Heat exchange station regulation and control method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of heat exchange station regulation and control, and discloses a heat exchange station regulation and control method and system based on artificial intelligence. The method comprises the following steps: collecting real-time meteorological monitoring data, real-time room temperature monitoring data and real-time heat exchange station monitoring data in a monitoring area, and constructing a real-time dynamic topological graph; inputting the real-time dynamic topological graph into a heat supply load prediction model, and performing heat supply load prediction to obtain a real-time heat supply load prediction result; inputting the real-time heat supply load prediction result and the real-time heat exchange station monitoring data into a regulation and control strategy generation model, and generating a regulation and control strategy to obtain a real-time heat exchange station regulation and control strategy; and the real-time heat exchange station regulation and control strategies are sent to the heat exchange stations in the monitoring area, the corresponding real-time heat exchange station regulation and control strategies are executed, and the heat exchange stations are regulated and controlled. According to the invention, the problems of dependence on a single data source, limited model prediction capability and lack of self-adaptive capability of a regulation and control strategy in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of heat exchange station control technology, specifically relating to a heat exchange station control method and system based on artificial intelligence. Background Technology

[0002] Heat exchange stations are typically located at the end of a heating (or cooling) system, close to the user area. Their main function is to transfer the heat from the centrally supplied primary heat medium (such as high-temperature water or steam) to the secondary heat medium (usually lower-temperature water) through heat exchange equipment. The heated secondary heat medium is then delivered to the user terminals (such as radiators, floor heating, air conditioning units, etc.) to meet the user's needs for heating, domestic hot water, or air conditioning.

[0003] Traditional heat exchange station control methods often rely on manual experience or simple preset programs, making it difficult to adapt to complex and dynamically changing actual heating demands. For example, changes in outdoor weather conditions (temperature, humidity, wind speed, etc.), the actual usage of buildings (indoor temperature, building type, etc.), and the operating status of the heating network itself (supply and return water temperature, flow rate, pressure, etc.) all significantly affect heating load and effectiveness. Traditional control methods cannot perceive these changes in real time and accurately, and make optimal adjustments, potentially leading to insufficient or excessive heating, resulting in energy waste, decreased user comfort, and increased operating costs.

[0004] In recent years, with the development of artificial intelligence technology, especially the application of deep learning and reinforcement learning in the control of complex systems, new ideas have been provided for solving the problems of heating system regulation. However, how to effectively integrate various real-time monitoring data, construct a model that can reflect the dynamic characteristics of the heating system, and generate a regulation strategy that takes into account both global optimization and local adaptability on this basis remains a challenging technical problem.

[0005] The shortcomings of existing technology: 1) Reliance on a single data source: Existing technologies rely solely on the operating data of the heat exchange station itself (such as supply and return water temperature, flow rate, and pressure), while ignoring key external factors that affect the heating effect. They cannot accurately judge changes in environmental heat dissipation and cannot directly understand the user's comfort needs and actual heat loss.

[0006] 2) Limited Model Predictive Capability: Existing load forecasting methods based on statistical regression, time series analysis, or simple physical models often struggle to accurately capture the complex nonlinear relationships between heating load and multiple factors such as weather, building characteristics, and user behavior. Prediction errors are significant, especially under special circumstances such as sudden weather changes, where prediction accuracy drops drastically.

[0007] 3) Lack of adaptive control strategies: Existing technologies still employ PID control, fuzzy control, or simple logic judgment based on preset rules. These strategies are often static and cannot be dynamically adjusted according to real-time changes in load, weather, and user needs. Summary of the Invention

[0008] To address the problems of existing technologies, such as reliance on a single data source, limited model prediction capabilities, and lack of adaptive control strategies, this invention aims to provide an artificial intelligence-based control method and system for heat exchange stations.

[0009] The technical solution adopted in this invention is as follows: An artificial intelligence-based control method for heat exchange stations includes the following steps: Collect real-time meteorological monitoring data, real-time room temperature monitoring data, and real-time heat exchange station monitoring data within the monitoring area, and construct a real-time dynamic topology map reflecting the heating situation in the monitoring area. The real-time dynamic topology map is input into the heating load prediction model built based on deep learning algorithm to predict the heating load and obtain the real-time heating load prediction result. The real-time heating load prediction results and real-time heat exchange station monitoring data are input into the control strategy generation model built based on reinforcement learning algorithm to generate control strategies and obtain the real-time heat exchange station control strategies. The real-time heat exchange station control strategy is sent to the heat exchange stations within the monitoring area, and the corresponding real-time heat exchange station control strategy is executed to control the heat exchange stations.

[0010] Furthermore, real-time meteorological monitoring data, real-time room temperature monitoring data, and real-time heat exchange station monitoring data are collected within the monitoring area, and a real-time dynamic topology map reflecting the heating situation in the monitoring area is constructed, including the following steps: Several micro weather stations were deployed within the monitoring area, room temperature sensors were installed inside several representative buildings within the monitoring area, and data acquisition terminals were installed at several heat exchange stations within the monitoring area. Using miniature weather stations, real-time meteorological monitoring data of the corresponding miniature weather stations is collected; using room temperature sensors, real-time room temperature monitoring data of the corresponding representative buildings is collected; using data acquisition terminals, real-time heat exchange station monitoring data of the corresponding heat exchange stations is collected and uploaded to the cloud data center. In the cloud data center, several real-time meteorological monitoring data, several real-time room temperature monitoring data and several real-time heat exchange station monitoring data are preprocessed to obtain several preprocessed real-time meteorological monitoring data, several preprocessed real-time room temperature monitoring data and several preprocessed real-time heat exchange station monitoring data. Based on several pre-processed real-time meteorological monitoring data, several pre-processed real-time room temperature monitoring data, and several pre-processed real-time heat exchange station monitoring data, a real-time dynamic topology map reflecting the heating situation in the monitoring area is constructed.

[0011] Furthermore, real-time meteorological monitoring data includes real-time meteorological indicator data and real-time meteorological station spatial location data; Real-time meteorological data includes outdoor temperature, relative humidity, wind speed, wind direction, atmospheric pressure, and solar radiation intensity at the corresponding location, collected by the micro weather station. Real-time room temperature monitoring data includes real-time room temperature index data and real-time room temperature spatial location data; Real-time room temperature data includes indoor temperature, indoor humidity, and building type of a representative building collected by room temperature sensors. Real-time heat exchange station monitoring data includes real-time heat exchange station index data and real-time heat exchange station spatial location data; Real-time heat exchange station indicator data includes the primary network supply and return water temperatures, secondary network supply and return water temperatures, flow rate, pressure, valve opening degree, and equipment status collected by the data acquisition terminal for the corresponding heat exchange station.

[0012] Furthermore, based on several pre-processed real-time meteorological monitoring data, several pre-processed real-time room temperature monitoring data, and several pre-processed real-time heat exchange station monitoring data, a real-time dynamic topology map reflecting the heating situation in the monitoring area is constructed, including the following steps: Based on several pre-processed real-time meteorological monitoring data, several pre-processed real-time room temperature monitoring data, and several pre-processed real-time heat exchange station monitoring data, several meteorological nodes, several room temperature monitoring nodes, and several heat exchange station nodes of the real-time dynamic topology are defined. Determine the edge relationship types of the real-time dynamic topology graph, construct the edges between several meteorological nodes, several room temperature monitoring nodes and several heat exchange station nodes based on the edge relationship types, and use quantification to determine the association strength of each edge. Based on several meteorological nodes, several room temperature monitoring nodes, and several heat exchange station nodes, combined with several corresponding edges and several correlation strengths, a real-time dynamic topology map reflecting the heating situation in the monitoring area is constructed.

[0013] Furthermore, edge relationship types include static edge relationship types and dynamic edge relationship types; Static edge relationship types include micro-weather station-micro-weather station, heat exchange station-heat exchange station, and representative building-representative building; Dynamic edge relationship types include micro-weather station-heat exchange station, weather station-representative building, and representative building-heat exchange station; Edges include static edges and dynamic edges; Static edges include meteorological-meteorological edges, heat exchange station-heat exchange station edges, and room temperature-room temperature edges; Dynamic edges include meteorological-heat exchange station edges, meteorological-room temperature edges, and room temperature-heat exchange station edges; Association strength includes static association strength and dynamic association strength; Static correlation strength includes meteorological-meteorological correlation strength, heat exchange station-heat exchange station correlation strength, and room temperature-room temperature correlation strength; The dynamic correlation strength includes the correlation strength between meteorology and heat exchange station, the correlation strength between meteorology and room temperature, and the correlation strength between room temperature and heat exchange station.

[0014] Furthermore, the heating load prediction model is constructed based on the GCN-GRU-MLP algorithm. The heating load prediction model includes a graph feature extraction module based on the GCN algorithm, a hidden state extraction module based on the GRU algorithm, and a heating load prediction module based on the MLP algorithm.

[0015] Furthermore, the real-time dynamic topology map is input into the heating load prediction model built based on a deep learning algorithm to predict the heating load and obtain the real-time heating load prediction results, including the following steps: Input the real-time dynamic topology map into the heating load prediction model built based on deep learning algorithms; The graph feature extraction module of the heating load prediction model is used to extract real-time graph features of the real-time dynamic topology graph and input them into the hidden state extraction module. The hidden state extraction module of the heating load prediction model is used to extract the real-time hidden state of the real-time graph features and input it into the heating load prediction module. Based on the real-time hidden state, the heating load prediction module of the heating load prediction model is used to predict the heating load and obtain the real-time heating load prediction results.

[0016] Furthermore, the regulation strategy generation model is constructed based on the HMARL algorithm, and the regulation strategy generation model includes a regional-level regulation strategy generation layer and an individual-level regulation strategy generation layer connected in sequence. The regional-level regulation strategy generation layer is set with a regional-level agent constructed based on the DQN algorithm and a first set of optimization objectives. The individual-level regulation strategy generation layer is set with several individual-level agents constructed based on the PPO algorithm and a corresponding second set of optimization objectives.

[0017] Furthermore, the real-time heating load forecast results and real-time heat exchange station monitoring data are input into the control strategy generation model constructed based on the reinforcement learning algorithm to generate the control strategy, thereby obtaining the real-time heat exchange station control strategy, including the following steps: The real-time heating load forecast results and real-time heat exchange station monitoring data are input into the regulation strategy generation model built based on reinforcement learning algorithm; Based on the real-time heating load forecast results and real-time heat exchange station monitoring data, the regional-level intelligent agent of the regional-level control strategy generation layer is used, combined with the first set of optimization objectives, to generate the control strategy and obtain the real-time regional-level heat exchange station control strategy. Based on the real-time heating load forecast results, real-time heat exchange station monitoring data, and real-time regional heat exchange station control strategies, several individual-level intelligent agents in the individual-level control strategy generation layer are used in combination with the corresponding second optimization target set to generate control strategies, thereby obtaining the real-time individual-level heat exchange station control strategy for each heat exchange station. By integrating real-time regional-level heat exchange station control strategies and several real-time individual-level heat exchange station control strategies, a real-time heat exchange station control strategy is obtained.

[0018] An artificial intelligence-based heat exchange station control system is provided to implement a heat exchange station control method. The system includes a dynamic topology graph construction unit, a heating load prediction unit, a control strategy generation unit, and a control strategy execution unit connected in sequence.

[0019] The beneficial effects of this invention are as follows: This invention provides an artificial intelligence-based heat exchange station control method and system. By collecting real-time data from multiple dimensions, including meteorological data, room temperature data, and heat exchange station operation data, and constructing a dynamic topology map, it can more comprehensively and accurately grasp the current state of the heating system and changes in the external environment. This overcomes the limitations of a single data source and provides a more reliable and refined data foundation for subsequent prediction and decision-making, making control more targeted. The invention employs a deep learning-based heating load prediction model, which can learn and capture the complex nonlinear relationships between multiple factors such as meteorological data, room temperature data, and heat exchange station status and the load. Compared to traditional models, it has higher prediction accuracy, especially performing better in dealing with complex situations such as sudden weather changes. It provides more reliable future load expectations, enabling control strategies to make more reasonable preparations and adjustments in advance, reducing supply-demand mismatch. The control strategy generation model, built based on reinforcement learning algorithms, can generate control strategies based on real-time prediction results and heat exchange station operation data. Data is used to generate dynamic control strategies that not only consider the current state but also adapt to future changes. These strategies are no longer rigid preset programs but dynamically optimized, highly adaptable, and high-quality policies that can more effectively balance heating performance and energy consumption. A hierarchical reinforcement learning strategy is employed, establishing regional and individual-level agents and considering different sets of optimization objectives. Regional-level agents are responsible for global coordination and optimization, while individual-level agents perform fine-tuning based on regional guidance and heat exchange station conditions. This achieves a good combination of global optimization and local adaptation, avoiding the drawbacks of simple unified control or completely decentralized control, and improving the collaborative efficiency and overall performance of heat exchange station control throughout the heating area. Through more accurate prediction and more optimized, adaptive, and coordinated control, energy waste can be significantly reduced, the overall operating cost of heating stations can be lowered, and heating quality can be improved, ensuring user comfort. Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of the heat exchange station control method based on artificial intelligence in this invention.

[0021] Figure 2 This is a structural block diagram of the heat exchange station control system based on artificial intelligence in this invention. Detailed Implementation

[0022] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1: like Figure 1 As shown, this embodiment provides a heat exchange station control method based on artificial intelligence, including the following steps: S1: Collect real-time meteorological monitoring data, real-time room temperature monitoring data, and real-time heat exchange station monitoring data within the monitoring area, and construct a real-time dynamic topology map reflecting the heating situation in the monitoring area, including the following steps: S1-1: Deploy several micro weather stations within the monitoring area, install room temperature sensors inside several representative buildings within the monitoring area, and install data acquisition terminals at several heat exchange stations within the monitoring area. S1-2: Use a micro weather station to collect real-time meteorological monitoring data of the corresponding micro weather station, use a room temperature sensor to collect real-time room temperature monitoring data of the corresponding representative building, use a data acquisition terminal to collect real-time heat exchange station monitoring data of the corresponding heat exchange station, and upload it to the cloud data center. Real-time meteorological monitoring data includes real-time meteorological indicator data and real-time meteorological station spatial location data; Real-time meteorological data includes outdoor temperature, relative humidity, wind speed, wind direction, atmospheric pressure, and solar radiation intensity at the corresponding location, collected by the micro weather station. Real-time room temperature monitoring data includes real-time room temperature index data and real-time room temperature spatial location data; Real-time room temperature data includes indoor temperature, building type, and user behavior patterns of representative buildings collected by room temperature sensors. Real-time heat exchange station monitoring data includes real-time heat exchange station index data and real-time heat exchange station spatial location data; Real-time heat exchange station indicator data includes the primary network supply and return water temperature, secondary network supply and return water temperature, flow rate, pressure, valve opening degree, and equipment status of the corresponding heat exchange station, collected by the data acquisition terminal. S1-3: In the cloud data center, preprocessing is performed on several real-time meteorological monitoring data, several real-time room temperature monitoring data and several real-time heat exchange station monitoring data to obtain several preprocessed real-time meteorological monitoring data, several preprocessed real-time room temperature monitoring data and several preprocessed real-time heat exchange station monitoring data. Preprocessing includes: Data cleaning: Detecting and handling missing values ​​(e.g., using interpolation, mean imputation, etc.) and outliers (e.g., using statistical methods or domain knowledge to identify, correct, or remove them); Data alignment: Align data from different sources and at different frequencies according to a unified timestamp, for example, to a time point every 5 minutes; for high-frequency data, the frequency can be reduced by averaging, maximizing, minimizing or interpolating; for low-frequency data, the frequency can be increased by repeating padding or interpolating. Data standardization / normalization: Standardize or normalize data of different dimensions (such as Z-Score standardization) to eliminate the influence of dimensions and facilitate subsequent model processing; S1-4: Based on several pre-processed real-time meteorological monitoring data, several pre-processed real-time room temperature monitoring data, and several pre-processed real-time heat exchange station monitoring data, construct a real-time dynamic topology map reflecting the heating situation in the monitoring area, including the following steps: S1-4-1: Based on several pre-processed real-time meteorological monitoring data, several pre-processed real-time room temperature monitoring data, and several pre-processed real-time heat exchange station monitoring data, define several meteorological nodes, several room temperature monitoring nodes, and several heat exchange station nodes in the real-time dynamic topology map. Heat exchange station nodes ;in, for t The first time the data was collected i Preprocessed real-time heat exchange station monitoring data corresponding to the node; t For time indication; i For node indication; For the first i Real-time spatial location data of the heat exchange station at the node; The primary water supply and return temperature; The supply and return water temperatures for the secondary network; For traffic; For pressure; Valve opening degree; Device status; Meteorological monitoring nodes ;in, for t The first time the data was collected i Preprocessed real-time meteorological monitoring data corresponding to the node; t For time indication; i For node indication; For the first i Real-time meteorological spatial location data of nodes; Outdoor temperature; Relative humidity; Wind speed; Wind direction; Atmospheric pressure; Solar radiation intensity; Room temperature monitoring nodes ;in, for t The first time the data was collected i Preprocessed real-time room temperature monitoring data corresponding to the node; t For time indication; i For node indication; For the first i Real-time room temperature spatial location data of the nodes; Indoor temperature; Building type; User behavior patterns; S1-4-2: Determine the edge relationship types of the real-time dynamic topology graph, construct the edges between several meteorological nodes, several room temperature monitoring nodes and several heat exchange station nodes based on the edge relationship types, and use quantification to determine the association strength of each edge. Edge relationship types include static edge relationship types and dynamic edge relationship types; Static edge relationship types include micro-weather station-micro-weather station, heat exchange station-heat exchange station, and representative building-representative building; Dynamic edge relationship types include micro-weather station-heat exchange station, weather station-representative building, and representative building-heat exchange station; Edges include static edges and dynamic edges; Static edges include meteorological-meteorological edges, heat exchange station-heat exchange station edges, and room temperature-room temperature edges; Dynamic edges include meteorological-heat exchange station edges, meteorological-room temperature edges, and room temperature-heat exchange station edges; Association strength includes static association strength and dynamic association strength; Static correlation strength includes meteorological-meteorological correlation strength, heat exchange station-heat exchange station correlation strength, and room temperature-room temperature correlation strength; Meteorological-Meteorological Correlation Strength ;in, To be based on different micro weather stations and The obtained meteorological-meteorological spatial distance; This is the meteorological-meteorological spatial distance coefficient. If the meteorological-meteorological spatial distance exceeds the meteorological-meteorological spatial distance threshold, then... =0, at other times =1; Heat exchange station-heat exchange station correlation strength ;in, To be based on different heat exchange stations and The obtained spatial distance between heat exchange stations; This is the spatial distance coefficient between heat exchange stations. If the spatial distance between heat exchange stations exceeds the spatial distance threshold, then... =0, at other times =1; Room temperature-room temperature correlation strength ;in, To be based on different representative buildings and The obtained room temperature-room temperature spatial distance; This is the room temperature-room temperature spatial distance coefficient. If the room temperature-room temperature spatial distance exceeds the room temperature-room temperature spatial distance threshold, then... =0, at other times =1; Dynamic correlation strength includes meteorological-heat exchange station correlation strength, meteorological-room temperature correlation strength, and room temperature-heat exchange station correlation strength; Meteorological-Heat Exchange Station Correlation Strength ;in, The function for obtaining the correlation strength between meteorology and heat exchange stations can be a physical model function (such as a physical heat conduction model) or a data-driven model function (such as linear regression or neural networks). This is the coefficient for the influence of heat loss; According to and The obtained spatial distance between the meteorological and heat exchange station; Weather-room temperature correlation strength ;in, The function for obtaining the correlation strength between meteorology and room temperature can be a physical model function (such as a physical heat transfer model) or a data-driven model function (such as linear regression or neural networks). for Building type Corresponding thermal inertia; According to and The obtained meteorological-room temperature spatial distance; Room temperature-heat exchange station correlation strength ;in, This is the function for obtaining the correlation strength between room temperature and heat exchange station, and the data-driven model function (such as linear regression, neural network). According to and The obtained room temperature - spatial distance of the heat exchange station; The coefficient representing the room temperature-heat exchange station service relationship; The data-driven model used above is a pre-trained association strength prediction model. By learning from a large number of samples with association strength labels, it can achieve the function of predicting association strength based on the input relevant data. The formula for updating the dynamic association strength is:

[0024] In the formula, , for t , t The node at time -1 i To the node j The updated dynamic correlation strengths include meteorological-heat exchange station correlation strength, meteorological-room temperature correlation strength, and room temperature-heat exchange station correlation strength; j For node indication; It is the attenuation factor; The functions for obtaining correlation strength include the meteorological-heat exchange station correlation strength function, the meteorological-room temperature correlation strength function, and the room temperature-heat exchange station correlation strength function. for t The first time the data was collected i, j Preprocessed monitoring data corresponding to the node; S1-4-3: Based on several meteorological nodes, several room temperature monitoring nodes, and several heat exchange station nodes, combined with several corresponding edges and several association strengths, a real-time dynamic topology map reflecting the heating situation in the monitoring area is constructed. The formula is:

[0025] In the formula, for t Real-time dynamic topology graph at any given moment; A set of nodes, including , , ; for t The set of edges at each moment; for t The preprocessed monitoring data matrix of nodes at each time point, including , , ; for t The correlation strength matrix at time step (i.e., time step), including all static and dynamic correlation strengths, is represented as an adjacency matrix. form; S2: Input the real-time dynamic topology map into the heating load prediction model built based on deep learning algorithm to predict the heating load and obtain the real-time heating load prediction results, including the following steps: S2-1: Input the real-time dynamic topology map into the heating load prediction model built based on deep learning algorithm in the cloud data center; The heating load prediction model is built based on the Graph Convolutional Network (GCN) - Gated Recurrent Unit (GRU) - Multi-Layer Perceptron (MLP) algorithm. The heating load prediction model includes a graph feature extraction module based on the GCN algorithm, a hidden state extraction module based on the GRU algorithm, and a heating load prediction module based on the MLP algorithm. S2-2: Using the graph feature extraction module of the heating load prediction model, extract the real-time graph features of the real-time dynamic topology graph and input them into the hidden state extraction module, including the following steps: S2-2-1: Association strength matrix of real-time dynamic topology graph adjacency matrix The form is normalized to obtain the normalized adjacency matrix; The formula is:

[0026] In the formula, for t The normalized adjacency matrix at time step; for t The augmented adjacency matrix at time step; for N× N The identity matrix; N The total number of nodes; for The degree matrix; S2-2-2: Based on the normalized adjacency matrix The graph feature extraction module of the heating load prediction model is used to extract the preprocessed monitoring data matrix of the real-time dynamic topology map. Real-time node characteristics Then, perform multi-layer graph convolution to obtain the final node embedding representation; The formula is:

[0027] In the formula, , for t The first moment l +1、 l Real-time node features of convolutional layers; For activation functions; For the first l The weight matrix of each convolutional layer; l This is an indicator of the number of convolutional layers. go through L After multiple convolutional layers, the final node embedding representation is obtained. These vectors capture the spatiotemporal context information of nodes in the graph structure; S2-2-3: Final embedding representation of all nodes Perform graph-level aggregation features (such as averaging or max pooling the embeddings of all nodes) to obtain real-time graph features of the real-time dynamic topology graph, and input them into the hidden state extraction module. The formula is:

[0028] In the formula, for t Real-time graph features at any given moment; It is an aggregate function; S2-3: Use the hidden state extraction module of the heating load prediction model to extract the real-time hidden state of the real-time graph features and input it into the heating load prediction module. The formula is:

[0029] In the formula, for t、t- The real-time hidden state and historical hidden state at a given moment; These are internal variables of the GRU; For GRU weights and biases; For activation functions; The product of Hadamard; S2-4: Based on the real-time hidden state, use the heating load prediction module of the heating load prediction model to predict the heating load and obtain the real-time heating load prediction result. The formula is:

[0030] In the formula, The predicted future heating load vector, i.e. the real-time heating load prediction result, includes the predicted total regional heating load and the predicted sub-heating loads divided by heat exchange stations; For output layer weights and biases; S3: Input the real-time heating load forecast results and real-time heat exchange station monitoring data into the control strategy generation model based on reinforcement learning algorithm in the cloud data center to generate the control strategy and obtain the real-time heat exchange station control strategy, including the following steps: S3-1: Input the real-time heating load prediction results and real-time heat exchange station monitoring data into the regulation strategy generation model constructed based on reinforcement learning algorithm; The regulation strategy generation model is constructed based on the Hierarchical Multi-Agent Reinforcement Learning (HMARL) algorithm. The regulation strategy generation model includes a regional-level regulation strategy generation layer and an individual-level regulation strategy generation layer connected in sequence. The regional-level regulation strategy generation layer is set with a regional-level agent constructed based on the Deep Q-Network (DQN) algorithm and a first set of optimization objectives. The individual-level regulation strategy generation layer is set with several individual-level agents constructed based on the Proximal Policy Optimization (PPO) algorithm and a corresponding second set of optimization objectives. Each individual-level agent corresponds to a heat exchange station. S3-2: Based on the real-time heating load forecast results and real-time heat exchange station monitoring data, the regional-level intelligent agent of the regional-level control strategy generation layer is used, combined with the first set of optimization objectives, to generate the control strategy and obtain the real-time regional-level heat exchange station control strategy. The state space of the regional intelligent agent includes the predicted total regional heating load in the real-time heating load forecast results, the average supply water temperature, average return water temperature, and total flow rate of all heat exchange stations obtained from real-time heat exchange station monitoring data. The action space of a regional intelligent agent, with discrete or continuous actions: Discrete: maintain the current strategy, increase the overall water supply temperature setting, decrease the overall water supply temperature setting, start / stop the regional auxiliary heat source; Continuous: set a regional water supply temperature baseline, or set a regional power distribution coefficient. The first set of optimization objectives is: to minimize the total energy consumption of the region, to minimize the deviation between the average room temperature of the region and the set value, and to maximize the uniformity of heating in the region. S3-3: Based on the real-time heating load forecast results, real-time heat exchange station monitoring data, and real-time regional heat exchange station control strategies, several individual-level intelligent agents in the individual-level control strategy generation layer are used in combination with the corresponding second optimization target set to generate control strategies, thereby obtaining the real-time individual-level heat exchange station control strategy for each heat exchange station. The state space of an individual-level intelligent agent includes the predicted sub-heating load of the current heat exchange station in the real-time heating load prediction results, the real-time regional heat exchange station control strategy, and the real-time heat exchange station monitoring data of the current heat exchange station. The action space of an individual-level intelligent agent is usually a continuous action, representing specific adjustment decisions: water supply temperature adjustment action, valve opening adjustment action, circulating pump frequency adjustment action, setting the target value of the water supply temperature of this station; The second set of optimization objectives is: to minimize the energy consumption of this heat exchange station, to maintain the supply and return water temperature difference of this heat exchange station within a reasonable range, and to minimize the deviation between the room temperature and the set value in the service area of ​​this heat exchange station. S3-4: Integrate the real-time regional heat exchange station control strategy and several real-time individual heat exchange station control strategies to obtain the real-time heat exchange station control strategy. S4: Send the real-time heat exchange station control strategy to the heat exchange stations within the monitoring area, execute the corresponding real-time heat exchange station control strategy, and control the heat exchange stations, including the following steps: S4-1: Send the real-time heat exchange station control strategy to all heat exchange stations within the monitoring area. Based on the real-time regional heat exchange station control strategy and the real-time individual heat exchange station control strategy of the current heat exchange station, generate the real-time control command (such as pulse width modulation (PWM) signal and frequency setpoint) for the current heat exchange station. S4-2: Sends real-time control commands to the actuators (valves, pumps, etc.) of the heat exchange station via fieldbus or industrial Ethernet. S4-3: The actuator regulates the heat exchange station according to real-time control commands.

[0031] Example 2: like Figure 2 As shown, this embodiment provides an artificial intelligence-based heat exchange station control system for implementing heat exchange station control methods. The system includes a dynamic topology graph construction unit, a heating load prediction unit, a control strategy generation unit, and a control strategy execution unit connected in sequence.

[0032] The dynamic topology map construction unit is used to collect real-time meteorological monitoring data, real-time room temperature monitoring data, and real-time heat exchange station monitoring data within the monitoring area, and to construct a real-time dynamic topology map reflecting the heating situation of the monitoring area. The heating load prediction unit is used to input the real-time dynamic topology map into the heating load prediction model built based on deep learning algorithm, perform heating load prediction, and obtain real-time heating load prediction results. The regulation strategy generation unit is used to input the real-time heating load prediction results and real-time heat exchange station monitoring data into the regulation strategy generation model built based on reinforcement learning algorithm, generate regulation strategies, and obtain real-time heat exchange station regulation strategies. The control strategy execution unit is used to send the real-time heat exchange station control strategy to the heat exchange stations within the monitoring area, execute the corresponding real-time heat exchange station control strategy, and control the heat exchange stations.

[0033] This invention provides an artificial intelligence-based heat exchange station control method and system. By collecting real-time data from multiple dimensions, including meteorological data, room temperature data, and heat exchange station operation data, and constructing a dynamic topology map, it can more comprehensively and accurately grasp the current state of the heating system and changes in the external environment. This overcomes the limitations of a single data source and provides a more reliable and refined data foundation for subsequent prediction and decision-making, making control more targeted. The invention employs a deep learning-based heating load prediction model, which can learn and capture the complex nonlinear relationships between multiple factors such as meteorological data, room temperature data, and heat exchange station status and the load. Compared to traditional models, it has higher prediction accuracy, especially performing better in dealing with complex situations such as sudden weather changes. It provides more reliable future load expectations, enabling control strategies to make more reasonable preparations and adjustments in advance, reducing supply-demand mismatch. The control strategy generation model, built based on reinforcement learning algorithms, can generate control strategies based on real-time prediction results and heat exchange station operation data. Data is used to generate dynamic control strategies that not only consider the current state but also adapt to future changes. These strategies are no longer rigid preset programs but dynamically optimized, highly adaptable, and high-quality policies that can more effectively balance heating performance and energy consumption. A hierarchical reinforcement learning strategy is employed, establishing regional and individual-level agents and considering different sets of optimization objectives. Regional-level agents are responsible for global coordination and optimization, while individual-level agents perform fine-tuning based on regional guidance and heat exchange station conditions. This achieves a good combination of global optimization and local adaptation, avoiding the drawbacks of simple unified control or completely decentralized control, and improving the collaborative efficiency and overall performance of heat exchange station control throughout the heating area. Through more accurate prediction and more optimized, adaptive, and coordinated control, energy waste can be significantly reduced, the overall operating cost of heating stations can be lowered, and heating quality can be improved, ensuring user comfort.

[0034] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. An artificial intelligence-based heat exchange station regulation method, characterized in that: The method comprises the following steps: Collecting real-time meteorological monitoring data, real-time room temperature monitoring data and real-time heat exchange station monitoring data in the monitoring area, and constructing a real-time dynamic topology graph reflecting the heating situation of the monitoring area; Inputting the real-time dynamic topology graph into a heating load prediction model constructed based on a deep learning algorithm to perform heating load prediction and obtain real-time heating load prediction results; Inputting the real-time heating load prediction results and real-time heat exchange station monitoring data into a regulation strategy generation model constructed based on a reinforcement learning algorithm to perform regulation strategy generation and obtain real-time heat exchange station regulation strategies; Sending the real-time heat exchange station regulation strategies to the heat exchange stations in the monitoring area to execute corresponding real-time heat exchange station regulation strategies and regulate the heat exchange stations.

2. The heat exchange station regulation method based on artificial intelligence according to claim 1, characterized in that: Collecting real-time meteorological monitoring data, real-time room temperature monitoring data and real-time heat exchange station monitoring data in the monitoring area, and constructing a real-time dynamic topology graph reflecting the heating situation of the monitoring area, comprising the following steps: Deploying a plurality of micro weather stations inside the monitoring area, deploying room temperature sensors inside a plurality of representative buildings in the monitoring area, and installing data acquisition terminals in a plurality of heat exchange stations in the monitoring area; Using the micro weather stations to collect real-time meteorological monitoring data of the corresponding micro weather stations, using the room temperature sensors to collect real-time room temperature monitoring data of the corresponding representative buildings, using the data acquisition terminals to collect real-time heat exchange station monitoring data of the corresponding heat exchange stations, and uploading to a cloud data center; In the cloud data center, pre-processing a plurality of real-time meteorological monitoring data, a plurality of real-time room temperature monitoring data and a plurality of real-time heat exchange station monitoring data to obtain a plurality of pre-processed real-time meteorological monitoring data, a plurality of pre-processed real-time room temperature monitoring data and a plurality of pre-processed real-time heat exchange station monitoring data; According to a plurality of pre-processed real-time meteorological monitoring data, a plurality of pre-processed real-time room temperature monitoring data and a plurality of pre-processed real-time heat exchange station monitoring data, a real-time dynamic topology graph reflecting the heating situation of the monitoring area is constructed.

3. The heat exchange station regulation method based on artificial intelligence according to claim 2, characterized in that: The real-time meteorological monitoring data includes real-time meteorological index data and real-time meteorological station spatial position data; The real-time meteorological index data includes outdoor temperature, relative humidity, wind speed, wind direction, atmospheric pressure and solar radiation intensity collected by the micro weather station at the corresponding position; The real-time room temperature monitoring data includes real-time room temperature index data and real-time room temperature spatial position data; The real-time room temperature index data includes indoor temperature, indoor humidity and building type of the corresponding representative building collected by the room temperature sensor; The real-time heat exchange station monitoring data includes real-time heat exchange station index data and real-time heat exchange station spatial position data; The real-time heat exchange station index data includes primary network supply and return water temperature, secondary network supply and return water temperature, flow, pressure, valve opening and equipment state of the corresponding heat exchange station collected by the data acquisition terminal.

4. The heat exchange station regulation method based on artificial intelligence according to claim 3, characterized in that: According to a plurality of pre-processed real-time meteorological monitoring data, a plurality of pre-processed real-time room temperature monitoring data and a plurality of pre-processed real-time heat exchange station monitoring data, a real-time dynamic topology graph reflecting the heating situation of the monitoring area is constructed, comprising the following steps: According to a plurality of pre-processed real-time meteorological monitoring data, a plurality of pre-processed real-time room temperature monitoring data and a plurality of pre-processed real-time heat exchange station monitoring data, a plurality of meteorological nodes, a plurality of room temperature monitoring nodes and a plurality of heat exchange station nodes of a real-time dynamic topology graph are defined; A plurality of edge relationship types of the real-time dynamic topology graph are determined, and edges between the plurality of meteorological nodes, the plurality of room temperature monitoring nodes and the plurality of heat exchange station nodes are constructed according to the plurality of edge relationship types, and the correlation strength of each edge is quantified; According to the plurality of meteorological nodes, the plurality of room temperature monitoring nodes and the plurality of heat exchange station nodes, the corresponding plurality of edges and the plurality of correlation strengths are combined to construct a real-time dynamic topology graph reflecting the heating condition of the monitoring area.

5. The heat exchange station regulation method based on artificial intelligence according to claim 4, characterized in that: The edge relationship type includes a static edge relationship type and a dynamic edge relationship type; The static edge relationship type includes a micro weather station-micro weather station, a heat exchange station-heat exchange station and a representative building-representative building; The dynamic edge relationship type includes a micro weather station-heat exchange station, a weather station-representative building and a representative building-heat exchange station; The edge includes a static edge and a dynamic edge; The static edge includes a meteorological-meteorological edge, a heat exchange station-heat exchange station edge and a room temperature-room temperature edge; The dynamic edge includes a meteorological-heat exchange station edge, a meteorological-room temperature edge and a room temperature-heat exchange station edge; The correlation strength includes a static correlation strength and a dynamic correlation strength; The static correlation strength includes a meteorological-meteorological correlation strength, a heat exchange station-heat exchange station correlation strength and a room temperature-room temperature correlation strength; The dynamic correlation strength includes a meteorological-heat exchange station correlation strength, a meteorological-room temperature correlation strength and a room temperature-heat exchange station correlation strength.

6. The heat exchange station regulation method based on artificial intelligence according to claim 5, characterized in that: The heating load prediction model is constructed based on the GCN-GRU-MLP algorithm, and the heating load prediction model includes a graph feature extraction module constructed based on the GCN algorithm, a hidden state extraction module constructed based on the GRU algorithm and a heating load prediction module constructed based on the MLP algorithm.

7. The heat exchange station regulation method based on artificial intelligence according to claim 6, characterized in that: The real-time dynamic topology graph is input into the heating load prediction model constructed based on the deep learning algorithm to perform heating load prediction and obtain a real-time heating load prediction result, including the following steps: The real-time dynamic topology graph is input into the heating load prediction model constructed based on the deep learning algorithm; The real-time graph features of the real-time dynamic topology graph are extracted using the graph feature extraction module of the heating load prediction model and input into the hidden state extraction module; The real-time hidden state of the real-time graph features is extracted using the hidden state extraction module of the heating load prediction model and input into the heating load prediction module; According to the real-time hidden state, the heating load prediction module of the heating load prediction model is used to perform heating load prediction to obtain a real-time heating load prediction result.

8. The heat exchange station regulation method based on artificial intelligence according to claim 7, characterized in that: The regulation strategy generation model is constructed based on the HMARL algorithm, and comprises a region-level regulation strategy generation layer and an individual-level regulation strategy generation layer connected in sequence.

9. The heat exchange station regulation method based on artificial intelligence according to claim 8, characterized in that: The real-time heating load prediction result and the real-time heat exchange station monitoring data are input into the regulation strategy generation model constructed based on the reinforcement learning algorithm to generate a regulation strategy, and a real-time heat exchange station regulation strategy is obtained, including the following steps: The real-time heating load prediction result and the real-time heat exchange station monitoring data are input into the regulation strategy generation model constructed based on the reinforcement learning algorithm; According to the real-time heating load prediction result and the real-time heat exchange station monitoring data, the region-level agent of the region-level regulation strategy generation layer is used to generate a regulation strategy in combination with the first optimization target set, and a real-time region-level heat exchange station regulation strategy is obtained; According to the real-time heating load prediction result, the real-time heat exchange station monitoring data and the real-time region-level heat exchange station regulation strategy, the individual-level agents of the individual-level regulation strategy generation layer are used to generate a regulation strategy in combination with the corresponding second optimization target set, and a real-time individual-level heat exchange station regulation strategy of each heat exchange station is obtained; The real-time region-level heat exchange station regulation strategy and the real-time individual-level heat exchange station regulation strategies are integrated to obtain a real-time heat exchange station regulation strategy.

10. An artificial intelligence-based heat exchange station regulation system for implementing the heat exchange station regulation method according to any one of claims 1-9, characterized in that: The system comprises a dynamic topology graph construction unit, a heating load prediction unit, a regulation strategy generation unit and a regulation strategy execution unit connected in sequence.

Images