boilers for such heat stations, and some heat stations have heat users such as hospitals and schools, which need to arrange small gas-fired boilers to ensure the reliability and stability of heating.

[0073] The location and quantity of the gas access point inside the station are determined, the supply pressure, pipe diameter, and type (such as natural gas, coal gas, etc.) and composition of the gas are understood, and these parameters directly affect the selection and operating efficiency of small gas-fired boilers. It is checked whether the gas metering device has been equipped inside the station, and whether its accuracy and range meet the future use requirements of small gas-fired boilers, and the stability and reliability of gas supply are understood, including whether there is seasonal gas supply tension or pressure fluctuation, etc.

[0074] The distribution of the surrounding gas pipeline outside the heat station is investigated, including the direction, burial depth, and distance to the nearest access point of the heat station, the gas supply capacity and remaining gas supply of the surrounding gas pipeline are understood, and whether it can meet the gas demand of the newly added small gas-fired boiler is evaluated. The distribution position and operation of the surrounding gas supply facilities (such as gas pressure regulating stations, gas filling stations, etc.) are understood, as well as their distance and connection mode with the heat station, and the safety and emergency guarantee capacity of gas supply are analyzed.

[0075] In practical applications, the process of generating each heat station image includes:

[0076] 1) Feature library construction:

[0077] Collect the in-station and out-station data and self-operation data of the heat station from various data sources. For geographic spatial data, it can be obtained through geographic information system (GIS) technology; gas supply condition data can be obtained from relevant information of the gas company; heat network pipeline layout data can be obtained from design drawings or relevant engineering documents of the heat station. Self-operation data is extracted from the monitoring system, historical record documents, etc. of the heat station. Integrate these data from different sources and store them in a unified database for subsequent processing;

[0078] 2) Feature extraction:

[0079] Geographic spatial feature extraction: From the geographic spatial data of the in-station and out-station where small gas boilers can be arranged, extract key features such as the latitude and longitude of the heat station, the land area, the spatial distribution proportion of each functional area in the station, the area and shape coefficient of the available land outside the station, and the distance to the surrounding main roads. These features can describe the geographic location and spatial layout characteristics of the heat station;

[0080] Gas supply feature extraction: From the surrounding gas supply condition data, extract features such as the mean value and fluctuation range of the gas supply pressure, the size of the gas pipe diameter, the coding value of the gas type, the stability index of the gas supply (such as the number of gas supply interruptions, the pressure variation coefficient, etc.), and the distance to the nearest gas pressure regulating station. These features can reflect the reliability and applicability of gas supply;

[0081] Heat network pipeline layout feature extraction: From the heat network pipeline layout data, extract features such as the total length of the heat network pipeline, the statistical characteristics of the pipe diameter distribution (such as the average pipe diameter, the maximum pipe diameter, etc.), the laying method code of the pipeline (such as 1 for overhead and 2 for buried), the heat preservation performance index of the pipeline (such as the thermal conductivity, the heat loss rate, etc.), and the heat network pipeline coverage density of different heating areas. These features help to understand the structure and transmission efficiency of the heat network;

[0082] Self-operation feature extraction: From the self-operation data of the heat station, extract features such as the frequency and type distribution of historical fault records, the mean value and peak value of historical heat supply and demand, the daily / monthly / yearly change trend of energy consumption, the upper and lower limits of the heat load interval of the users under jurisdiction, the distribution proportion of heat user types (such as the proportion of residents, businesses, industries, etc.), the number and proportion of important heat users, etc. These features can reflect the operation status and service object characteristics of the heat station;

[0083] 3) Image label library generation:

[0084] Thermal station geospatial feature labels: Based on the results of geospatial feature extraction, geospatial feature labels are generated for thermal stations. For example, if a thermal station is located in a city center, it can be labeled as "city center type"; if it occupies a large area and has a lot of available open space, it can be labeled as "spacious type"; depending on the distance from the main road, it can be labeled as "convenient transportation type" or "inconvenient transportation type", etc.

[0085] Gas supply feature label: Generate a gas supply feature label based on the results of gas supply feature extraction. If the gas supply pressure is stable and the pipe diameter is large, it can be labeled as "good gas supply type"; if the gas supply has frequent fluctuations or insufficient pressure, it can be labeled as "unstable gas supply type";

[0086] Energy consumption behavior feature labels for thermal power stations: Based on the results of energy consumption behavior feature extraction for thermal power stations, energy consumption behavior feature labels are generated. For example, if historical fluctuations in heat supply and demand are small and energy consumption is relatively stable, it can be labeled as "stable energy consumption"; if peak heat supply and demand are high and energy consumption fluctuates dramatically within a specific time period, it can be labeled as "fluctuating energy consumption"; based on the main time period distribution of energy consumption, it can be labeled as "daytime energy consumption" or "nighttime energy consumption";

[0087] Heating station static attribute labels: Generate static attribute labels based on some basic static attributes of the heating station, such as the construction year of the heating station, the heating company it belongs to, and the nature of the heating area (such as residential, commercial, industrial, etc.). For example, a relatively new heating station can be labeled "Newly built heating station", a station belonging to a large heating company can be labeled "Large company owned", and a heating area mainly in residential areas can be labeled "Residential heating type", etc.

[0088] 4) Thermal power station portrait generation:

[0089] Determine label weights: Use expert experience, hierarchical analysis, principal component analysis, and other methods to determine the weights of various profile labels. For example, for heating stations located in areas with tight gas supply, the weight of gas supply feature labels may be relatively high; for heating stations serving large industrial users, the weights of labels related to heat user type and importance may be assigned larger values.

[0090] Comprehensive portrait generation: Generate a portrait of the thermal power station based on each label and its weight value. Multiply the value of each label by its corresponding weight, and then add them together to obtain a comprehensive feature vector. Based on this feature vector, classify and describe the thermal power station. For example, a thermal power station has a high score for the geospatial feature label, a medium score for the gas supply feature label, a low score for the energy consumption feature label, and a medium score for the static attribute label. The weights of each label are 0.3, 0.2, 0.3, and 0.2 respectively. Based on these scores and weights, the thermal power station can be comprehensively described as a thermal power station with a superior geographical location but energy consumption behavior that needs to be optimized, and gas supply and static attributes are at a medium level. Different portrait descriptions can be set for different score intervals according to actual conditions to more intuitively understand the characteristics and conditions of the thermal power station.

[0091] In actual applications, the process of selecting key thermal power stations includes:

[0092] 1) Evaluate the thermal station profile:

[0093] Analyze geographic spatial characteristics: Prioritize heating stations located at the end or edge of the heating network, where the network pipeline coverage density is low. Deploying small gas boilers in these locations can effectively compensate for heating shortages, improve heating performance, and reduce heat loss from long-distance heating.

[0094] Investigate gas supply characteristics: Choose a heating station with a good gas supply, such as stable gas supply pressure, large pipe diameter, and close proximity to a gas pressure regulating station. Good gas supply conditions ensure stable operation of small gas boilers, reducing the risk of increased operating costs and heating interruptions caused by gas supply problems.

[0095] Study energy consumption characteristics: Focus on thermal power stations with large fluctuations in energy consumption, especially during the peak winter heating season, where the imbalance between heat supply and demand is prominent and where heat shortages have historically occurred. Deploying small gas boilers can provide additional heat during peak periods, meeting user needs and improving heating quality.

[0096] Considering static properties: For heating stations whose heating areas are mainly for important users (such as hospitals, schools, government agencies, etc.), or whose heating areas are large and have a large number of heat users, special consideration should be given. Ensuring the stability and reliability of heating in these areas is crucial. Small gas boilers can be used as backup or supplementary heat sources to enhance the safety of the heating system.

[0097] 2) Economic cost analysis:

[0098] Initial investment cost: Calculate the initial investment costs for equipment purchase, installation and commissioning, and associated pipeline modifications for deploying small gas boilers at each thermal power station. For thermal power stations with complex site conditions and requiring extensive pipeline modifications, the initial investment cost may be high and requires careful evaluation.

[0099] Operating costs: Analyze the gas costs, maintenance costs, labor costs, etc. required to operate a small gas boiler. Consider the energy usage characteristics of different thermal power stations, predict their operating time and energy consumption, and evaluate long-term operating costs. For example, a thermal power station with stable energy consumption and low load has relatively low operating costs and is more suitable for the deployment of a small gas boiler.

[0100] Energy-saving benefits: Evaluate the energy-saving effects of deploying small gas-fired boilers on the entire heating system. For example, consider whether it can reduce the long-distance heating losses of the source-side thermal power plant, improve energy efficiency, and reduce the overall energy consumption costs of the system. Priority should be given to thermal power stations that can achieve cascaded energy utilization and improve system energy efficiency through small gas-fired boilers.

[0101] Cost-benefit ratio: Calculate the cost-benefit ratio of deploying small gas boilers at each heating station, that is, the ratio of the expected energy savings and benefits from improved heating quality to the investment and operating costs. Select heating stations with the highest cost-benefit ratio for deployment to ensure that the economic benefits are maximized while meeting heating demand;

[0102] 3) Comprehensive assessment and decision-making

[0103] Establish an evaluation index system: Based on various aspects of the thermal power station's profile and economic cost factors, establish a comprehensive evaluation index system and clarify the weight of each indicator. For example, in areas with high heating quality requirements, the weight of indicators involving important users in the thermal power station's energy consumption behavior characteristics and static properties can be appropriately increased; in economically cost-sensitive areas, the weight of investment cost and operating cost indicators can be increased;

[0104] Quantitative evaluation: Each heating station is quantitatively scored according to the evaluation index system, and the various characteristics and economic cost factors of the heating station profile are converted into specific numerical values. For example, a heating station with a stable gas supply can be given a higher score on the gas supply characteristic index, and a heating station with a low initial investment cost can be given a higher score on the economic cost index;

[0105] Comprehensive ranking: Ranking each heating station based on its comprehensive score, selecting the top-ranked stations as key stations for the placement of small gas boilers. At the same time, the selection results should be appropriately adjusted and optimized based on actual conditions, such as the overall planning of the heating system and the capital budget, to ensure that the selected key heating stations can both meet heating demand and achieve efficient operation of the multi-heat source heating system within economic cost constraints.

[0106] In this embodiment, S2 includes:

[0107] The source-side thermal power plant is used as an active heat source. Based on the heat load results of each thermal power station, the thermal power plant shall give priority to meeting the heat load demand by increasing the operating load of the unit and adjusting the extraction steam volume, circulating water volume or combustion volume of the heating unit. If the thermal power plant still cannot meet the heat load demand after adjustment, indicating that the supply capacity of the thermal power plant has been exceeded, then each small gas boiler shall be used as a passive heat source. Based on the heat load shortage value, the gas combustion volume shall be gradually increased to supplement additional heat so that the water supply of the small gas boiler and the water supply of the thermal power station heat exchanger are mixed to reach the required water supply temperature, thereby collaboratively meeting the heat load demand of each thermal station.

[0108] In addition, when some equipment in the thermal power plant fails, equipment is under maintenance, or abnormalities occur in the heating network, the heating output of small gas boilers will be increased and the operating parameters of the heating network will be adjusted to optimize the distribution of heat media and coordinate heating with the thermal power plant.

[0109] In this embodiment, S3 includes:

[0110] Multiple pre-configured edge computing terminals at heating stations acquire relevant operational and meteorological data that influence the heat load of each heating station. These data are then trained using a pre-set machine learning algorithm and a heat load prediction model for each heating station. The model then obtains the heat load demand value for each heating station during each time period and feeds it back to the pre-configured cloud-based heating dispatch center.

[0111] The cloud-based heating dispatch center obtains the heat load demand value of each heating station at each time period, as well as the heating capacity and heating cost data of thermal power plants and small gas boilers at each heating station. With the goal of meeting the heat load demand of each heating station, minimizing heating costs and carbon emissions, and taking the equipment heating capacity and the heat load supply and demand balance of each heating station as constraints, a heat load optimization allocation model is established, which is expressed as:

[0112]

[0113] T is the total dispatch period; N is the number of heating stations; T hs,j,t 、T gb,j,t are the hot water temperatures output by the thermal power plant and the small gas boiler to the jth thermal power station at time t; T sv,g,j,t is the water supply temperature setting value of the jth thermal power station at time t; M is the number of equipment in the multi-heat source heating system; C i,t is the operation and maintenance cost of the i-th device at time t; hs The price of fuel required for the thermal power plant; gb The price of gas required for a small gas boiler; P hs,t P is the operating power of the thermal power plant unit at time t; gb,t is the operating power of the small gas boiler at time t; hs ,ξ gbare the carbon emission factors of thermal power plants and small gas boilers, respectively;

[0114] Solve the heat load optimization distribution model and output the heat load output of the source-side thermal power plant and the heat load output distribution results of the small gas boilers in each thermal power station in each period.

[0115] It should be noted that the edge computing terminal is installed in the thermal power station and is responsible for collecting operational data and meteorological data that affect the load of the thermal power station in real time, and performing preliminary processing on the collected data, including data cleaning, data normalization and other operations, to provide reliable data input for the training of the machine learning algorithm. In combination with the pre-set machine learning algorithm support vector machine and thermal power station portrait, model training is performed locally to establish a thermal load prediction model for each thermal power station, obtain the thermal load demand value of each thermal power station in each period and upload it to the cloud-based heating dispatching center.

[0116] The edge computing terminal needs to have network communication capabilities and can establish a connection with the cloud-based heating dispatch center through Ethernet, 4G / 5G, and other network methods to upload the processed data to the cloud. The following factors need to be considered when setting up the edge computing terminal:

[0117] 1) Scale and complexity of thermal power stations: Large thermal power stations have large variations in heat load, numerous devices, and heavy data collection and processing tasks. They may require separate edge computing terminals to ensure real-time processing and analysis of large amounts of data and accurately predict heat loads. Small thermal power stations, on the other hand, have smaller data volumes and relatively simple operations. Multiple small thermal power stations may consider sharing a single edge computing terminal.

[0118] 2) Data transmission and communication costs: If the thermal power stations are concentrated and the communication distance between stations is short and the cost is low, the collected data from multiple thermal power stations can be aggregated and processed on an edge computing terminal to reduce hardware investment and data transmission costs. If the thermal power stations are dispersed, configuring an edge computing terminal for each thermal power station can avoid the high cost and signal attenuation caused by long-distance data transmission;

[0119] 3) Reliability and redundancy requirements: For key heating stations in the heating system, to ensure heat load forecasting and system operation reliability, even if they are small in scale, it is possible to set up separate edge computing terminals to achieve independent data processing and backup, preventing the operation of multiple heating stations from being affected by failures in shared terminals.

[0120] 4) Computing Power and Resource Requirements: Evaluate the computing power and resources required for edge computing terminals based on the complexity of each heating station's heat load prediction model, data processing volume, and real-time requirements. If the combined data processing requirements of multiple heating stations exceed the capabilities of a single edge computing terminal, separate terminals will be required for each heating station or a subset of stations.

[0121] One edge computing terminal per heating station: This solution is suitable for large, complex, or critical heating stations. These stations typically have high heat loads, numerous devices, and complex operating conditions. They require independent edge computing terminals to ensure real-time and accurate data processing, as well as system reliability and stability. For example, heating stations serving major heating areas with large heat load fluctuations, or heating stations serving areas with critical users who have extremely high heat quality requirements and cannot tolerate data processing delays or failures.

[0122] Multiple heating stations, one edge computing terminal: For some small, concentrated heating stations, multiple stations can share a single edge computing terminal. For example, in a residential community or small commercial district, multiple small heating stations with relatively stable heat loads and close proximity can aggregate data onto a single edge computing terminal for unified processing. This reduces hardware costs and operational complexity while still meeting the basic requirements for heat load forecasting and system operation.

[0123] The cloud-based heating dispatching center is responsible for receiving the heat load demand values ​​of each thermal power station in each time period uploaded by each edge computing terminal, and at the same time collecting data such as the heating capacity and heating cost of thermal power plants and small gas boilers in each thermal power station, and storing these data in the database; it should be noted that the goal of meeting the heat load demand of each thermal power station is indirectly expressed by meeting the water supply temperature set value of the thermal power station. When the water supply temperature meets the standard, it indirectly indicates that the heat load demand meets the standard.

[0124] When solving the heat load optimization distribution model, a genetic optimization algorithm is used to solve the model to determine the heat load distribution results of the thermal power plant and small gas boilers in different time periods, including:

[0125] Step 1) Initialize the population code:

[0126] The heating power of each heat source of thermal power plants and small gas boilers is used as a decision variable, and the heating power of all heat sources is combined into a chromosome using real number coding.

[0127] Generate initial population: randomly generate a certain number of chromosomes within the feasible range of heating capacity of each heat source to form the initial population;

[0128] Step 2) Fitness evaluation:

[0129] For each chromosome in the population (i.e., a heat load distribution scheme), calculate its objective function value. To facilitate the operation of the genetic algorithm, the objective function is usually converted into a fitness function. The larger the fitness value, the better the heat load distribution scheme corresponding to the chromosome.

[0130] Step 3) Select an action:

[0131] Roulette wheel selection is used to select individuals from the current population to enter the mating pool. The probability of each individual being selected is proportional to its fitness value. The higher the fitness value, the greater the probability of being selected.

[0132] Step 4) Crossover operation:

[0133] Randomly select two individuals in the mating pool as parents, and perform crossover operations using single-point crossover or multi-point crossover to generate offspring individuals;

[0134] Step 5) Mutation operation:

[0135] Perform mutation operations on the offspring individuals obtained after crossover. Randomly select one or more genes in the offspring chromosomes (i.e., the heating power of the heat source) and make small random changes within their feasible range;

[0136] Step 6) Constraint processing:

[0137] After crossover and mutation operations, individuals that do not meet the constraints may be generated. These individuals need to be repaired to meet the equipment heating capacity and heat load supply and demand balance constraints. For example, for individuals that do not meet the equipment heating capacity constraint, the out-of-range heating power is adjusted to within the feasible range; for individuals that do not meet the heat load supply and demand balance constraint, the heating power of each heat source is proportionally adjusted so that its sum equals the total heat load demand.

[0138] Step 7) Update the population:

[0139] The offspring individuals that have undergone selection, crossover, mutation and constraint processing replace some individuals in the current population to form a new population;

[0140] Step 8) Termination condition judgment:

[0141] Determine whether the termination conditions are met. Common termination conditions include: reaching the maximum number of iterations, or the fitness value no longer increasing. If the termination conditions are met, the algorithm ends and outputs the individual with the largest fitness value in the current population as the optimal solution; otherwise, return to step 2 and continue iterating.

[0142] In this embodiment, the heat load supply and demand balance constraint condition satisfied by each heat station is expressed as:

[0143]

[0144] Q hs,t is the thermal load output of the power plant at time t; Q gb,t is the heat load output of the small gas boiler at time t; Q hl,t is the heat loss when the heat network transmits heat at time t; Q j,t is the heat load demand of the jth thermal power station at time t.

[0145] In this embodiment, S4 also includes: allocating the thermal load output of each unit in the thermal power plant according to the thermal load output distribution results of the source-side thermal power plant in each time period and combining the operating characteristics of each unit in the thermal power plant, and then adjusting the unit operating parameters according to the thermal load output allocated to each unit.

[0146] In this embodiment, the S4 includes:

[0147] In the second stage, each heating station in the multi-heat source heating system is considered as an intelligent agent, responsible for controlling the regulating equipment of each heating station, the number of operating small gas boilers, and related execution equipment according to the heat load output distribution results of the small gas boilers in each heating station in each time period;

[0148] Set the state space, action space, and reward of the agent, and model the control problem of each thermal power station's regulating equipment, the number of small gas boilers in operation, and related execution equipment as a Markov decision problem;

[0149] The cloud-based heating dispatch center introduces a deep reinforcement learning algorithm model. After initializing the algorithm parameters, the total number of steps for agent training is set, with each step corresponding to a moment.

[0150] At each moment, when the algorithm model is offline trained, each intelligent agent interacts with the simulation environment of the multi-heat source heating system and generates corresponding actions according to the current state and behavioral strategy. The environment then feeds back the new state and reward value based on the action of the intelligent agent and stores it in the experience replay pool as an experience tuple. A batch of experience data is then randomly sampled from the experience replay pool to maximize the cumulative reward, update the algorithm model parameters, and gradually learn a near-optimal control action strategy until the algorithm model converges or the preset number of training times is reached, completing the offline training of each thermal station intelligent agent.

[0151] The trained deep reinforcement learning algorithm models of each thermal station agent are encapsulated into executable agent programs and sent to the corresponding thermal station edge computing terminals.

[0152] In practical applications, the MASAC algorithm is used as the deep reinforcement learning algorithm model. The SAC algorithm, also known as the Flexible Actor-Critic algorithm, is a deep reinforcement learning algorithm that integrates the actor-critic framework, off-policy updating, and the maximum entropy principle. The actor-critic framework consists of two parts: an actor and a critic. The actor generates actions based on the current policy, while the critic evaluates the quality of these actions. Off-policy updating allows the agent to learn from data collected from behavioral policies that differ from the current policy, enhancing data utilization. The maximum entropy principle encourages exploration by introducing an entropy term in policy optimization.

[0153] The single-agent SAC algorithm is extended to multiple agents to form the MASAC algorithm model. The specific training process includes:

[0154] 1) Initialization phase:

[0155] When training begins, the model parameters of each agent are initialized, including the parameters of the policy network (actor) and the value network (critic), as well as the experience replay buffer and temperature parameters.

[0156] 2) Environmental interaction stage:

[0157] At each time step, each agent observes its current state;

[0158] The agent decides the action to perform based on its migration policy (generated by the policy network);

[0159] After all agents have taken actions, the environment calculates the immediate reward for each agent based on these actions and transitions to a new state;

[0160] 3) Experience Collection Phase:

[0161] Each agent uses the experience replay buffer to collect historical environment information, including observed states, executed actions, rewards obtained, and new states. This information will serve as a dataset for subsequent training;

[0162] 4) Model update phase:

[0163] Policy Optimization: The Bellman equation is used to update the policy network, and gradient descent is used to adjust the policy network parameters to maximize the agent's long-term discounted reward. At the same time, the entropy term encourages the agent to explore a wider range of policy spaces, avoiding premature convergence to a local optimum.

[0164] Q-value optimization: Use a target Q-value network to calculate the target Q-value, and then update the parameters of the Q-value network through gradient descent so that the Q-value network can more accurately estimate the action-value function;

[0165] 5) Repeat training:

[0166] The process of environmental interaction, experience collection, and model update is repeated continuously. As training progresses, the agent gradually learns better strategies to optimize the overall goal.

[0167] 6) End of training:

[0168] When the preset training end conditions are met, such as the number of training steps reaches a certain threshold, the performance indicators of the intelligent agent reach the expected level, or the training curve tends to be stable, the training process ends. The policy network obtained at this time is the trained model and can be used in actual decision-making and control tasks.

[0169] In this embodiment, the state space of the intelligent agent includes: the operation data of the thermal power station itself and the small gas boilers, environmental data and the heat load output distribution results of the small gas boilers in each thermal power station;

[0170] The action space of the agent includes: the opening of the regulating valve associated with each thermal power station, the speed of the water pump, the start and stop status of the small gas boiler arranged in the thermal power station, the opening of the gas flow regulating valve, the operating parameters of the water pump supporting the small gas boiler, and the combustion power of the burner;

[0171] The rewards of the intelligent agent include: optimal stability of heating temperature, highest gas utilization efficiency and minimum operating cost.

[0172] In this embodiment, the offline training of the algorithm model further includes: introducing an imitation learning mechanism, whereby the intelligent agent actively learns the experience data of heating experts to obtain an initial behavior strategy.

[0173] In this embodiment, S5 includes:

[0174] The edge computing terminal at the thermal power station observes local information of the relevant thermal power station, including the supply and return water temperature, supply and return water pressure, circulating water flow, the heat load output distribution results of the small gas boilers in each thermal power station, outdoor temperature, gas flow, combustion efficiency, and furnace pressure of the small gas boilers;

[0175] The intelligent agent in the edge computing terminal of the thermal power station makes online intelligent decisions on the equipment control instructions for the thermal power station and small gas boilers based on the local information of the thermal power station.

[0176] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, 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 portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that 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 using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0177] In addition, the functional modules in each embodiment 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. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of each embodiment 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.

[0178] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A multi-heat source heating operation method based on multi-agent and distributed gas boilers, characterized in that: It includes: S1. Obtain multi-dimensional factors of the thermal power station's internal and external operations, and select some thermal power stations to configure at least one small gas boiler each, forming a multi-heat source heating system with the source-side thermal power plant; S2. Establish a multi-heat source heating system operation mechanism: use the source-side thermal power plant as an active heat source to meet the basic heat load requirements of each heating station, and use each small gas boiler as a passive heat source to provide additional heat to make up for the heat load shortfall; S3. In the first stage, the pre-configured edge computing terminal at the thermal power station predicts the heat load demand value of each thermal power station during each time period and feeds it back to the cloud-based heating dispatch center. The cloud-based heating dispatch center then establishes a heat load optimization allocation model based on the heating capacity, heating cost, and carbon emissions of the thermal power plant and each small gas-fired boiler. It then outputs the heating load allocation results for each time period for the source-side thermal power plant and each small gas-fired boiler at each thermal power station. In the second phase, each heating station in the multi-source heating system is considered an intelligent agent. The cloud-based heating dispatch center uses a deep reinforcement learning algorithm to perform offline training on each heating station intelligent agent. The intelligent agent gradually learns a near-optimal control strategy through continuous interaction with the heating system. The trained intelligent agent for each heating station is then sent to the corresponding edge computing terminal of the heating station. S5. The edge computing terminal of the thermal power station observes the local information of the relevant thermal power station, generates equipment control instructions for the thermal power station and small gas boilers, and makes online intelligent decisions.

2. The multi-heat source heating operation method according to claim 1, characterized in that: Said S1 comprises: Obtaining data on the heating station both inside and outside the station, including geospatial data on where small gas boilers can be placed inside and outside the station, surrounding gas supply conditions, and heating network pipeline layout; Obtain the thermal power station's own operating data, including historical fault records of the thermal power station, historical heat supply and demand data, energy consumption data of the thermal power station, heat load ranges of subordinate users, and the types and importance of heat users; Based on the internal and external data of the thermal power station and its own operation data, a thermal power station feature library is constructed, and after feature extraction, the key features of the thermal power station are obtained; Based on the key features of the heating station, a thermal station portrait label library is generated, including the thermal station geospatial feature label, gas supply feature label, thermal station energy consumption feature label, and thermal station static attribute label. The weight value of each label is combined to generate a portrait of each thermal station. Taking into account the profiles of each thermal power station and the economic cost constraints, some key thermal power stations were selected, each of which was equipped with at least one small gas boiler, and formed a multi-heat source heating system with the source-side thermal power plant.

3. The multi-heat source heating operation method according to claim 1, characterized in that: The S2 includes: The source-side thermal power plant is used as an active heat source. Based on the heat load results of each thermal power station, the thermal power plant shall give priority to meeting the heat load demand by increasing the operating load of the unit and adjusting the extraction steam volume, circulating water volume or combustion volume of the heating unit. If the thermal power plant still cannot meet the heat load demand after adjustment, indicating that the supply capacity of the thermal power plant has been exceeded, then each small gas boiler shall be used as a passive heat source. Based on the heat load shortage value, the gas combustion volume shall be gradually increased to supplement additional heat so that the water supply of the small gas boiler and the water supply of the thermal power station heat exchanger are mixed to reach the required water supply temperature, thereby collaboratively meeting the heat load demand of each thermal station. In addition, when some equipment in the thermal power plant fails, equipment is under maintenance, or abnormalities occur in the heating network, the heating output of small gas boilers will be increased and the operating parameters of the heating network will be adjusted to optimize the distribution of heat media and coordinate heating with the thermal power plant.

4. The multi-heat source heating operation method according to claims 1 and 2, characterized in that: The S3 includes: Multiple pre-configured edge computing terminals at heating stations acquire relevant operational and meteorological data that influence the heat load of each heating station. These data are then trained using a pre-set machine learning algorithm and a heat load prediction model for each heating station. The model then obtains the heat load demand value for each heating station during each time period and feeds it back to the pre-configured cloud-based heating dispatch center. The cloud-based heating dispatch center obtains the heat load demand value of each heating station at each time period, as well as the heating capacity and heating cost data of thermal power plants and small gas boilers at each heating station. With the goal of meeting the heat load demand of each heating station, minimizing heating costs and carbon emissions, and taking the equipment heating capacity and the heat load supply and demand balance of each heating station as constraints, a heat load optimization allocation model is established, which is expressed as: T is the total dispatch period; N is the number of heating stations; T hs,j,t 、T gb,j,t are the hot water temperatures output from the thermal power plant and the small gas boiler to the jth thermal power station at time t; T sv,g,j,t is the water supply temperature setting value of the jth thermal power station at time t; M is the number of equipment in the multi-heat source heating system; C i,t is the operation and maintenance cost of the i-th device at time t; hs The price of fuel required for the thermal power plant; gb The price of gas required for a small gas boiler; P hs,t P is the operating power of the thermal power plant unit at time t; gb,t is the operating power of the small gas boiler at time t; hs ,ξ gb are the carbon emission factors of thermal power plants and small gas boilers, respectively; Solve the heat load optimization distribution model and output the heat load output of the source-side thermal power plant and the heat load output distribution results of the small gas boilers in each thermal power station in each period.

5. The multi-heat source heating operation method according to claim 4, characterized in that: The constraints of heat load supply and demand balance satisfied by each heat station are expressed as: Q hs,t is the thermal load output of the power plant at time t; Q gb,t is the heat load output of the small gas boiler at time t; Q hl,t is the heat loss when the heat network transmits heat at time t; Q j,t is the heat load demand of the jth thermal power station at time t.

6. The multi-heat source heating operation method according to claim 1, characterized in that: The S4 also includes: allocating the heat load output of each unit in the thermal power plant according to the heat load output distribution results of the source-side thermal power plant in each time period and combining the operating characteristics of each unit in the thermal power plant, and then adjusting the unit operating parameters according to the heat load output allocated to each unit.

7. The multi-heat source heating operation method according to claim 1, characterized in that: The S4 includes: In the second stage, each heating station in the multi-heat source heating system is considered as an intelligent agent, responsible for controlling the regulating equipment of each heating station, the number of operating small gas boilers, and related execution equipment according to the heat load output distribution results of the small gas boilers in each heating station in each time period; Set the state space, action space, and reward of the agent, and model the control problem of each thermal power station's regulating equipment, the number of small gas boilers in operation, and related execution equipment as a Markov decision problem; The cloud-based heating dispatch center introduces a deep reinforcement learning algorithm model. After initializing the algorithm parameters, the total number of steps for agent training is set, with each step corresponding to a moment. At each moment, when the algorithm model is offline trained, each intelligent agent interacts with the simulation environment of the multi-heat source heating system and generates corresponding actions according to the current state and behavioral strategy. The environment then feeds back the new state and reward value based on the action of the intelligent agent and stores it in the experience replay pool as an experience tuple. A batch of experience data is then randomly sampled from the experience replay pool to maximize the cumulative reward, update the algorithm model parameters, and gradually learn a near-optimal control action strategy until the algorithm model converges or the preset number of training times is reached, completing the offline training of each thermal station intelligent agent. The trained deep reinforcement learning algorithm models of each thermal station agent are encapsulated into executable agent programs and sent to the corresponding thermal station edge computing terminals.

8. The multi-heat source heating operation method according to claim 7, characterized in that: The state space of the intelligent agent includes: the operation data of the thermal power station itself and the small gas boilers, environmental data and the heat load output distribution results of the small gas boilers in each thermal power station; The action space of the agent includes: the opening of the regulating valve associated with each thermal power station, the speed of the water pump, the start and stop status of the small gas boiler arranged in the thermal power station, the opening of the gas flow regulating valve, the operating parameters of the water pump supporting the small gas boiler, and the combustion power of the burner; The rewards of the intelligent agent include: optimal stability of heating temperature, highest gas utilization efficiency and minimum operating cost.

9. The multi-heat source heating operation method according to claim 7, characterized in that: The offline training of the algorithm model also includes: introducing an imitation learning mechanism, where the intelligent agent actively learns the experience data of heating experts to obtain an initial behavior strategy.

10. The multi-heat source heating operation method according to claim 1, characterized in that: The S5 includes: The edge computing terminal at the thermal power station observes local information of the relevant thermal power station, including the supply and return water temperature, supply and return water pressure, circulating water flow, the heat load output distribution results of the small gas boilers in each thermal power station, outdoor temperature, gas flow, combustion efficiency, and furnace pressure of the small gas boilers; The intelligent agent in the edge computing terminal of the thermal power station makes online intelligent decisions based on the local information of the thermal power station and issues equipment control instructions for the thermal power station and small gas boilers.

Images