Self-adaptive optimization control method and system for characteristics of circulating pump of heat exchange station

Abstract

The invention discloses a heat exchange station circulating pump characteristic adaptive optimization control method and system, and relates to the technical field of circulating pump control, and the method comprises the steps: building a building terminal heat transfer model and a building energy balance model based on the temperature distribution of the surface of a radiator, and then building a building load prediction model; performing numerical modeling on the performance curve of the heat exchange station circulating pump, and establishing a circulating pump performance model; based on a depth deterministic strategy gradient algorithm, establishing a neural network control model for self-adaptive control of the circulating pump of the heat exchange station; establishing a state, an action and a reward function of an agent of the neural network control model; and training the neural network control model to obtain a trained control model, and performing real-time control on the heat exchange station circulating pump based on the trained control model. The technical problems that a traditional control method of the circulating pump is low in operation efficiency and serious in energy waste are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of circulating pump control, and in particular to an adaptive optimization control method and system for the characteristics of a circulating pump in a heat exchange station. Background Technique

[0002] In a central heating system, as a key node connecting the primary heat network and the user side, the operation control of a heat exchange station directly affects the energy efficiency level and operation stability of the heating system. The secondary side circulating pump, as the main power equipment in the heat exchange station, is responsible for driving hot water to circulate between user buildings to meet the heat load requirements of radiators or floor heating systems at the end of each building. The adjustment of its operating frequency is not only related to the comfort of the building room temperature, but also directly determines the energy consumption level and operation economy of the system.

[0003] In current engineering practices, the control methods of secondary side circulating pumps mainly rely on fixed-frequency operation or simple PID control. Some systems make rough adjustments according to the temperature difference between the supply and return water or the supply water pressure on the secondary side. These traditional control methods have the advantages of simple implementation and rapid response, but generally have the following limitations: lack of the ability to perceive the dynamic changes of the system heat load, unable to accurately adjust the pump speed according to the actual demand at the end, prone to "over-supply" or "under-supply" phenomena, affecting heating comfort; the pump frequency setting depends on manual experience or on-site trial adjustment, lacking a real-time adaptive mechanism, with a fixed adjustment strategy, and difficult to cope with the complex and changeable characteristics of the system operation state; the control target is single, usually only focusing on the supply water temperature or pressure difference, and failing to incorporate the energy efficiency characteristics of the pump into the optimization target, resulting in the long-term operation of the circulating pump in an inefficient working condition and serious energy waste; traditional control is difficult to uniformly consider the two goals of "load matching" and "energy efficiency optimization", and it is difficult to achieve a balance between system adjustment and energy conservation. Summary of the Invention

[0004] The purpose of the present invention is to provide an adaptive optimization control method and system for the characteristics of a circulating pump in a heat exchange station to solve at least one of the above technical problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for adaptively optimizing the characteristics of a circulating pump in a heat exchange station, including: establishing a building end heat transfer model and a building energy balance model based on the temperature distribution on the surface of the radiator; establishing a building load prediction model based on the building end heat transfer model and the building energy balance model; numerically modeling the performance curve of the circulating pump in the heat exchange station to establish a circulating pump performance model; establishing a neural network control model for the adaptive control of the circulating pump in the heat exchange station based on the deep deterministic policy gradient algorithm; establishing the state, action, and reward functions of the intelligent agent of the neural network control model based on the building load prediction model and the circulating pump performance model; training the neural network control model based on the state, action, and reward functions of the intelligent agent to obtain a trained control model, and performing real-time control on the circulating pump in the heat exchange station based on the trained control model.

[0006] Further, the building end heat transfer model includes: ; In the formula, is the temperature distribution of the radiator in the vertical direction, is time, is the position of the radiator in the vertical direction, is the height of the radiator, is the volume of the radiator, is the volume flow rate of water flowing through the radiator at time is the heat transfer coefficient of the radiator, is the heat exchange area of the radiator, is the specific heat capacity of water, is the density of water, is the indoor temperature of the building room.

[0007] Further, the building energy balance model includes an external wall energy balance equation, an internal wall energy balance equation, and an indoor air energy balance equation; where the external wall energy balance equation includes: ; the internal wall energy balance equation includes: ; the indoor air energy balance equation includes: ; In the formula, is the specific heat capacity of the external wall, is the mass of the external wall, is the temperature of the external wall at time is the area of the external wall, is the convective heat transfer coefficient between the inner surface of the external wall and the indoor air, is Indoor temperature at a certain moment is the convective heat transfer coefficient between the outer surface of the exterior wall and the outdoor air is Outdoor temperature at a certain moment is the absorptivity of the exterior wall surface to solar radiation is Solar radiation intensity at a certain moment is the specific heat capacity of the interior wall is the mass of the interior wall is Temperature of the interior wall at a certain moment is the area of the interior wall is the specific heat capacity of the indoor air is the mass of the indoor air is the heat transfer coefficient of the exterior window is the area of the exterior window

[0008] Furthermore, the building load prediction model includes: ; In the formula, is Predicted heat load of the building at a certain moment is Actual heat load of the building at a certain moment is Indoor temperature of the building at a certain moment is Outdoor temperature at a certain moment is Solar radiation intensity at a certain moment is Outdoor relative humidity at a certain moment is the coefficient to be estimated by the model is the residual term Among them, the calculation formula for the actual heat load of the building includes: ; In the formula, is Actual heat load of the building at a certain moment is the specific heat capacity of water is the density of water is the flow rate of the circulation pump is Supply water temperature on the secondary side at a certain moment is Return water temperature on the secondary side at a certain moment The coefficient to be estimated by the model is determined through training with historical operation data

[0009] Furthermore, the circulation pump performance model includes: ; In the formula, is the efficiency of the circulation pump, is the flow rate of the circulation pump, is the head of the circulation pump, is the head coefficient of the circulation pump, is the effective power of the circulation pump, is the density of water, is the acceleration due to gravity, is the shaft power of the circulation pump, is the motor efficiency of the circulation pump, is the electric power of the circulation pump, is the ratio of the operating frequency to the rated frequency of the circulation pump.

[0010] Further, the neural network control model includes a policy network and a value network; wherein, the policy network includes two layers of fully connected networks and one layer of rectified linear unit activation network; the value network includes a value function network.

[0011] Further, the state of the agent of the neural network control model includes: ; In the formula, is the state vector of the agent at time is the flow rate of the circulation pump at time is the pressure difference between the inlet and outlet of the circulation pump at time is the frequency of the circulation pump at time is the electric power of the circulation pump at time is the secondary side supply water temperature at time is the secondary side return water temperature at time is the predicted heat load of the building at time; The actions of the agent of the neural network control model include: ; In the formula, is the adjustment increment of the circulation pump frequency at time is the maximum frequency change step allowed for the action, is the change in the circulation pump frequency.

[0012] The reward function of the agent of the neural network control model includes: ; In the formula, is the reward function at time is The actual heat load of the building at a certain moment, is the electric power of the circulating pump at a certain moment, and is a weighting coefficient.

[0013] In a second aspect, an embodiment of the present invention further provides an adaptive optimization control system for the characteristics of a heat exchange station circulating pump, including: a first establishment module, a second establishment module, a third establishment module, a fourth establishment module, a fifth establishment module, and a control module; wherein, the first establishment module is configured to establish a building end heat transfer model and a building energy balance model based on the temperature distribution on the surface of the radiator; the second establishment module is configured to establish a building load prediction model based on the building end heat transfer model and the building energy balance model; the third establishment module is configured to numerically model the performance curve of the heat exchange station circulating pump to establish a circulating pump performance model; the fourth establishment module is configured to establish a neural network control model for the adaptive control of the heat exchange station circulating pump based on the deep deterministic policy gradient algorithm; the fifth establishment module is configured to establish the state, action, and reward functions of the agent of the neural network control model based on the building load prediction model and the circulating pump performance model; the control module is configured to train the neural network control model based on the state, action, and reward functions of the agent to obtain a trained control model, and perform real-time control on the heat exchange station circulating pump based on the trained control model.

[0014] In a third aspect, an embodiment of the present invention further provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, the method provided by the embodiment of the present invention is implemented.

[0015] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method provided by the embodiment of the present invention is implemented.

[0016] The present invention provides a method and system for adaptively optimizing the characteristics of a circulating pump in a heat exchange station. By introducing a reinforcement learning agent, the system can real-time sense the operating state of the system, dynamically adjust the operating frequency of the circulating pump, and continuously approach the high-efficiency operating range of the circulating pump working condition point. While ensuring the stable supply of building heat load, it significantly reduces the energy consumption of the circulating pump. Compared with traditional fixed logic or PID control methods, the present invention has the ability of self-learning and self-adaptation, can cope with the changes of complex working conditions such as load fluctuations and system aging, and improves the robustness and intelligence level of the system operation. At the same time, by constructing an end-to-end intelligent control strategy, the present invention reduces the dependence on the debugging of empirical parameters and manual intervention, helps to realize the standardization, automation and remote control of pump station control, and alleviates the technical problems of low operating efficiency and serious energy waste existing in the traditional control method of circulating pumps. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for the specific embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of a method for adaptively optimizing the characteristics of a circulating pump in a heat exchange station provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of a system for adaptively optimizing the characteristics of a circulating pump in a heat exchange station provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0020] Embodiment 1

[0021] Figure 1 It is a flowchart of a method for adaptively optimizing the characteristics of a circulating pump in a heat exchange station provided by an embodiment of the present invention. As Figure 1 shown, the method specifically includes the following steps: Step S102, based on the temperature distribution on the surface of the radiator, establish a heat transfer model at the end of the building and a building energy balance model.

[0022] Step S104: Based on the building end heat transfer model and the building energy balance model, establish a building load prediction model.

[0023] Step S106: Numerically model the performance curve of the circulating pump in the heat exchange station to establish a circulating pump performance model.

[0024] Step S108: Based on the deep deterministic policy gradient algorithm, establish a neural network control model for the adaptive control of the circulating pump in the heat exchange station.

[0025] Step S110: Based on the building load prediction model and the circulating pump performance model, establish the state, action, and reward functions of the agent of the neural network control model.

[0026] Step S112: Based on the state, action, and reward functions of the agent, train the neural network control model to obtain the trained control model, and perform real-time control on the circulating pump in the heat exchange station based on the trained control model.

[0027] Specifically, the specific heat capacity of the water in the radiator is about 10 times that of its steel shell. Therefore, the steel shell is ignored to establish a dynamic heat transfer model of the radiator. The temperature distribution on the surface of the radiator decreases along the vertical direction and is uniformly distributed in the horizontal direction. Its heat dissipation process can be simplified to one-dimensional heat transfer. The building end heat transfer model includes: ; In the formula, T(t, y) is the temperature distribution of the radiator in the vertical direction, in °C; t is the time, in s; y is the position of the radiator in the vertical direction, Y is the height of the radiator, in m; V is the volume of the radiator, in m³; q(t) is the volume flow rate of the water flowing through the radiator at time t, in m³ / s; K is the heat transfer coefficient of the radiator, in W / (m²·°C); F is the heat exchange area of the radiator, in m²; c is the specific heat capacity of water, in kJ / (kg·°C); ρ is the density of water, in kg / m³; T in is the indoor temperature of the building room, in °C.

[0028] Specifically, the building energy balance model includes the external wall energy balance equation, the internal wall energy balance equation, and the indoor air energy balance equation; among them, For the external wall (including the roof), only one-dimensional heat transfer in the thickness direction is considered. The external wall energy balance equation includes: ; In the formula, is the specific heat capacity of the external wall, in J / (kg·°C); is the mass of the external wall, in kg; is the temperature of the external wall at time is the area of the external wall, in m2 ; is the convective heat transfer coefficient between the inner surface of the exterior wall and the indoor air, with the unit of W / (m 2 ·K); is the indoor temperature at the moment, with the unit of °C; is the convective heat transfer coefficient between the outer surface of the exterior wall and the outdoor air, with the unit of W / (m 2 ·K); is the outdoor temperature at the moment, with the unit of °C; is the absorptivity of the exterior wall surface to solar radiation; is the solar radiation intensity at the moment, with the unit of W / m 2 .

[0029] For the interior wall (including the floor), only one-dimensional heat transfer in the thickness direction is considered, and heat transfer between adjacent rooms and between upper and lower floors is ignored. The energy balance equation of the interior wall includes: ; In the formula, is the specific heat capacity of the interior wall, with the unit of J / (kg·°C); is the mass of the interior wall, with the unit of kg; is the temperature of the interior wall at the moment, with the unit of °C; is the area of the interior wall, with the unit of m 2 .

[0030] The energy balance equation of the indoor air includes: ; In the formula, is the specific heat capacity of the indoor air, with the unit of J / (kg·K); is the mass of the indoor air, with the unit of kg; is the heat transfer coefficient of the exterior window, with the unit of W / (m 2 ·K); is the area of the exterior window, with the unit of m 2 ; is the outdoor temperature at the moment, with the unit of °C.

[0031] To achieve the active optimization control of the circulation pump operation, it is necessary to first obtain the predicted value of the building load. The building terminal load directly determines the required heat and circulation flow of the system. In this invention, multiple linear regression is used to perform short-term prediction on the load, providing a basis for subsequent pump control decisions. Multiple linear regression constructs a linear combination model to fit the relationship between the building heat load at the terminal and a set of influencing factors. Specifically, the building load prediction model includes: ; In the formula, is The predicted heating load of the building at time t, in kW; is The actual heating load of the building at time t, in kW; is The indoor temperature of the building at time t, in °C; is The solar radiation intensity at time t, in W / m²; is The outdoor relative humidity at time t, in %; are the coefficients to be estimated by the model, is the residual term, following a normal distribution; wherein, the calculation formula for the actual heating load of the building includes: ; In the formula, is The actual heating load of the building at time t, in kW; is the specific heat capacity of water, in kJ / (kg·°C); is the density of water, in kg / m 3 ; is the flow rate of the circulation pump, in m 3 / s; is The supply water temperature on the secondary side at time t, in °C; is The return water temperature on the secondary side at time t, in °C.

[0032] The coefficients to be estimated by the model are determined through training with historical operation data. Specifically, the initial building load prediction model is trained with historical operation data. When sufficient data for the current heating season is collected, a new model is continuously regressed using the new data, and the weather forecast data for the next time step and the indoor data for the current time step are input to predict the building load for the next time step. The prediction results will be directly input into the reinforcement learning controller to assist in optimizing the pump speed control strategy.

[0033] Specifically, the circulation pump performance model includes: ; In the formula, is the efficiency of the circulation pump, in %; is the flow rate of the circulation pump, in m 3 / s; is the head of the circulation pump, in m·H2O; is the head coefficient of the circulation pump, is the effective power of the circulation pump, in kW; is the density of water, in kg / m 3 ; is the acceleration due to gravity, 9.81 m / s²; is the shaft power of the circulating pump, with the unit of kW; is the motor efficiency of the circulating pump, with the unit of %; is the electric power of the circulating pump, with the unit of kW; is the ratio of the operating frequency to the rated frequency of the circulating pump.

[0034] Specifically, in order to enable the deep reinforcement learning algorithm to effectively explore and optimize the operating state of the circulating pump, a numerical model of the circulating pump performance curve is established, the head expression of the circulating pump at different flow rates is established, and the response model of the reinforcement learning interaction environment is constructed therefrom. At a certain constant frequency, binomial fitting is used to obtain the flow-head characteristic curve of the circulating pump: ; In actual projects, variable-frequency circulating pumps are generally used for secondary network circulating pumps to reduce transmission and distribution energy consumption. According to the similarity law, when the same circulating pump operates at variable speed, the performance parameters between similar operating points satisfy the following formula: ; where the subscript 0 represents the parameters of the circulating pump at the rated frequency of 50 Hz, and no subscript represents the parameters of the circulating pump at the operating frequency. is the rotational speed of the circulating pump, is the frequency of the circulating pump. is the frequency ratio, representing the ratio of the operating frequency to the rated frequency of the circulating pump. Substituting the proportional relationship of each parameter of the circulating pump in Equation (2) into Equation (1), the flow-head relationship formula of the variable-frequency circulating pump is obtained: ; The effective power of the circulating pump indicates that the total energy obtained by the fluid passing through the pump per unit time is: ; The shaft power of the circulating pump indicates that the mechanical power transmitted by the motor to the shaft of the circulating pump is: ; Using the efficiency of the circulating pump indicates the degree to which the input shaft power is utilized by the fluid as: ; By using binomial fitting, the flow-efficiency characteristic curve of the circulating pump is obtained as: ; In the formula, A, B, and C are the efficiency coefficients of the circulating pump.

[0035] At different frequencies, the flow rate and pressure passing through the operating point are binomially fitted to obtain the system resistance characteristic curve. The system resistance characteristic curve reflects the relationship between the flow resistance in the pipe network and the flow rate, and the resistance pressure drop is proportional to the square of the flow rate: ; In the formula: is the system at the flow rate The required head below, unit: m·H2O; is the system resistance coefficient, comprehensively reflecting the influence of pipeline length, diameter, local resistance, valve opening, etc. on water flow, unit: s 2 / m 5 .

[0036] Specifically, the neural network control model includes a policy network Actor and a value network Critic; among them, the policy network is a policy network, which inputs the current system state and outputs continuous actions, that is, the control frequency of the circulating pump, including two layers of fully connected networks and one layer of rectified linear unit activation network (Rectified Linear Unit, ReLU); the value network includes a value function network, which inputs and actions, and outputs the Q value (i.e., the future cumulative reward estimate) of this action in this state, and is a three-layer structure.

[0037] Reinforcement learning is an intelligent learning method that learns the optimal behavior strategy by interacting with the environment, continuously trying mistakes and accumulating experience. Its basic idea comes from the "reward and punishment mechanism" in psychology: the intelligent agent tries different actions, and gradually improves the behavior strategy according to the rewards or punishments given by the environment to maximize the long-term cumulative return. The embodiment of the present invention adopts the Deep Deterministic Policy Gradient (DDPG) algorithm to realize the adaptive operating point control of the circulating pump. DDPG is a reinforcement learning method based on the actor-critic structure and is suitable for continuous action space scenarios. The control variable of the circulating pump is the frequency conversion frequency, which is a typical continuous control problem, so it is particularly suitable to adopt DDPG.

[0038] Specifically, the state of the intelligent agent of the neural network control model is the state observed by the intelligent agent at each control step, including: ; In the formula, is the state vector of the intelligent agent at time is the flow rate of the circulating pump at time, unit: m 3 / s; is the pressure difference between the inlet and outlet of the circulating pump at time, unit: m·H2O; is the frequency of the circulating pump at time, unit: Hz; is the electric power of the circulating pump at time, unit: kW; is the secondary side supply water temperature at time, unit: °C; is the secondary side return water temperature at time, unit: °C; For the predicted heat load of the building at time, in kW.

[0039] The action of the agent in the neural network control model, that is, how the agent controls the environment. In the embodiments of the present invention, the action is the frequency increment of the circulation pump, specifically including: ; Wherein, is the adjustment increment of the circulation pump frequency at time, in Hz; is the maximum frequency change step allowed for the action, in Hz; is the change amount of the circulation pump frequency.

[0040] The reward function of the agent in the neural network control model will significantly affect the convergence of the deep reinforcement learning and the final control performance, and determine the control optimization goal. The goal of the embodiments of the present invention is to make the circulation pump operate in the high-efficiency area on the premise of matching the terminal heat load. Specifically, the reward function includes: ; Wherein, is the reward function at time, is the actual heat load of the building at time, in kW; is the weighting coefficient.

[0041] Specifically, the neural network control model further includes environmental feedback: the neural network control model receives the frequency , and obtains the flow rate and the head by using the flow rate-head characteristic curve of the circulation pump and the system characteristic curve, and further calculates the effective power ; obtains the operating efficiency of the circulation pump at the current flow rate by using the flow rate-efficiency characteristic curve of the circulation pump, and further calculates the electric power of the circulation pump through the effective power; calculates the return water temperature of the least-utilized user as the building return water temperature and the indoor temperature of the least-utilized user by means of the building terminal radiator model and the least-utilized user of the building through the current flow rate and the supply water temperature , and further calculates the actual heat load of the building and predicts the building heat load at the next moment.

[0042] Specifically, the update strategy of the Actor-Critic structure in the neural network control model is as follows: Initialize the experience pool Replay Buffer, and randomly explore and collect , in each round, sample a training batch from the experience pool and update the Critic network. The goal is to minimize the temporal difference error: ; Wherein, is the loss function of the Critic network; are the parameters to be optimized by the Critic network; is the size of the sampled batch; is the state of the i-th piece of experience data; is the action of the i-th piece of experience data; is the value estimation of the current Critic network for ; value estimation; is the target value of the i-th piece of experience; .

[0043] Wherein, is the immediate reward obtained from the i-th piece of experience; is the discount factor, which controls the influence degree of future rewards; is the next state after executing the action at the i-th step; is the action given by the target Actor network at ; is the value evaluation of the target Critic network for ; value evaluation; are the parameters of the target Actor and target Critic networks respectively.

[0044] Use policy gradient to optimize the Actor network. The goal is to maximize the value given by the Critic network: ; Wherein, is the gradient of the objective function (total expected reward) of the policy network with respect to its parameter ; are the parameters of the Actor network; is the size of the sampled batch; is the value function given by the Critic network; is the gradient of the function with respect to the action a, and is calculated on the action output by the current policy of the Actor network; is the gradient of the action output by the policy network with respect to its parameter; is the policy objective function.

[0045] Specifically, the soft replacement of the target network includes: To ensure stable training, a target network is introduced. It is updated smoothly and iteratively, following the main network slowly to avoid drastic oscillations during the policy update process: ; wherein, are the parameters of the target network; are the parameters of the current main network; is the soft update factor.

[0046] The training of the neural network control model goes through multiple rounds, repeating the sampling - update - soft replacement process. When the average reward rises steadily and the control frequency tends to be reasonable, it indicates that the policy converges. The finally derived Actor network can be used as the control policy to output the frequency output frequency in real - time systems according to the state, for real - time adjustment of the circulating pump.

[0047] The embodiment of the present invention provides a method for adaptively optimizing the characteristics of a circulating pump in a heat exchange station. A neural network model composed of a policy network (Actor) and a value network (Critic) is constructed to control the frequency of the circulating pump in real - time for efficient energy - saving operation. The system state design covers the key parameters of the circulating pump operation, including flow rate, pressure difference, frequency, electric power, supply and return water temperatures, and building predicted heat load, constituting the state vector perceived by the agent at each control step. The control action is set as the adjustment increment of the pump frequency to guide the dynamic adjustment of the operating point of the circulating pump. The reward function combines two objectives: load tracking accuracy and operation efficiency deviation, aiming to guide the system to operate in the efficient range while meeting the heat load demand. The environmental model provides a real - interaction environment by feedbacking various state information based on the flow - head characteristic curve, flow - efficiency characteristic curve, system resistance curve, building terminal model, and building user model of the pump. During the training process, the experience replay mechanism and the temporal difference method are used to update the Critic network, the Actor network is optimized by policy gradient, and at the same time, a target network is introduced for soft replacement to enhance the stability and convergence of the training. When the average reward of the agent tends to be stable and the control behavior is reasonable during the training process, it indicates that the policy converges. The finally trained Actor network can be deployed as an online control policy to output the frequency of the circulating pump according to the real - time state, realizing the efficient and adaptive operation of the system.

[0048] Embodiment Two

[0049] Figure 2 is a schematic diagram of an adaptive optimization control system for the characteristics of a circulating pump in a heat exchange station according to the embodiment of the present invention. As Figure 2 shown, the system includes: a first establishment module 10, a second establishment module 20, a third establishment module 30, a fourth establishment module 40, a fifth establishment module 50, and a control module 60.

[0050] Specifically, a first model building module 10 is configured to build a building end heat transfer model and a building energy balance model based on the temperature distribution on the surface of the radiator; A second model building module 20 is configured to build a building load prediction model based on the building end heat transfer model and the building energy balance model; A third model building module 30 is configured to numerically model the performance curve of the circulating pump in the heat exchange station to build a circulating pump performance model; A fourth model building module 40 is configured to build a neural network control model for adaptive control of the circulating pump in the heat exchange station based on the deep deterministic policy gradient algorithm; A fifth model building module 50 is configured to build the state, action, and reward functions of the agent of the neural network control model based on the building load prediction model and the circulating pump performance model; A control module 60 is configured to train the neural network control model based on the state, action, and reward functions of the agent to obtain a trained control model, and perform real-time control on the circulating pump in the heat exchange station based on the trained control model.

[0051] The present invention also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the method provided by the embodiment of the present invention is implemented.

[0052] The present invention also provides a computer-readable storage medium, where the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method provided by the embodiment of the present invention is implemented.

[0053] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0054] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only includes an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An adaptive optimization control method for the characteristics of a circulating pump in a heat exchange station, characterized in that, including: Based on the temperature distribution on the surface of the radiator, establish a building terminal heat transfer model and a building energy balance model; Based on the building terminal heat transfer model and the building energy balance model, establish a building load prediction model; Conduct numerical modeling on the performance curve of the circulating pump in the heat exchange station to establish a circulating pump performance model; Based on the deep deterministic policy gradient algorithm, establish a neural network control model for the adaptive control of the circulating pump in the heat exchange station; Based on the building load prediction model and the circulating pump performance model, establish the state, action, and reward functions of the agent of the neural network control model; Based on the state, action, and reward functions of the agent, train the neural network control model to obtain a trained control model, and perform real-time control on the circulating pump in the heat exchange station based on the trained control model.

2. The method according to claim 1, wherein: The building terminal heat transfer model includes: ； In the formula, is the temperature distribution of the radiator in the vertical direction, is the time, is the position of the radiator in the vertical direction, is the height of the radiator, is the volume of the radiator, is the volume flow rate of water flowing through the radiator at time is the heat transfer coefficient of the radiator, is the heat exchange area of the radiator, is the specific heat capacity of water, is the density of water, is the indoor temperature of the building room.

3. The method according to claim 1, wherein: The building energy balance model includes an external wall energy balance equation, an internal wall energy balance equation, and an indoor air energy balance equation; among them, The external wall energy balance equation includes: ; The internal wall energy balance equation includes: ; The indoor air energy balance equation includes: ; In the formula, is the specific heat capacity of the exterior wall, is the mass of the exterior wall, is the temperature of the exterior wall at time is the area of the exterior wall, is the convective heat transfer coefficient between the inner surface of the exterior wall and the indoor air, is the indoor temperature at time is the convective heat transfer coefficient between the outer surface of the exterior wall and the outdoor air, is the outdoor temperature at time is the absorptivity of the exterior wall surface to solar radiation, is the solar radiation intensity at time is the specific heat capacity of the interior wall, is the mass of the interior wall, is the temperature of the interior wall at time is the area of the interior wall, is the specific heat capacity of the indoor air, is the mass of the indoor air, is the heat transfer coefficient of the exterior window, is the area of the exterior window.

4. The method according to claim 1, wherein: The building load prediction model includes: ； In the formula, is the predicted heat load of the building at time is the actual heat load of the building at time is the indoor temperature of the building at time is the outdoor temperature at time is the solar radiation intensity at time is the outdoor relative humidity at time is the coefficient to be estimated by the model, and is the residual term; Among them, the calculation formula for the actual heat load of the building includes: ; Wherein, is the actual heat load of the building at the moment, the specific heat capacity of water, is the flow rate of the circulation pump, is the secondary side supply water temperature at the moment, is the secondary side return water temperature at the moment; The coefficients to be estimated in the model are determined by training with historical operation data.

5. The method according to claim 1, wherein: The performance model of the circulation pump includes: ; In the formula, is the efficiency of the circulation pump, is the flow rate of the circulation pump, is the head of the circulation pump, is the head coefficient of the circulation pump, is the effective power of the circulation pump, is the density of water, is the acceleration due to gravity, is the shaft power of the circulation pump, is the motor efficiency of the circulation pump, is the electric power of the circulation pump, is the ratio of the operating frequency to the rated frequency of the circulation pump.

6. The method according to claim 1, characterized in that: The neural network control model includes a policy network and a value network; among them, the policy network includes two layers of fully connected networks and one layer of rectified linear unit activation network; the value network includes a value function network.

7. The method according to claim 1, characterized in that: The state of the agent of the neural network control model includes: ； Wherein, is the time state vector of the said agent, is the flow rate of the circulating pump at is the pressure difference between the inlet and outlet of the circulating pump at is the frequency of the circulating pump at is the electric power of the circulating pump at is the secondary side supply water temperature at is the secondary side return water temperature at is the predicted heat load of the building at The actions of the agent of the neural network control model include: ; In the formula, is the adjustment increment of the circulating pump frequency at time the maximum frequency change step allowed for the action, is the change amount of the circulating pump frequency; The reward function of the agent in the neural network control model includes: ; In the formula, is the moment reward function, is the actual heat load of the building at the moment, is the electric power of the circulating pump at the moment, is the weighting coefficient.

8. An adaptive optimization control system for the characteristics of a circulating pump in a heat exchange station, characterized in that, including: A first establishment module, a second establishment module, a third establishment module, a fourth establishment module, a fifth establishment module, and a control module; among them, The first establishment module is used to establish a building terminal heat transfer model and a building energy balance model based on the temperature distribution on the surface of the radiator; The second establishment module is used to establish a building load prediction model based on the building terminal heat transfer model and the building energy balance model; The third establishment module is used to conduct numerical modeling on the performance curve of the circulating pump in the heat exchange station to establish a circulating pump performance model; The fourth establishment module is used to establish a neural network control model for the adaptive control of the circulating pump in the heat exchange station based on the deep deterministic policy gradient algorithm; The fifth establishment module is used to establish the state, action, and reward functions of the agent of the neural network control model based on the building load prediction model and the circulating pump performance model; The control module is used to train the neural network control model based on the state, action, and reward functions of the agent to obtain a trained control model, and perform real-time control on the circulating pump in the heat exchange station based on the trained control model.

9. An electronic device, characterized in that, including: A memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, it implements the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by the processor, it implements the method according to any one of claims 1-7.

Images