Heating energy-saving building, heat exchange station and boiler room linkage control system

Abstract

The invention discloses a linkage control system for a heating energy-saving building, a heat exchange station and a boiler room, and belongs to the technical field of heating control. Comprising an acquisition and integration module, a map construction module, a demand prediction module, a multi-criterion optimization module, a flow regulation and control module, a strategy generation module, a decision enhancement module, a dynamic adjustment module, a feedback adjustment module and a monitoring and maintenance module. The energy loss in the heat transfer process can be reduced, the overall system efficiency is improved, the flexibility and the response speed of the system are improved, resource waste is avoided, the energy consumption is reduced, the low-carbon operation target of the heating system is facilitated, the innovativeness and the global optimization capability of a linkage control strategy are improved, and the energy consumption is reduced. The method is suitable for high-complexity and high-dimension system optimization, the performability and the actual landing performance of the strategy are ensured, and the sensitivity of the system to external environment fluctuation is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of heating control, and in particular to a heating energy-saving building, a heat exchange station and a boiler room linkage control system. Background Art

[0002] The heating system is an important part of modern urban infrastructure and is widely used in residential, commercial buildings and industrial facilities. As an important part of energy consumption, the efficiency and energy-saving level of the heating system directly affect the city's carbon emissions and the quality of life of residents. With the intensification of global warming and the energy crisis, how to minimize energy consumption while ensuring heating needs has become an important research direction in the field of heating. In recent years, with the development of the Internet of Things, big data and artificial intelligence technologies, the research on intelligent heating control systems has made breakthrough progress. However, existing methods generally have problems such as decentralized system modules, insufficient synergy and unsatisfactory dynamic optimization effects. In particular, in the linkage control of buildings, heat exchange stations and boiler rooms, a unified comprehensive optimization framework has not yet been formed. With the deepening of smart city construction and energy transformation, heating energy saving has become a key issue that needs to be solved urgently.

[0003] The existing linkage control system for heating energy-saving buildings, heat exchange stations and boiler rooms has high energy loss in the heat transfer process, which reduces the overall system efficiency, and the system's flexibility and response speed are poor, which easily leads to waste of resources; in addition, the existing linkage control strategy for heating energy-saving buildings, heat exchange stations and boiler rooms has low innovation and global optimization capabilities, and cannot be applied to high-complexity and high-dimensional system optimization, and cannot guarantee the feasibility and practical implementation of the strategy. Therefore, we propose a linkage control system for heating energy-saving buildings, heat exchange stations and boiler rooms. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a heating energy-saving building, heat exchange station and boiler room linkage control system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A heating energy-saving building, heat exchange station and boiler room linkage control system, including a collection and integration module, a map construction module, a demand prediction module, a multi-criteria optimization module, a flow control module, a strategy generation module, a decision reinforcement module, a dynamic adjustment module, a feedback regulation module and a monitoring and maintenance module;

[0007] The acquisition and integration module is used to collect multi-source data from different sources and integrate them in real time to build a dynamic view of the global status;

[0008] The graph construction module is used to construct an entity relationship network including buildings, heat exchange stations and boiler rooms;

[0009] The demand forecasting module is used to forecast future heating demand based on multiple groups of dynamic factors;

[0010] The multi-criteria optimization module is used to output a preliminary heating resource allocation plan;

[0011] The flow control module is used to control and optimize the flow of circulating water between buildings, heat exchange stations and boiler rooms;

[0012] The strategy generation module is used to integrate demand forecasting and water cycle optimization results to generate a linkage control strategy;

[0013] The decision-making reinforcement module is used to simulate and evaluate multiple linkage control strategies;

[0014] The dynamic adjustment module is used to handle real-time changes in heating demand and external environmental fluctuations, and dynamically optimize the control strategy;

[0015] The feedback adjustment module is used for users to view the current heating status and historical energy consumption, and adjust the heating parameters according to user feedback;

[0016] The monitoring and maintenance module is used to perform real-time status monitoring of boilers, heat exchange stations and building heating equipment and generate energy efficiency evaluation reports.

[0017] As a further solution of the present invention, the acquisition and integration module constructs a dynamic view of the global state in the following specific steps:

[0018] S1.1: Collect environmental data such as temperature and humidity inside the building and external meteorological conditions, as well as water temperature and pressure of the heat exchange station and fuel consumption and emission data of the boiler room. Then calculate the mean of each type of data, identify outliers in each type of data based on the 3σ principle, and process outliers and missing values ​​in each type of data through linear interpolation.

[0019] S1.2: Normalize data from different sources, align data from different devices and regions according to the timestamps of each type of data, and then integrate various types of data from different data sources based on the weighted average model to form a unified global dynamic view. Then, based on the various types of data collected in real time from buildings, heat exchange stations, and boiler rooms, summarize the status of buildings, heat exchange stations, and boiler rooms to update the global dynamic view in real time.

[0020] As a further solution of the present invention, the graph construction module constructs an entity relationship network including buildings, heat exchange stations and boiler rooms in the following specific steps:

[0021] S2.1: Based on the constructed global dynamic view, detailed data from different devices, regions, and data sources are collected, each group of data is classified according to business needs, and entity information in each type of data is identified through NLP technology, and each group of identified entities is used as a node;

[0022] S2.2: According to the device operation mechanism and sensor data, obtain the physical and logical relationships between entities, use the physical and logical relationships between each group of entities as edges, connect the corresponding nodes through the generated edges, and assign corresponding attributes to each entity according to the device operation mechanism and sensor data to generate an entity relationship network;

[0023] S2.3: Based on the entity relationship network, inference rules are generated through semantic inference rules. When real-time data flows in, the entity relationship network uses real-time data to dynamically update the entity attributes and relationships in the network.

[0024] As a further solution of the present invention, the demand forecasting module predicts future heating demand based on multiple groups of dynamic factors in the following specific steps:

[0025] S3.1: Extract historical heating data from the entity relationship network, and preprocess each set of heating data into multi-dimensional time series data to form the corresponding input data. Then, standardize the input data and divide it into training set, validation set and test set, and create a set of demand forecasting models based on CNN architecture;

[0026] S3.2: According to the set training batches, the training set data is input into the demand forecasting model in sequence. The model input layer receives the formatted training set data and extracts the feature data in the time dimension through the one-dimensional convolution kernel in the convolution layer. The pooling layer downsamples the extracted feature data to reduce the feature dimension and reduce the computational complexity.

[0027] S3.3: The training data is repeatedly convolved through multiple sets of convolutional layers and pooling layers for pooling operations, and the final feature data is input into a fully connected layer. The fully connected layer then expands the extracted feature data into a one-dimensional vector and outputs the final predicted heat demand through an output layer;

[0028] S3.4: Use MSE as the loss function to calculate the deviation between the predicted value and the true value, and input the deviation value from the model output layer according to the chain rule, and calculate the gradient of the deviation value for each layer of the model, and then use the Adam algorithm to update the model parameters, calculate the evaluation index through the validation set and the test set, and judge the model performance. If the performance of the demand forecasting model does not meet the preset requirements, retrain until the performance meets the requirements and stop training;

[0029] S3.5: The latest heating data collected in real time is input into the demand forecasting model, and the demand forecasting model outputs the heat demand for multiple time steps in the future through forward propagation.

[0030] As a further solution of the present invention, the flow control module controls and optimizes the circulation water flow between the building, the heat exchange station and the boiler room. The specific steps are as follows:

[0031] S4.1: Collect the water flow, water temperature, pressure and pump performance parameters of the circulating water flow system between the building, heat exchange station and boiler room from the entity relationship network, and form a hydraulic network topology based on the connection relationship between the building, heat exchange station and boiler room, initialize the flow and path resistance of the circulation path, obtain multiple groups of flow plans by random generation, and form a group of populations with the generated plans, and generate multiple groups of water droplets at the same time, where the position of each water droplet represents a flow plan;

[0032] S4.2: Randomly generate the positions of multiple groups of water drops, set the fitness function based on the goal of minimizing the total energy consumption of the circulating water system, and then calculate the fitness value of each group of water drops corresponding to the flow scheme through the fitness function, and sort the population according to the fitness value. The water drop with the highest fitness is the main river, and the remaining water drops are tributaries, and their initial flow is allocated according to the tributary fitness value;

[0033] S4.3: Based on the fitness values ​​of the main river and its tributaries, the distance between the main river and the tributaries is calculated, two random factors are generated in the range of [0, 1], and based on the generated random factors, the characteristic values ​​and positions of the tributaries are continuously adjusted by simulating the linear update of sample characteristics, so that the tributaries are constantly close to the main river to form new tributaries;

[0034] S4.4: Based on the average or median of the current tributary fitness value, set the fitness threshold, calculate the fitness value of each current tributary through the objective function, and compare the calculated fitness value with the threshold value respectively. If it is less than the threshold, the corresponding sample will be eliminated and will no longer participate in the subsequent optimization process. A coefficient greater than 1 is selected as the precipitation factor by random selection. Then, the precipitation factor is amplified for tributaries greater than or equal to the preset threshold to increase the weight of high fitness samples in each test sample;

[0035] S4.5: After each round of iteration, the fitness of all main rivers and tributaries is recalculated according to the fitness function, and the main rivers and tributaries are updated. Then, the position update and main river update operations are performed again until the maximum number of iterations is reached, or the increment of the fitness of each water droplet is less than the set convergence threshold, or the screened water droplets meet the preset quality requirements. Then, the iterative update is stopped, and the water droplets with the highest fitness value are screened as the optimal flow plan, and the circulating water flow is regulated based on the optimal flow plan.

[0036] As a further solution of the present invention, the specific steps of simulation evaluation of the decision reinforcement module are as follows:

[0037] S5.1: Collect the current building temperature status, heat exchange station status, boiler room status and environmental status to construct the corresponding state space, and create the corresponding action space according to the control strategy that can be taken in each state in the state space, and use the current system operation status as the root node of the search tree;

[0038] S5.2: For the root node state, according to the available actions in the action space, through the dynamic model of the heating system, calculate the new state information that the current state transfers to after executing the random action. For each state after the transfer, add a new node to the search tree, and evaluate the immediate reward of taking the action from the current state to the new state, and accumulate the initial value of the reward as the immediate reward;

[0039] S5.3: Starting from the root node, the optimal node with the highest current UCB value is gradually selected along the constructed search tree through the UCB strategy. When the selected node has not been fully expanded, an unexecuted action is selected based on the current state to generate a new node. According to the control strategy and system model, any random action is executed from the current state, and new state information is generated, which is added to the search tree as a new node;

[0040] S5.4: Starting from the expanded new node, execute each action of the corresponding path in a random or heuristic manner until the simulation or heating stable state of the predetermined time period is reached, and calculate the cumulative rewards of all states experienced during the simulation process, and backtrack the cumulative rewards obtained from the simulation to all nodes involved in the tree;

[0041] S5.5: Repeat the selection, expansion, simulation and backtracking steps multiple times until the reward value change of each node converges to the preset threshold, collect all the linkage control strategies in the search tree, and select the path with the highest cumulative reward as the current optimal strategy, and apply it to the actual linkage control system.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention collects water flow, water temperature, pressure and various parameters of water pump performance of the circulating water flow system between buildings, heat exchange stations and boiler rooms from the entity relationship network, and forms a hydraulic network topology according to the connection relationship between buildings, heat exchange stations and boiler rooms, initializes the flow and path resistance of the circulation path, obtains multiple groups of flow schemes by random generation, and forms a group of populations with the generated schemes, and generates multiple groups of water droplets at the same time, wherein the position of each water droplet represents a flow scheme, and randomly generates the positions of multiple groups of water droplets, and sets the fitness function based on the goal of minimizing the total energy consumption of the circulating water system, and then calculates the fitness value of each group of water droplets corresponding to the flow scheme through the fitness function, and takes the water droplet with the highest fitness as the main river, and the remaining water droplets as tributaries, and allocates their initial flow according to the tributary fitness value, and calculates the distance between the main river and the tributary based on the fitness value of the main river and the tributary, and continuously adjusts the characteristic value and the tributary position of the tributary to form a new tributary, and based on the average value of the current tributary fitness value Or median, set the fitness threshold, and eliminate tributaries with fitness values ​​less than the threshold. A coefficient greater than 1 is selected as the precipitation factor by random selection. Then, the precipitation factor is amplified for tributaries greater than or equal to the preset threshold to increase the weight of high fitness samples in each test sample. After each round of iteration, the fitness of all main rivers and tributaries is recalculated according to the fitness function, and the main river and tributaries are updated. Then, the position update and main river update operations are performed again until the maximum number of iterations is reached, or the increment of the fitness of each water droplet is less than the set convergence threshold, or the screened water droplets meet the preset quality requirements, then the iterative update is stopped, and the water droplets with the highest fitness value are screened as the optimal flow scheme. Based on the optimal flow scheme, the circulating water flow is regulated, which can reduce the energy loss in the heat transfer process, improve the overall system efficiency, enhance the flexibility and response speed of the system, avoid waste of resources, reduce energy consumption, and help the heating system achieve the low-carbon operation goal.

[0044] 2. The present invention collects the current building temperature state, heat exchange station state, boiler room state and environmental state to construct a corresponding state space, and creates a corresponding action space according to the control strategy that can be taken in each state in the state space. The current system operation state is used as the root node of the search tree. Starting from the root node, the optimal node with the highest current UCB value is gradually selected along the constructed search tree through the UCB strategy. When the selected node has not been fully expanded, an unexecuted action is selected based on the current state to generate a new node. According to the control strategy and system model, any random action is executed from the current state, and new state information is generated. The state information is added as a new node to the search tree. Starting from the expanded new node, a random or heuristic action is used to generate a new node. The various actions of the corresponding path are executed in a simulated manner until the simulation or heating stable state is reached in the predetermined time period, and the cumulative rewards of all states experienced during the simulation are calculated. The cumulative rewards obtained from the simulation are back-tracked to all nodes involved in the tree, and the selection, expansion, simulation and backtracking steps are repeated many times until the reward value change of each node converges to the preset threshold. All linkage control strategies in the search tree are collected, and the path with the highest cumulative reward is selected as the current optimal strategy, which is applied to the actual linkage control system to improve the innovation and global optimization ability of the linkage control strategy. It is suitable for high-complexity and high-dimensional system optimization, ensure the executability and practical implementation of the strategy, and reduce the sensitivity of the system to external environmental fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0046] Figure 1 This is a system block diagram of a heating energy-saving building, heat exchange station and boiler room linkage control system proposed by the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] Example 1

[0049] Reference Figure 1 , a heating energy-saving building, heat exchange station and boiler room linkage control system, including an acquisition and integration module, a graph construction module, a demand prediction module, a multi-criteria optimization module, a flow control module, a strategy generation module, a decision reinforcement module, a dynamic adjustment module, a feedback regulation module and a monitoring and maintenance module.

[0050] The acquisition and integration module is used to collect multi-source data from different sources and integrate them in real time to build a dynamic view of the global status.

[0051] Specifically, the environmental data of internal temperature, humidity and external meteorological conditions of the building, as well as the water temperature, water pressure of the heat exchange station and the fuel consumption and emission data of the boiler room are collected. Then, the data mean of each type of data is calculated separately, and the outliers in each type of data are identified based on the 3σ principle. The outliers and missing values ​​in each type of data are processed by linear interpolation. The data from different sources are normalized, and the data of different devices and regions are aligned according to the timestamps of each type of data. Then, the various types of data from different data sources are integrated based on the weighted average model to form a unified global dynamic view. Then, based on the various types of data collected in real time, the status of the buildings, heat exchange stations and boiler rooms are summarized to update the global dynamic view in real time.

[0052] The graph building module is used to construct an entity relationship network including buildings, heat exchange stations and boiler rooms.

[0053] Specifically, based on the constructed global dynamic view, detailed data from different devices, regions, and data sources are collected, each group of data is classified according to business needs, and entity information in each type of data is identified through NLP technology. The identified groups of entities are used as nodes, and the physical and logical relationships between entities are obtained according to the device operation mechanism and sensor data. The physical and logical relationships between the groups of entities are used as edges, and the corresponding nodes are connected through the generated edges. Corresponding attributes are assigned to each entity according to the device operation mechanism and sensor data to generate an entity relationship network. Based on the entity relationship network, reasoning rules are generated through semantic reasoning rules. When real-time data flows in, the entity relationship network uses real-time data to dynamically update the entity attributes and relationships in the network.

[0054] The demand forecasting module is used to predict future heating demand based on multiple sets of dynamic factors.

[0055] Specifically, historical heating data is extracted from the entity relationship network, and each group of heating data is preprocessed into multi-dimensional time series data to constitute the corresponding input data. Then, the input data is standardized and divided into training set, validation set and test set. A set of demand forecasting models is created based on the CNN architecture. According to the set training batches, the training set data is input into the demand forecasting model in sequence. The model input layer receives the formatted training set data and extracts the feature data in the time dimension through the one-dimensional convolution kernel in the convolution layer. The pooling layer downsamples the extracted feature data to reduce the feature dimension and reduce the computational complexity. The training data is repeatedly convolved through multiple groups of convolution layers and pooling layers for pooling operations, and the final feature The feature data is input into the fully connected layer, and then the fully connected layer expands the extracted feature data into a one-dimensional vector, and outputs the final predicted heat demand through the output layer. MSE is used as the loss function to calculate the deviation between the predicted value and the true value, and the deviation value is input from the model output layer according to the chain rule, and the gradient of the deviation value for each layer of the model is calculated. The Adam algorithm is then used to update the model parameters, and the evaluation index is calculated through the validation set and the test set to judge the model performance. If the performance of the demand forecasting model does not meet the preset requirements, it is retrained until the performance meets the standards and the training is stopped. The latest heating data collected in real time is input into the demand forecasting model, and the demand forecasting model outputs the heat demand for multiple time steps in the future through forward propagation.

[0056] The multi-criteria optimization module is used to output a preliminary heating resource allocation plan.

[0057] Example 2

[0058] Reference Figure 1 , a heating energy-saving building, heat exchange station and boiler room linkage control system, including an acquisition and integration module, a graph construction module, a demand prediction module, a multi-criteria optimization module, a flow control module, a strategy generation module, a decision reinforcement module, a dynamic adjustment module, a feedback regulation module and a monitoring and maintenance module.

[0059] The flow control module is used to control and optimize the circulation water flow between buildings, heat exchange stations and boiler rooms.

[0060] Specifically, the water flow, water temperature, pressure and various parameters of the circulating water flow system between the building, the heat exchange station and the boiler room are collected from the entity relationship network, and the hydraulic network topology is formed according to the connection relationship between the building, the heat exchange station and the boiler room, and the flow and path resistance of the circulation path are initialized. Multiple groups of flow schemes are obtained by random generation, and the generated schemes are grouped into a group of populations. At the same time, multiple groups of water droplets are generated, where the position of each water droplet represents a flow scheme. The positions of multiple groups of water droplets are randomly generated, and the fitness function is set based on the goal of minimizing the total energy consumption of the circulating water system. Then, the fitness value of the flow scheme corresponding to each group of water droplets is calculated by the fitness function, and the population is sorted according to the fitness value. The water drop with the highest fitness is the main river, and the remaining water drops are tributaries. The initial flow of the tributaries is allocated according to their fitness values. Based on the fitness values ​​of the main river and the tributaries, the distance between the main river and the tributaries is calculated, and two random factors are generated in the range of [0, 1]. Based on the generated random factors, the characteristic values ​​of the tributaries are continuously adjusted by simulating the linear update of the sample characteristics. and the position of tributaries, so that the tributaries and the main river are constantly approaching to form new tributaries. Based on the average or median of the current tributary fitness values, a fitness threshold is set. The fitness values ​​of the current tributaries are calculated through the objective function, and the calculated fitness values ​​are compared with the threshold values ​​respectively. If the fitness values ​​are less than the threshold, the corresponding samples are eliminated and no longer participate in the subsequent optimization process. A coefficient greater than 1 is selected as the precipitation factor by a random selection method. Then, the precipitation factor is amplified for tributaries greater than or equal to the preset threshold to increase the weight of high fitness samples in each test sample. After each round of iteration, the fitness of all main rivers and tributaries is recalculated according to the fitness function, and the main river and tributaries are updated. Then, the position update and main river update operations are performed again until the maximum number of iterations is reached, or the increment of the fitness of each water droplet is less than the set convergence threshold, or the screened water droplets meet the preset quality requirements, the iterative update is stopped, and the water droplets with the highest fitness value are screened out as the optimal flow scheme, and the circulating water flow is regulated based on the optimal flow scheme.

[0061] The strategy generation module is used to integrate demand forecasting and water cycle optimization results to generate linkage control strategies; the decision enhancement module is used to simulate and evaluate multiple linkage control strategies.

[0062] Specifically, the current building temperature state, heat exchange station state, boiler room state and environmental state are collected to construct the corresponding state space, and the corresponding action space is created according to the control strategies that can be adopted in each state in the state space. The current system operation state is used as the root node of the search tree. For the root node state, according to the available actions in the action space, the dynamic model of the heating system is used to calculate the new state information that the current state transfers to after executing random actions. For each state after transfer, a new node is added to the search tree, and the instant reward for taking actions from the current state to the new state is evaluated, and the initial value of the accumulated reward is the instant reward. Starting from the root node, the optimal node with the highest current UCB value is gradually selected along the constructed search tree through the UCB strategy. When the selected node has not been fully expanded When a new node is generated, an unexecuted action is selected based on the current state. According to the control strategy and system model, any random action is executed from the current state, and new state information is generated. The state information is added to the search tree as a new node. Starting from the expanded new node, various actions of the corresponding path are executed in a random or heuristic manner until the simulation or heating stable state of the predetermined time period is reached, and the cumulative rewards of all states experienced during the simulation are calculated. The cumulative rewards obtained by simulation are back-tracked to all nodes involved in the tree. The selection, expansion, simulation and backtracking steps are repeated many times until the reward value change of each node converges to the preset threshold. All linkage control strategies in the search tree are collected, and the path with the highest cumulative reward is selected as the current optimal strategy, which is applied to the actual linkage control system.

[0063] The dynamic adjustment module is used to handle real-time changes in heating demand and external environmental fluctuations, and dynamically optimize control strategies; the feedback adjustment module allows users to view the current heating status and historical energy consumption, and adjust heating parameters based on user feedback; the monitoring and maintenance module is used to perform real-time status monitoring of boilers, heat exchange stations and building heating equipment, and generate energy efficiency evaluation reports.

Claims

1. A heating energy-saving building, heat exchange station and boiler room linkage control system, characterized in that: It includes collection and integration module, graph construction module, demand forecasting module, multi-criteria optimization module, flow control module, strategy generation module, decision reinforcement module, dynamic adjustment module, feedback regulation module and monitoring and maintenance module; The acquisition and integration module is used to collect multi-source data from different sources and integrate them in real time to build a dynamic view of the global status; The graph construction module is used to construct an entity relationship network including buildings, heat exchange stations and boiler rooms; The demand forecasting module is used to forecast future heating demand based on multiple groups of dynamic factors; The multi-criteria optimization module is used to output a preliminary heating resource allocation plan; The flow control module is used to control and optimize the circulation water flow between the building, the heat exchange station and the boiler room; The strategy generation module is used to integrate demand forecasting and water cycle optimization results to generate a linkage control strategy; The decision-making reinforcement module is used to simulate and evaluate multiple linkage control strategies; The dynamic adjustment module is used to handle real-time changes in heating demand and external environmental fluctuations, and dynamically optimize the control strategy; The feedback adjustment module is used for users to view the current heating status and historical energy consumption, and adjust the heating parameters according to user feedback; The monitoring and maintenance module is used to perform real-time status monitoring of boilers, heat exchange stations and building heating equipment and generate energy efficiency evaluation reports.

2. A heating energy-saving building, heat exchange station and boiler room linkage control system according to claim 1, characterized in that: The specific steps of the acquisition and integration module to construct a dynamic view of the global state are as follows: S1.1: Collect environmental data such as temperature and humidity inside the building and external meteorological conditions, as well as water temperature and pressure of the heat exchange station and fuel consumption and emission data of the boiler room. Then calculate the mean of each type of data, identify outliers in each type of data based on the 3σ principle, and process outliers and missing values ​​in each type of data through linear interpolation. S1.2: Normalize data from different sources, align data from different devices and regions according to the timestamps of each type of data, and then integrate various types of data from different data sources based on the weighted average model to form a unified global dynamic view. Then, based on the various types of data collected in real time from buildings, heat exchange stations, and boiler rooms, summarize the status of buildings, heat exchange stations, and boiler rooms to update the global dynamic view in real time.

3. A heating energy-saving building, heat exchange station and boiler room linkage control system according to claim 2, characterized in that: The specific steps of the graph construction module to construct an entity relationship network including buildings, heat exchange stations and boiler rooms are as follows: S2.1: Based on the constructed global dynamic view, detailed data from different devices, regions, and data sources are collected, each group of data is classified according to business needs, and entity information in each type of data is identified through NLP technology, and each group of identified entities is used as a node; S2.2: According to the device operation mechanism and sensor data, obtain the physical and logical relationships between entities, use the physical and logical relationships between each group of entities as edges, connect the corresponding nodes through the generated edges, and assign corresponding attributes to each entity according to the device operation mechanism and sensor data to generate an entity relationship network; S2.3: Based on the entity relationship network, inference rules are generated through semantic inference rules. When real-time data flows in, the entity relationship network uses real-time data to dynamically update the entity attributes and relationships in the network.

4. A heating energy-saving building, heat exchange station and boiler room linkage control system according to claim 3, characterized in that: The specific steps of the demand forecasting module for forecasting future heating demand based on multiple groups of dynamic factors are as follows: S3.1: Extract historical heating data from the entity relationship network, and preprocess each set of heating data into multi-dimensional time series data to form the corresponding input data. Then, standardize the input data and divide it into training set, validation set and test set, and create a set of demand forecasting models based on CNN architecture; S3.2: According to the set training batches, the training set data is input into the demand forecasting model in sequence. The model input layer receives the formatted training set data and extracts the feature data in the time dimension through the one-dimensional convolution kernel in the convolution layer. The pooling layer downsamples the extracted feature data to reduce the feature dimension and reduce the computational complexity. S3.3: The training data is repeatedly convolved through multiple sets of convolutional layers and pooling layers for pooling operations, and the final feature data is input into a fully connected layer. The fully connected layer then expands the extracted feature data into a one-dimensional vector and outputs the final predicted heat demand through an output layer; S3.4: Use MSE as the loss function to calculate the deviation between the predicted value and the true value, and input the deviation value from the model output layer according to the chain rule, and calculate the gradient of the deviation value for each layer of the model, and then use the Adam algorithm to update the model parameters, calculate the evaluation index through the validation set and the test set, and judge the model performance. If the performance of the demand forecasting model does not meet the preset requirements, retrain until the performance meets the requirements and stop training; S3.5: The latest heating data collected in real time is input into the demand forecasting model, and the demand forecasting model outputs the heat demand for multiple future time steps through forward propagation.

5. A heating energy-saving building, heat exchange station and boiler room linkage control system according to claim 3, characterized in that: The specific steps of the flow control module to control and optimize the circulating water flow between the building, the heat exchange station and the boiler room are as follows: S4.1: Collect the water flow, water temperature, pressure and pump performance parameters of the circulating water flow system between the building, heat exchange station and boiler room from the entity relationship network, and form a hydraulic network topology based on the connection relationship between the building, heat exchange station and boiler room, initialize the flow and path resistance of the circulation path, obtain multiple groups of flow plans by random generation, and form a group of populations with the generated plans, and generate multiple groups of water droplets at the same time, where the position of each water droplet represents a flow plan; S4.2: Randomly generate the positions of multiple groups of water drops, set the fitness function based on the goal of minimizing the total energy consumption of the circulating water system, and then calculate the fitness value of each group of water drops corresponding to the flow scheme through the fitness function, and sort the population according to the fitness value. The water drop with the highest fitness is the main river, and the remaining water drops are tributaries, and their initial flow is allocated according to the tributary fitness value; S4.3: Based on the fitness values ​​of the main river and its tributaries, the distance between the main river and the tributaries is calculated, two random factors are generated in the range of [0, 1], and based on the generated random factors, the characteristic values ​​and positions of the tributaries are continuously adjusted by simulating the linear update of sample characteristics, so that the tributaries are constantly close to the main river to form new tributaries; S4.4: Based on the average or median of the current tributary fitness value, set the fitness threshold, calculate the fitness value of each current tributary through the objective function, and compare the calculated fitness value with the threshold value respectively. If it is less than the threshold, the corresponding sample will be eliminated and will no longer participate in the subsequent optimization process. A coefficient greater than 1 is selected as the precipitation factor by random selection. Then, the precipitation factor is amplified for tributaries greater than or equal to the preset threshold to increase the weight of high fitness samples in each test sample; S4.5: After each round of iteration, the fitness of all main rivers and tributaries is recalculated according to the fitness function, and the main rivers and tributaries are updated. Then, the position update and main river update operations are performed again until the maximum number of iterations is reached, or the increment of the fitness of each water droplet is less than the set convergence threshold, or the screened water droplets meet the preset quality requirements. Then, the iterative update is stopped, and the water droplets with the highest fitness value are screened as the optimal flow plan, and the circulating water flow is regulated based on the optimal flow plan.

6. A heating energy-saving building, heat exchange station and boiler room linkage control system according to claim 1, characterized in that: The specific steps of the decision reinforcement module simulation evaluation are as follows: S5.1: Collect the current building temperature status, heat exchange station status, boiler room status and environmental status to construct the corresponding state space, and create the corresponding action space according to the control strategy that can be taken in each state in the state space, and use the current system operation status as the root node of the search tree; S5.2: For the root node state, according to the available actions in the action space, through the dynamic model of the heating system, calculate the new state information that the current state transfers to after executing the random action. For each state after the transfer, add a new node to the search tree, and evaluate the immediate reward of taking the action from the current state to the new state, and accumulate the initial value of the reward as the immediate reward; S5.3: Starting from the root node, the optimal node with the highest current UCB value is gradually selected along the constructed search tree through the UCB strategy. When the selected node has not been fully expanded, an unexecuted action is selected based on the current state to generate a new node. According to the control strategy and system model, any random action is executed from the current state, and new state information is generated, which is added to the search tree as a new node; S5.4: Starting from the expanded new node, execute each action of the corresponding path in a random or heuristic manner until the simulation or heating stable state of the predetermined time period is reached, and calculate the cumulative rewards of all states experienced during the simulation process, and backtrack the cumulative rewards obtained from the simulation to all nodes involved in the tree; S5.5: Repeat the selection, expansion, simulation and backtracking steps multiple times until the reward value change of each node converges to the preset threshold, collect all the linkage control strategies in the search tree, and select the path with the highest cumulative reward as the current optimal strategy, and apply it to the actual linkage control system.

Images