Solar energy and ground source heat pump coupling heat supply method and system based on machine learning

Abstract

The invention discloses a solar energy and ground source heat pump coupling heat supply method and system based on machine learning, and relates to the technical field of heat supply energy management, and the method comprises the steps: taking the minimum system energy consumption as the target, and using a double-delay depth deterministic strategy gradient neural network model to optimize the output distribution of the solar energy and the ground source heat pump. The model adapts to dynamic working conditions, solar energy is utilized preferentially, and the high-energy-consumption operation time of the ground source heat pump is shortened. Multi-dimensional data are collected through the Internet of Things and an environment monitoring system to construct a state space, a heat supply temperature deviation and equipment operation efficiency reward function is introduced, it is guaranteed that the heat supply temperature is stable, efficient operation of the ground source heat pump is guaranteed, and complex working conditions are dealt with. And full-process automation is realized by relying on a machine learning model, a load distribution sequence is generated through an optimal heat supply load distribution model, and dynamic adjustment is automatically completed by combining a PI controller to output total demand power.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heat energy management, more particularly to a solar energy and ground source heat pump coupled heat supply method and system based on machine learning. BACKGROUND

[0002] In recent years, the core development goal of the heating industry is to transition to clean and low-carbon. Since traditional fossil energy heating accounts for a large proportion of energy consumption and carbon emissions, energy saving and emission reduction has become a priority for the heating industry. Under this background, solar energy and ground source heat pump coupled heat supply systems are considered to be promising solutions. It can effectively solve the intermittency problem of single solar heat supply and the high energy consumption problem of single ground source heat pump heat supply while achieving low emissions, low cost and high efficiency, so it has received extensive attention and research.

[0003] In the coupled heat supply system, energy management is a key factor affecting the economy of the system. The energy saving mechanism of the coupled heat supply system is to optimize the output distribution of the two heat sources in the high efficiency range, and use the ground source heat pump to compensate for the gap between the solar energy output and the current heat demand. Therefore, it is very important to develop an energy management strategy for the solar energy-ground source heat pump coupled heat supply system to improve its adaptability under different environmental conditions. This means that the use of energy needs to be optimized so that the solar heat supply system and the ground source heat pump work more efficiently under different operating conditions. Through a reasonable energy management strategy, the best heat supply output and energy utilization rate can be ensured under different heat supply scenarios, thereby improving the economy of the system.

[0004] Many current energy management strategies for coupled heat supply systems are designed based on standard heat supply operating conditions and pay less attention to real operating data; the heat supply system is part of the energy system, and its operating conditions are closely related to the environment. Therefore, it is of great significance to optimize the energy management strategy in combination with real operating data and real-time operating conditions so that the optimized strategy can adapt to various operating conditions and achieve energy management optimization of the coupled heat supply system.

[0005] Therefore, how to propose a solar energy and ground source heat pump coupled heat supply method and system based on machine learning to overcome the defects in the prior art is a problem that technicians in the field need to solve urgently. SUMMARY

[0006] Therefore, the present application provides a solar energy and ground source heat pump coupled heat supply method and system based on machine learning to overcome the defects in the prior art. In order to achieve the above purpose, the present application adopts the following technical solutions: A solar energy and ground source heat pump coupled heat supply method based on machine learning, comprising: The heat supply load distribution model is constructed, real-time data is obtained, and the real-time data is input into the heat supply load distribution model to obtain an optimal heat supply load distribution scheme at each time in a future period of time, and an optimal load distribution sequence is formed. Based on the optimal load distribution sequence, total heat supply demand power at each time in the future period of time is calculated to obtain a total demand power sequence. Based on the total demand power sequence, the solar heat supply system output sequence and the ground source heat pump system output sequence are determined with the minimum system energy consumption as the target. Based on the solar heat supply system output sequence and the ground source heat pump system output sequence, the coupled heat supply system is controlled.

[0007] Optionally, the heat supply load distribution model comprises: The double-delay deep deterministic policy gradient neural network model is initialized, the double-delay deep deterministic policy gradient neural network model constructs a state space with the current ambient temperature, the solar radiation intensity, the soil temperature, the heat supply terminal demand temperature, the system energy storage device state of charge, the solar collector efficiency and the ground source heat pump COP value, constructs an action space with the solar heat supply system output proportion, the reward function is constructed by the system total energy consumption reward function, the heat supply temperature deviation reward function, the solar utilization rate reward function and the equipment operation efficiency reward function, and after training, the Actor target network is used as the optimal heat supply load distribution model.

[0008] Optionally, the double-delay deep deterministic policy gradient neural network model comprises: The expression of the double-delay deep deterministic policy gradient neural network model is as follows: ; Wherein, S represents the state space, A represents the action space, and R represents the reward function. ; Wherein, represents the current ambient temperature, represents the current solar radiation intensity, represents the current soil temperature, represents the current heat supply terminal demand temperature, represents the current system energy storage device state of charge, represents the current solar collector efficiency, represents the current ground source heat pump COP value. ; Wherein, represents the current solar heat supply system output proportion. ; wherein, is a weight factor representing the total energy consumption reward function of the system per unit time, is a weight factor representing the heating temperature deviation reward function per unit time, is a weight factor representing the solar energy utilization rate reward function per unit time, is a weight factor representing the equipment operation efficiency reward function per unit time, is the total energy consumption reward function of the system per unit time, is the heating temperature deviation reward function per unit time, is the solar energy utilization rate reward function per unit time, is the equipment operation efficiency reward function per unit time.

[0009] Optionally, the total energy consumption reward function per unit time is the sum of the energy consumption of the solar heating system and the energy consumption of the ground source heat pump system per unit time step.

[0010] Optionally, the expression of the heating temperature deviation reward function per unit time is as follows: ; wherein, , respectively represent the maximum and minimum water supply temperatures allowed by the heating terminal, represents the actual water supply temperature at the current time, and exp represents the exponential function.

[0011] Optionally, the expression of the solar energy utilization rate reward function per unit time is as follows: ; wherein, represents the solar energy utilization threshold, and k represents the solar energy utilization reward coefficient.

[0012] Optionally, the expression of the equipment operation efficiency reward function per unit time is as follows: ; wherein, represents the ground source heat pump operation efficiency threshold, and m represents the equipment efficiency reward coefficient.

[0013] Optionally, based on the optimal load distribution sequence, the total heating demand power at each time in the future period of time is calculated to obtain a total demand power sequence, including: calculating the theoretical heating demand power according to the heating terminal demand temperature, the building area, and the heat load index of each future time; The PI controller model in the system outputs the total heat supply demand power at each time in the future period of time based on the load distribution ratio in the optimal load distribution sequence, and a total demand power sequence is formed.

[0014] Optionally, the total demand power sequence is used to determine the solar heat supply system output sequence and the ground source heat pump system output sequence with the minimum system energy consumption as the target, and the following cost function is used for control: ; ; wherein, is the instantaneous energy consumption of the ground source heat pump, is the instantaneous energy consumption of the solar heat supply system, c is an equivalent factor, is the solar heat supply system output sequence, is the ground source heat pump system output, is the equivalent heat value of the solar heat supply system, represents the state of charge of the energy storage device at the initial time of the cycle, U is the solar heat supply system output sequence in the future period of time, and l represents the optimization objective function, and J represents the cost function.

[0015] Optionally, a solar heat supply system and a ground source heat pump coupled heat supply system based on machine learning comprises: a model construction module configured to construct a heat supply load distribution model; a training module configured to obtain real-time data and input the real-time data into the heat supply load distribution model to obtain an optimal heat supply load distribution scheme at each time in the future period of time and form an optimal load distribution sequence; a prediction module configured to calculate the total heat supply demand power at each time in the future period of time based on the optimal load distribution sequence and obtain a total demand power sequence; a constraint module configured to determine the solar heat supply system output sequence and the ground source heat pump system output sequence with the minimum system energy consumption as the target based on the total demand power sequence; a control module configured to control the coupled heat supply system based on the solar heat supply system output sequence and the ground source heat pump system output sequence.

[0016] According to the technical solution, compared with the prior art, the solar heat supply system and the ground source heat pump coupled heat supply method and system based on machine learning have the following beneficial effects: This invention proposes a machine learning-based method for coupled solar and ground-source heat pump heating, comprising: constructing a heating load allocation model, acquiring real-time data, inputting it into the heating load allocation model to obtain the optimal heating load allocation scheme for each moment in the future period, forming an optimal load allocation sequence; based on the optimal load allocation sequence, calculating the total heating demand power for each moment in the future period, obtaining a total demand power sequence; based on the total demand power sequence, determining the output sequence of the solar heating system and the output sequence of the ground-source heat pump system with the goal of minimizing system energy consumption; and controlling the coupled heating system based on the output sequences of the solar heating system and the ground-source heat pump system.

[0017] This invention (1) takes minimizing system energy consumption as its core objective and uses a dual-delay deep deterministic strategy gradient neural network model to accurately optimize the output allocation between solar energy and ground source heat pumps. The model can adapt to dynamic changes in operating conditions such as solar irradiance and ambient temperature in real time, prioritizing the maximization of solar energy utilization (guided by a solar energy utilization reward function) and reducing the operating time of ground source heat pumps in high-energy-consumption ranges.

[0018] (2) Real-time collection of multi-dimensional data such as ambient temperature, soil temperature, and heating terminal demand temperature through the Internet of Things and environmental monitoring system to construct a comprehensive state space; at the same time, the heating temperature deviation reward function and the equipment operation efficiency reward function are introduced to ensure that the heating terminal temperature is stable within a reasonable range and that the ground source heat pump operates in a high-efficiency range, effectively coping with complex working conditions such as intermittent solar energy and sudden changes in ambient temperature.

[0019] (3) Relying on machine learning models to achieve full-process automation of load forecasting, power calculation, output allocation and system control: The load allocation sequence for a future period of time is generated in advance by the optimal heating load allocation model, and the total demand power is accurately output by the PI controller, so that dynamic adjustment can be completed without manual intervention.

[0020] (4) Through multi-dimensional reward function collaborative optimization, the solar energy utilization rate reward function incentivizes the system to operate at irradiance ≥ Make full use of solar energy resources and reduce ineffective charging and discharging of energy storage devices; the equipment operation efficiency reward function avoids the ground source heat pump from wasting energy in the low efficiency range; the total system energy consumption reward function controls energy consumption from a global perspective. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 A machine learning-based solar-geothermal heat pump coupled heating method flowchart is provided. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0024] The embodiments of the present application disclose a machine learning-based solar-geothermal heat pump coupled heating method, as shown in the figure, comprising: Figure 1 constructing a heating load distribution model, obtaining real-time data, inputting into the heating load distribution model, obtaining an optimal heating load distribution scheme at each time in the future period, forming an optimal load distribution sequence; Based on the optimal load distribution sequence, the total heating demand power at each time in the future period is calculated, and a total demand power sequence is obtained; Based on the total demand power sequence, the solar heating system output sequence and the geothermal heat pump system output sequence are determined with the minimum system energy consumption as the target; Based on the solar heating system output sequence and the geothermal heat pump system output sequence, the coupled heating system is controlled.

[0025] Further, the construction of the heating load distribution model comprises training of a double-delay deep deterministic policy gradient neural network model to obtain an optimal heating load distribution model: The double-delay deep deterministic policy gradient neural network model is initialized, the double-delay deep deterministic policy gradient neural network model constructs a state space with the current ambient temperature, solar radiation intensity, soil temperature, heating terminal demand temperature, system energy storage device state of charge, solar collector efficiency and geothermal heat pump COP value, and constructs an action space with the solar heating system output ratio; the reward function is constructed by the system total energy consumption reward function, the heating temperature deviation reward function, the solar utilization rate reward function and the equipment operation efficiency reward function; after training, the Actor target network is used as the optimal heating load distribution model.

[0026] ​Further, the real-time data includes real-time model input data obtained based on an Internet of Things and / or an environmental monitoring system. The real-time model input data obtained based on the Internet of Things and / or the environmental monitoring system includes current environmental temperature, solar radiation intensity, soil temperature, heating terminal demand temperature, system energy storage device state of charge, solar collector efficiency, and ground source heat pump COP value. After obtaining the required real-time data, the real-time data is input as a parameter of a state space into the optimal heating load distribution model, and finally, a corresponding solar heating system output ratio is output from an action space, so as to obtain an optimal load distribution scheme and form an optimal load distribution sequence.

[0027] Further, based on the total demand power sequence, the solar heating system output sequence and the ground source heat pump system output sequence are determined with the minimum system energy consumption as the target, wherein the sum of the solar heating system output and the ground source heat pump system output at each time in a future period of time is equal to the total heating demand power.

[0028] Further, the double-delay deep deterministic policy gradient neural network model includes: The expression of the double-delay deep deterministic policy gradient neural network model is as follows: ; wherein S represents a state space, A represents an action space, and R represents a reward function; ; wherein, represents a current time environmental temperature, represents a current time solar radiation intensity, represents a current time soil temperature, represents a current time heating terminal demand temperature, represents a current time system energy storage device state of charge, represents a current time solar collector efficiency, represents a current time ground source heat pump COP value; ; wherein, represents a current time solar heating system output ratio; ; wherein, represents a weight factor of a system total energy consumption reward function per unit time, represents a weight factor of a heating temperature deviation reward function per unit time, represents a weight factor of a solar utilization rate reward function per unit time, represents a weight factor of an equipment operation efficiency reward function per unit time, is the total energy consumption reward function of the system per unit time, is the heating temperature deviation reward function per unit time, is the solar energy utilization rate reward function per unit time, is the equipment operation efficiency reward function per unit time.

[0029] Further, the total energy consumption reward function of the system per unit time is the sum of the solar heating system energy consumption and the ground source heat pump system energy consumption per unit time step.

[0030] Further, the expression of the heating temperature deviation reward function per unit time is as follows: ; wherein, , respectively represent the maximum and minimum water supply temperatures allowed by the heating terminal, represents the actual water supply temperature at the current time, and exp represents the exponential function.

[0031] Further, the expression of the solar energy utilization rate reward function per unit time is as follows: ; wherein, represents the solar energy utilization threshold, and k represents the solar energy utilization reward coefficient.

[0032] Further, the expression of the equipment operation efficiency reward function per unit time is as follows: ; wherein, represents the ground source heat pump operation efficiency threshold, and m represents the equipment efficiency reward coefficient.

[0033] Further, based on the optimal load distribution sequence, the total heating demand power at each time in the future period of time is calculated to obtain a total demand power sequence, including: According to the heating terminal demand temperature, the building area, and the heat load index at each future time, the theoretical heating demand power is calculated; Combined with the load distribution proportion in the optimal load distribution sequence, the PI controller model in the system outputs the total heating demand power at each time in the future period of time to form a total demand power sequence, and the corresponding expression is as follows: ; wherein, is the total demand power, is the proportional coefficient, is the difference between the heating terminal demand temperature and the actual temperature, is the integral coefficient.

[0034] Further, the total demand power sequence is used to determine the solar heat supply system output sequence and the ground source heat pump system output sequence with the minimum system energy consumption as the target, and the following cost function is used for control: ; ; Wherein, is the instantaneous energy consumption of the ground source heat pump, is the instantaneous energy consumption of the solar heat supply system, and c is an equivalent factor, is the solar heat supply system output sequence, is the ground source heat pump system output, is the equivalent heat value of the solar heat supply system, represents the state of charge of the energy storage device at the initial time of the cycle, U is the solar heat supply system output sequence in the future period of time, and l represents the optimization objective function, and J represents the cost function.

[0035] In the specific embodiment, when the solar heat supply system output sequence and the ground source heat pump system output sequence are determined, the total demand power sequence in the future period of time is used to construct an optimal control problem, the optimal action sequence is solved, and the solar heat supply system output sequence and the ground source heat pump system output sequence in the future period of time are determined, and the specific process includes: Step 1: Selection of state variables ; Selection of control variables ; The optimal control expression is established as follows: ; Wherein, is the capacity of the energy storage device, is the charging and discharging power of the energy storage device; Meanwhile, the following constraints are met: ; Wherein, , , , respectively represent the minimum output of the solar heat supply system, the maximum output of the solar heat supply system, the minimum output of the ground source heat pump, and the maximum output of the ground source heat pump; The sum of the solar heat supply system output and the ground source heat pump output of the coupled heat supply system in the future period of time is equal to the total demand power, that is, ; The cost function J is minimized to minimize the energy consumption of the coupled heat supply system in the future period of time; When solving, the solar heat supply system output, the ground source heat pump output, and the state of charge of the energy storage device must meet the respective ranges.

[0036] Step two: the above problem is transformed into the following general form: Optimization objective function and state transition equation are not typical quadratic forms, but satisfy the second order differentiable and contain inequality constraint functions about control and state variables and ; When solving, the optimization objective function is Taylor expanded, the state transition equation is linearized, and the inequality constraint function is converted into a new optimization objective function by the barrier function method L , the problem is transformed into a linear quadratic regulation problem, and the optimal action sequence of the solar heating system and the ground source heat pump in the future period is solved iteratively: Barrier function: ; ; Wherein, , , and are the first preset parameter, the second preset parameter, the third preset parameter and the fourth preset parameter respectively; thus the following new optimization objective function is obtained L : .

[0037] Step three: in the process of solving, through the new optimization objective function, the output of the solar heating system and are optimized to obtain the required output sequence of the solar heating system and the output sequence of the ground source heat pump system.

[0038] When the optimal heating load distribution model is used to output actions according to the input state space parameters, the steps are as follows: Under any initial action sequence, the initial state quantity is generated according to the forward recursion equation of the model, the initial objective function in the loop is obtained, the iteration step and the regularization coefficient are set, then a new series of control rates are obtained by the backward pass of the model, and the forward pass of the model is performed again, the new state quantity is calculated according to the control rate, and the new value is calculated; If the new value is greater than or equal to , the regularization coefficient is appropriately increased, and is set, and the backward pass and the forward pass are repeatedly executed. If new Value less than Then update the state space sequence, action space sequence, and objective function value; Let X = X NEW U=U NEW , ; where X NEW U=U NEW This represents the new state space sequence and action space sequence calculated after a new cycle; Increase the regularization coefficient, then make a judgment. - If the absolute value difference is less than the preset difference tol, then the action sequence and state sequence are output; otherwise, the backward and forward methods are executed repeatedly.

[0039] This embodiment fully considers the impact of real-time changes in heating environment information and system status information on the energy consumption of the coupled heating system. By combining the Internet of Things and environmental monitoring system, a more accurate load distribution scheme can be obtained, thereby achieving more precise and efficient heating control and better energy management of the coupled heating system.

[0040] In a specific embodiment, a Transfer Neural Network (TNN) can be introduced into the core process of determining the output sequence of the solar heating system and the ground source heat pump system based on the total demand power sequence. This effectively solves the problems of scarce data in new scenarios, poor adaptability across operating conditions, and slow iterative convergence of traditional optimization methods. By transferring optimization experience from existing scenarios, the solution of the optimal action sequence in new scenarios is accelerated, while improving the optimization accuracy under complex operating conditions. The specific optimization scheme is as follows: The optimized process consists of four stages: migration initialization, constraint adaptation, iterative solution, and result verification, as detailed below: Phase 1: Initialize the transfer neural network and obtain the initial output allocation scheme. (1) Constructing a transfer neural network structure The transfer neural network adopts a three-layer structure of feature extraction, domain adaptation, and output mapping. 1) Feature extraction layer: Input the total demand power sequence of similar historical scenes Environmental feature vectors (Including the average ambient temperature of historical scenes, the range of solar irradiance, and the trend of soil temperature changes), common features are extracted through a 2-layer fully connected network (ReLU activation function). (e.g., the power output ratio of ground source heat pumps in high-demand-low-solar-irradiance scenarios). 2) Domain adaptation layer: introduce domain adaptive loss function (Domain Adaptive Loss) to minimize the distribution difference between the current scene environment features (real-time environmental temperature, irradiance intensity, soil temperature of the current scene) and historical scene features The formula is as follows: ; Where N is the feature dimension, is the number of historical data samples, is the number of accumulated data samples in the current scene; 3) Output mapping layer: based on common features and adapted environment features, output the initial solar power output sequence and the initial ground source heat pump output sequence of the current scene, to ensure that the initial scheme meets the basic constraints .

[0041] (2) Similar scene screening rules

[0042] To ensure the effectiveness of the transferred experience, the historical similar scenes are screened by feature similarity threshold: calculate the cosine similarity between the current scene environment features and the historical scenes : ; Only historical scene data with ≥0.75 is retained as the transfer source, to avoid experience failure due to too large scene difference.

[0043] Phase 2: constraint condition transfer and adaptation

[0044] In the traditional step, constraints such as solar maximum output and ground source heat pump COP threshold need to be manually reset according to the current scene, which is prone to parameter adaptation deviation. After introducing the transfer neural network, the automatic adaptation of constraint parameters can be realized in the following ways: (1) constraint parameter transfer: extract the total demand power interval-constraint parameter correspondence from similar historical scenes, when , as the initial value of the constraint parameter of the current scene; (2) real-time adaptive adjustment: combine the current scene device status, such as real-time efficiency of solar collector and real-time COP value of ground source heat pump , and through the TNN output correction coefficient λ, fine-tune the initial constraint parameter: ;​ ; wherein, , is the average parameter of the device in similar historical scenarios, ensuring that the constraint parameter matches the actual performance of the current device.

[0045] Stage 3: Iterative solution based on DDPG-TNN cooperation

[0046] Based on the initial output sequence of the migrated neural network and the adapted constraint parameter, precise iterative optimization is performed through the DDPG model, with the following core improvements: (1) Introduce transfer trust weight into cost function To balance the influence of transfer experience and current data, introduce transfer trust weight into the original cost function, and adjust the formula as follows: ; wherein, is the deviation loss of the current iteration output sequence and the migrated initial sequence, and the formula is: ; Dynamic adjustment rule of <100) = 0.6 (prefer to trust transfer experience), and as the data accumulates ( ≥ 500) decreases to 0.1 (prefer to rely on current data), avoiding the interference of transfer experience on later optimization.

[0047] (2) Accelerate iterative convergence

[0048] Since the initial output sequence is close to the optimal solution (guided by transfer experience), the number of iterations of the DDPG model is reduced from the traditional 10-15 rounds to 3-5 rounds, and the step length of each iteration can be appropriately increased (from 0.01 to 0.03), while TNN is used to exclude obvious output combinations that do not meet the constraints in advance (such as , reducing invalid iteration calculation, and improving convergence efficiency by more than 60%.

[0049] Stage 4: Optimization result verification and transfer update

[0050] After each iteration solution, the following two steps are used to ensure the effectiveness of the results and update the experience library of the transfer neural network: Result verification: verify whether the final output sequence meets the minimum energy consumption goal (energy consumption reduction ≥ 8% compared with the initial scheme) and device safety constraints (such as , the heating quality constraint (heating end temperature deviation ≤±2℃), if not met, re-call TNN to adjust the initial scheme, and iterate again; Experience update: the total demand power sequence-optimal output sequence-environmental characteristics of the current scene are added to the historical experience library as new samples, and the network parameters are updated online every 24 hours to continuously improve the transfer optimization effect of subsequent scenes.

[0051] This embodiment realizes (1) data dependence reduction: in the early stage of a new scene (less than 50 data), a reliable initial optimization scheme can still be output through the transfer of experience, avoiding the dilemma that traditional methods cannot optimize without data, and shortening the system debugging period; (2) optimization precision improvement: under cross-condition scenes (such as extremely low temperature weather in winter), energy consumption optimization error is reduced, and the running time length ratio of the high-efficiency interval of the ground source heat pump is improved; (3) enhanced adaptive capability: the whole process of scene switching-constraint adaptation-optimization solution can be completed without human intervention, which is especially suitable for centralized heating management of multi-region and multi-type buildings, and the operation and maintenance flexibility is significantly improved.

[0052] In the specific embodiment, a solar-geothermal heat pump coupled heating system based on machine learning includes: A model construction module is configured to construct a heating load distribution model. A training module is configured to obtain real-time data, input the real-time data into the heating load distribution model, and obtain an optimal heating load distribution scheme at each time point in a future period of time to form an optimal load distribution sequence. A prediction module is configured to calculate total heating demand power at each time point in a future period of time based on the optimal load distribution sequence to obtain a total demand power sequence. A constraint module is configured to determine a solar heating system output sequence and a ground source heat pump system output sequence based on the total demand power sequence and taking the minimum system energy consumption as the target. A control module is configured to control the coupled heating system based on the solar heating system output sequence and the ground source heat pump system output sequence.

[0053] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0054] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A machine learning-based method for coupled solar and ground-source heat pump heating, characterized in that, include: A heating load allocation model is constructed, real-time data is obtained and input into the heating load allocation model to obtain the optimal heating load allocation scheme for each moment in the future period, forming the optimal load allocation sequence. Based on the optimal load allocation sequence, the total heating demand power at each moment in the future period is calculated to obtain the total demand power sequence. Based on the total demand power sequence, with the goal of minimizing system energy consumption, the output sequence of the solar heating system and the output sequence of the ground source heat pump system are determined. The coupled heating system is controlled based on the output sequence of the solar heating system and the output sequence of the ground source heat pump system.

2. The solar and ground source heat pump coupled heating method based on machine learning according to claim 1, characterized in that, The construction of the heating load distribution model includes: A dual-delay deep deterministic policy gradient neural network model is initialized. The state space of the dual-delay deep deterministic policy gradient neural network model is constructed with the current ambient temperature, solar irradiance, soil temperature, heating terminal demand temperature, system energy storage device state of charge, solar collector efficiency, and ground source heat pump COP value. The action space is constructed with the output ratio of the solar heating system. The reward function is constructed from the system total energy consumption reward function, heating temperature deviation reward function, solar energy utilization rate reward function, and equipment operating efficiency reward function. After training, the Actor target network is used as the optimal heating load allocation model.

3. The solar and ground source heat pump coupled heating method based on machine learning according to claim 2, characterized in that, The dual-delay deep deterministic strategy gradient neural network model includes: The expression for the dual-delay deep deterministic strategy gradient neural network model is as follows: ； Where S represents the state space, A represents the action space, and R represents the reward function; ； in, This indicates the current ambient temperature. This indicates the current solar irradiance. This indicates the current soil temperature. This indicates the current temperature demand at the heating terminal. This indicates the current state of charge of the system's energy storage device. This indicates the current efficiency of the solar collector. This indicates the COP value of the ground source heat pump at the current moment; ； in, This indicates the current percentage of output from the solar heating system. ； in, The weighting factor represents the total energy consumption reward function of the system per unit time. The weighting factor of the reward function representing the heating temperature deviation per unit time. The weighting factor of the reward function representing the solar energy utilization rate per unit time. The weighting factor represents the reward function for equipment operating efficiency per unit time. It is the system's total energy consumption reward function per unit time. It is the reward function for the deviation of heating temperature per unit time. It is the reward function for solar energy utilization rate per unit time. It is the equipment operating efficiency reward function per unit time.

4. The solar and ground source heat pump coupled heating method based on machine learning according to claim 3, characterized in that, The total system energy consumption reward function per unit time is the sum of the energy consumption of the solar heating system and the energy consumption of the ground source heat pump system within a unit time step.

5. The solar and ground source heat pump coupled heating method based on machine learning according to claim 3, characterized in that, The expression for the heating temperature deviation reward function per unit time is as follows: ； in, , These represent the maximum and minimum allowable water supply temperatures at the heating terminal, respectively. This represents the actual water supply temperature at the current moment, and exp represents the exponential function.

6. The solar and ground source heat pump coupled heating method based on machine learning according to claim 3, characterized in that, The expression for the solar energy utilization reward function per unit time is as follows: ； in, denoted by , where represents the threshold for solar energy utilization, and k represents the incentive coefficient for solar energy utilization.

7. The solar and ground source heat pump coupled heating method based on machine learning according to claim 3, characterized in that, The expression for the equipment operating efficiency reward function per unit time is as follows: ； in, represents the operating efficiency threshold of the ground source heat pump, and m represents the equipment efficiency bonus coefficient.

8. The solar and ground source heat pump coupled heating method based on machine learning according to claim 1, characterized in that, The process of calculating the total heating demand power at each moment within a future period based on the optimal load allocation sequence, resulting in the total demand power sequence, includes: The theoretical heating demand power is calculated based on the heating terminal demand temperature, building area, and heat load index at each future moment. By combining the load allocation ratio in the optimal load allocation sequence, the total heating demand power at each moment in the future is output through the PI controller model in the system, forming the total demand power sequence.

9. A machine learning-based solar and ground source heat pump coupled heating method according to claim 1, characterized in that, Based on the total demand power sequence, and with the goal of minimizing system energy consumption, the output sequences of the solar heating system and the ground source heat pump system are determined, and controlled using the following cost function: ； ； in, For the instantaneous energy consumption of a ground source heat pump, The instantaneous energy consumption of the solar heating system is given by c, where c is the equivalence factor. For the power output sequence of the solar heating system, To provide power for the ground source heat pump system, The equivalent calorific value of solar heating U represents the state of charge of the energy storage device at the beginning of the cycle, U is the output sequence of the solar heating system over a future period, l represents the objective function, and J represents the cost function.

10. A solar-ground source heat pump coupled heating system based on machine learning, characterized in that, include: Model building module: Used to build heating load distribution models; Training module: Used to acquire real-time data, input it into the heating load allocation model, obtain the optimal heating load allocation scheme for each moment in the future period, and form the optimal load allocation sequence; Prediction module: Used to calculate the total heating demand power at each time point in the future based on the optimal load allocation sequence, and obtain the total demand power sequence; Constraint module: Used to determine the output sequence of solar heating system and ground source heat pump system based on total demand power sequence, with the goal of minimizing system energy consumption; Control module: Used to control the coupled heating system based on the output sequence of the solar heating system and the output sequence of the ground source heat pump system.

Images