Two-stage optimization method for household wind and light storage system oriented to thermal comfort and energy efficiency collaboration

Abstract

The invention relates to the technical field of intelligent building and household energy management, and provides a household wind and light storage system two-stage optimization method for thermal comfort and energy efficiency collaboration, and the method comprises the following steps: S1, constructing a household wind and light storage system high-fidelity model, and precisely representing the coupling relation and operation constraint of electric and thermal multi-energy flow; s2, constructing a multi-objective optimization problem taking minimization of daily electricity consumption cost, maximization of renewable energy source self-use rate and precise maintenance of thermal comfort as the core; and S3, designing and solving a two-stage collaborative optimization framework: decomposing a global optimization problem into two successive stages of day-ahead scheduling and real-time control to realize efficient solving. According to the method, the source-load uncertainty is effectively stabilized, the power consumption cost is reduced, the renewable energy sources are efficiently consumed, the indoor thermal comfort degree is maintained at high precision, and the performance is remarkably superior to that of a traditional method.

Description

Technical Field

[0001] This invention relates to the field of intelligent building and home energy management technology, and in particular to an operation optimization method for an integrated solar, wind and energy storage system for residential buildings. Background Technology

[0002] The green and low-carbon transformation of the energy structure is a core issue of global sustainable development. As a major end-use energy consumer, the building sector's energy conservation and emission reduction are crucial for achieving carbon neutrality. Against this backdrop, home energy systems integrating solar photovoltaic, wind power, battery storage, and thermal energy storage are considered an effective way to improve energy self-sufficiency and reduce carbon emissions. By coordinating the production, storage, conversion, and consumption of energy within the system, home energy management systems can significantly improve operational economy and energy efficiency while meeting user needs.

[0003] However, such systems face multiple challenges in actual operation. First, renewable energy sources such as photovoltaic and wind power have significant intermittency and randomness, while residential loads (especially heating, ventilation, and air conditioning system loads) are highly dynamic and uncertain. This mismatch between the source and load sides poses significant difficulties for the stable and economical operation of the system. Second, there are complex electro-thermal multi-energy flow coupling relationships within the system. Electrical and thermal energy influence each other in the generation, storage, and consumption stages, making accurate modeling and coordinated scheduling difficult. Furthermore, users' demands for indoor thermal comfort are increasing, requiring a delicate trade-off between reducing energy consumption and ensuring comfort.

[0004] Currently, optimization methods for home energy management mainly include rule-based control, traditional optimization algorithms, and model predictive control (DRC). Rule-based control methods are logically simple and easy to implement, but lack intelligence and adaptability, making them difficult to handle dynamically changing environments. Traditional optimization algorithms (such as linear programming and mixed-integer programming) can find optimal solutions under specific models, but their performance heavily depends on the accuracy of the model. Simplifying highly nonlinear factors such as user behavior and equipment characteristics in home energy systems leads to performance degradation in real-world scenarios. Model predictive control employs a rolling optimization strategy, which can handle uncertainty to some extent, but it has a heavy computational burden and requires extremely high accuracy from the prediction model, making real-time application in complex home energy systems difficult. In recent years, artificial intelligence technologies such as deep reinforcement learning (DRL) have provided new ideas for solving these problems. Their model-free nature and long-term optimization perspective make them very suitable for handling high-dimensional, nonlinear decision-making problems in home energy management. However, existing DRL methods mostly focus on optimization at a single time scale, either performing only macro-level day-ahead scheduling or only minute-level real-time control, lacking an effective mechanism to organically combine forward-looking planning with real-time rapid response. This disconnect prevents the system from conducting global energy planning while dealing with real-time uncertainties, thus limiting further improvements in the overall system performance.

[0005] In summary, existing optimization methods for residential wind, solar, and energy storage management systems mainly suffer from the following three problems:

[0006] (1) Challenges of insufficient model accuracy and strong coupling of multiple energy flows;

[0007] (2) Lack of multi-timescale collaborative optimization mechanism;

[0008] (3) Weak multi-objective collaborative optimization capability;

[0009] Therefore, how to construct an efficient and robust home energy management framework that can synergistically optimize system economy, energy efficiency, and multi-regional thermal comfort under the dual uncertainties of renewable energy and load has become a key issue that urgently needs to be addressed in this technical field. Summary of the Invention

[0010] This invention addresses the technical challenges faced by existing home energy management methods in dealing with high-proportion renewable energy access and diversified user needs, such as insufficient model accuracy, lack of multi-timescale collaborative optimization mechanisms, and weak multi-objective (economic, energy efficiency, and comfort) collaborative optimization capabilities. It proposes a two-stage optimization method for home wind-solar-storage systems that focuses on the synergy between thermal comfort and energy efficiency. This method solves the multi-timescale coordination problem through a two-stage collaborative optimization of "day-ahead scheduling-real-time control," achieving multi-objective collaborative optimization of energy efficiency, economics, and comfort.

[0011] The two-stage optimization method for a home solar-wind storage system that aims to achieve synergy between thermal comfort and energy efficiency, as proposed in this invention, includes the following three sequential and closely related core steps.

[0012] S1. Construct a high-fidelity model of a residential solar-powered energy storage system:

[0013] An integrated model is established, including photovoltaic, wind turbine, battery energy storage system, thermal storage tank, heat pump and multi-zone HVAC system, to accurately characterize the coupling relationship and operational constraints of electric and heat multi-energy flow in the entire chain of generation, storage, conversion and consumption, and to provide an accurate physical basis for subsequent optimization.

[0014] This step is the cornerstone for subsequent problem definition and solution.

[0015] The high-fidelity model in step S1 specifically includes:

[0016] (1) Output power P of the solar photovoltaic power generation model pv,t (KW) is determined by the actual light intensity G a,t (W / m 2 ) and photovoltaic panel operating temperature T a,t (°C) is jointly determined, specifically as follows:

[0017] ;

[0018] Where, N pv It is the number of photovoltaic panels, η pv It is the attenuation coefficient, P pv This refers to the rated power (kW) of the photovoltaic panel under standard operating conditions. pv It is the power temperature coefficient, T ref It is the standard ambient temperature (°C), G ref Standard solar irradiance (W / m²) 2 );

[0019] (2) Output power P of the wind power generation model wt,t Wind speed V at the wheel hub w,t The decision is specifically expressed as follows:

[0020] ;

[0021] Among them, V in V ref V out These are the inlet velocity, rated velocity, and outlet velocity (m / s) of the fan, respectively. wt It is the rated power (KW) of the fan;

[0022] (3) State of charge (SOC) of the battery energy storage model t Represented as:

[0023] ;

[0024] Where, η ess It refers to the energy utilization efficiency of battery energy storage systems, E ess It is the energy storage system capacity (kWh), P ess,t It is the charging and discharging power (kW) of the energy storage system, and the State of Charge (SOC). t-1 It is the state of charge of the battery energy storage model at the previous time step (t-1), and Δt is the length of each time step (15 min).

[0025] Its operation must meet power constraints and state of charge constraints:

[0026] ;

[0027] in, and These are the maximum allowable charging power and maximum discharging power (kW) of the battery energy storage system, and the State of Charge (SOC). max and SOC min These are the maximum and minimum values ​​of the state of charge;

[0028] (4) The residual heat Q of the thermal storage tank model tst,t Represented as:

[0029] ;

[0030] Among them, Q tst,t-1 It is the heat energy (kWh) stored in the thermal storage tank at the previous moment, α tst It is the heat exchange coefficient between the thermal storage tank and the environment, η tst It is the energy utilization efficiency of the thermal storage tank, P tst,t It is the heat storage or heat release power (KW) of the heat storage tank at time t;

[0031] Its operation must meet power constraints and heat constraints:

[0032] ;

[0033] in and These are the maximum allowable heat storage capacity and maximum heat release capacity (kW) of the thermal storage tank, respectively. and These are the maximum and minimum values ​​(kWh) of the remaining energy in the thermal storage tank, respectively.

[0034] (5) Heating or cooling power Q of the heat pump model hp,t Its power consumption P hp,t (kW) and coefficient of performance (COP) tRelated, specifically:

[0035] ;

[0036] Where λ c1 , λ c2 and λ c3 T is a fitting parameter related to the physical characteristics of a heat pump. out,t It is the ambient temperature (°C) when the heat pump is working.

[0037] (6) The electrical load model includes multiple electrical devices, and the operating power of the i-th electrical load at time t is... (KW) is represented as:

[0038] ;

[0039] in, This indicates the switching state of the i-th electrical load at time t, where 0 represents off and 1 represents on; This is the rated operating power (kW) of the load.

[0040] (7) A thermal dynamic model of a multi-zone HVAC system describes the temperature T of each zone at time t. i,t (°C), and the heat exchange Q between the building envelope and the outdoor environment. i,t out The heat exchange Q between adjacent regions i,t adj The cooling capacity Q supplied to the area by the multi-zone HVAC system i,t HVAC And the internal heat Q generated by people and equipment inside the room. i,t in The correlation is defined by the following system of equations:

[0041] ;

[0042] Where ρ is the air density (kg / m³), C ρ It is the specific heat capacity of air at constant pressure (kJ / (kg·K)), V i A is the volume (m³) of region i. ZONEi and A i,s These are the contact area between the region and the outside world, and the contact area with adjacent regions (m²), respectively. s It is the number of adjacent regions, T out,t and T s,t These are the outdoor temperature at time t and the temperature of the adjacent area (°C), respectively. C It is the supply air temperature (°C), q l It is the infiltration air volume ratio (kg / s), Body m and Elec nThese are the heat generation power (W) of personnel and equipment, respectively, and N. m and N n These refer to the number of personnel and equipment, U trans1 It is the heat transfer coefficient between the area and the outdoors, U trans2 It is the heat transfer coefficient (W / (m²·K)) between a region and its adjacent regions, T i,t T is the temperature (°C) of region i at time t. i,t+1 m is the temperature (°C) of region i at the previous time (t+1). i,t It is the air supply volume for supply area i.

[0043] S2. Construct a multi-objective collaborative optimization problem:

[0044] With the core optimization objectives of minimizing daily electricity costs, maximizing the self-consumption rate of renewable energy, and accurately maintaining thermal comfort in multiple regions, the system defines the electrical power balance constraints, thermal power balance constraints, equipment operation boundary constraints, and comfort range constraints that the operation of a residential wind-solar-storage system must meet, forming a complete mathematical description of the optimization problem.

[0045] This step clarifies the optimization objective and boundary conditions.

[0046] The multi-objective collaborative optimization problem in step S2 is specifically defined as follows:

[0047] (1) Define the first optimization objective with economy as the core, that is, minimize the total daily operating cost C of the system. total Its composition consists of the cost of purchasing electricity from the grid, expressed as:

[0048] ;

[0049] Among them, P grid,t C is the power exchanged with the grid at time t (kW, positive for purchasing electricity, negative for selling electricity). grid,t It is the time-of-use electricity price (CNY / kWh) at time t;

[0050] (2) Define the second optimization objective with energy efficiency as the core, namely, maximizing the self-consumption rate of renewable energy R. SCR It represents the proportion of renewable energy generation that meets the total system load, and its expression is:

[0051] ;

[0052] Among them, P pv-eload,t P pv-hp,t P wt-eload,t and P wt-hp,t These represent the power (kW) supplied by the photovoltaic and wind turbines and the power of the heat pump at time t, respectively. eload,t P is the electrical load power (kW) at time t.hp,t It is the heat pump power (KW) at time t;

[0053] (3) Define the power balance constraints that the residential solar-wind-storage system must meet to ensure that the power supplied by all power sources is equal to the power of all loads at time t:

[0054] ;

[0055] Among them, P pv,t The photovoltaic power (kW) at time t, P wt,t The power of the wind turbine at time t (kW), P grid,t P is the power purchased by the power grid at time t (kW). ess,t The charge / discharge power (kW) of the battery energy storage system at time t, P hp,t It is the heat pump power (KW) at time t. It is the power (kW) of the i-th load at time t. It is the power (KW) of the fans in a multi-zone HVAC system;

[0056] (4) Define the heat power balance constraints that the residential solar-wind storage system must meet to ensure that the heat pump output power is balanced with the heat load and the power of the heat storage tank at time t:

[0057] ;

[0058] in, It is the power (kW) required by a multi-zone HVAC system to meet the heat load, P tst,t It is the thermal power (kW) provided by the thermal storage tank at time t;

[0059] (5) Define a third optimization objective centered on comfort, ensuring that the temperature T in each region i at time t is constant. i,t Maintain within the comfort zone:

[0060] ;

[0061] in, and These represent the upper and lower limits (°C) of the thermal comfort range, respectively.

[0062] (6) Define the start-stop time constraints for electrical loads:

[0063] ;

[0064] in, and The earliest start time and latest end time t of device i eload,i It is the device's on-time (i) and (h). eload,i It is the runtime of device i;

[0065] The multi-objective collaborative optimization problem is to collaboratively optimize the first optimization objective, the second optimization objective, and the third optimization objective under the premise of satisfying the above-mentioned constraints on power balance, heat power balance, and power load start-up and shutdown time.

[0066] S3. Design and solve a two-stage collaborative optimization framework:

[0067] The global optimization problem is decomposed into two interconnected stages: day-ahead scheduling and real-time control, to achieve efficient solution.

[0068] This step is the core of the method of this invention, which solves the problem of solving complex optimization problems through a time-scale strategy.

[0069] The specific process of step S3 includes:

[0070] (1) Day-ahead scheduling phase: Based on the renewable energy output and load forecast for the next 24 hours, a multi-objective genetic algorithm (NSGA-II) is used to optimize the start-up and shutdown times of dispatchable loads (such as electric vehicles, washing machines, etc.) with economy and energy efficiency as the primary objectives, and to generate a macro-level day-ahead energy scheduling plan;

[0071] This phase focuses on strategic energy planning, with a time resolution of 1 hour, and is implemented as follows:

[0072] (a) The scheduling time scale is set to 24 hours, with 1 hour as the basic scheduling period, i.e., T=24, Δt=1h;

[0073] (b) The optimization variable is the start-stop time series of each schedulable load:

[0074] ;

[0075] in, It is a binary variable representing whether the i-th schedulable load runs during time period t, N S This represents the total number of schedulable loads;

[0076] (c) The optimization variables are solved using a non-dominated sorting genetic algorithm (NSGA-II) with an elitist strategy. The fitness function of this algorithm is the multi-objective collaborative optimization problem defined in step S2, i.e., simultaneously optimizing the total daily operating cost C. total and renewable energy self-consumption rate R SCR ;

[0077] (d) The running parameters of the non-dominated sorting genetic algorithm with elitist strategy include: initializing the population size N. p =100, maximum number of generations I max =200, crossover probability P c=0.8, mutation probability P m =0.1;

[0078] (e) The constraints that the day-ahead scheduling phase must satisfy include:

[0079] (i) Scheduled load runtime window constraint: The actual runtime of each scheduleable load i must fall within its allowed working time range. Inside:

[0080] (ii) System power balance constraints: The power balance constraints defined in step S2 must be satisfied in each scheduling period t;

[0081] (iii) Equipment operation boundary constraints: The operation boundary constraints of the battery energy storage system, thermal storage tank and electrical load defined in steps S1 and S2 must be satisfied in each scheduling period t;

[0082] (f) The non-dominated sorting genetic algorithm with elite strategy continuously evolves the population through selection, crossover, and mutation genetic operations, and finally outputs a set of Pareto optimal solutions. Each solution represents a start-stop time scheme for a schedulable load. From the Pareto optimal solution set, the day-ahead scheduling plan is selected according to actual preferences as the macro-energy planning benchmark for the real-time control stage.

[0083] (2) Real-time control stage:

[0084] Within the macro framework of the recently generated scheduling plan, the charging and discharging power P of the battery energy storage system is analyzed using the maximum entropy deep reinforcement learning algorithm (SAC). ess,t The heat storage and release power P of the heat storage tank tst,t And the change in air supply volume Δm in each area of ​​the multi-zone HVAC system. i,t Perform rolling optimization and fine-tuning at the minute level (e.g., 15 minutes). This stage focuses on tactical real-time response to address uncertainties such as prediction errors, with the core objective of ensuring thermal comfort and further reducing operating costs. The SAC algorithm encourages exploration through its maximum entropy principle, effectively learning the optimal stochastic strategy in complex dynamic environments.

[0085] The specific implementation method of the real-time control stage is as follows:

[0086] (a) The control time scale is set to 15 minutes, i.e. the control step size Δt = 0.25 h. At each control time k, the agent interacts with the environment once.

[0087] (b) The real-time control problem is modeled as a Markov decision process (MDP), which includes the following elements:

[0088] (i) State space S: the state s at time kk ∈S includes renewable energy generation P pv,t and P wt,t Battery energy storage system state of charge (SOC) ess,t and the remaining heat Q of the thermal storage tank tst,t Electrical load power P eload,t Time-of-use electricity price C grid,t Indoor temperature T in each area i,t Outdoor ambient temperature T out,t Area occupancy status (Body) m and equipment heating power Elec n This can be expressed as a formula:

[0089] ;

[0090] (ii) Action space A: Action a at time k k ∈A includes the change in air supply volume Δm in each area. i,t Battery energy storage system charging and discharging power P ess,t And the heat storage and release power P of the thermal storage tank tst,t This can be expressed as a formula:

[0091] ;

[0092] (iii) Reward function R: reward r t This is a multi-objective weighted sum designed to guide the agent to simultaneously optimize thermal comfort, penalties for violating energy storage operation rules, and operating costs.

[0093] ;

[0094] Where r 1,t It is a penalty related to regional thermal comfort, when temperature T i,t Beyond the comfort zone Punishment will be imposed at that time:

[0095] ;

[0096] r 2,t This is a penalty for energy storage operations during non-economic periods, preventing battery energy storage and thermal storage tanks from discharging during off-peak electricity prices or charging during peak electricity prices, when the electricity price is lower than... It is off-peak electricity pricing, and the electricity price is higher than [previous price]. Peak electricity pricing:

[0097] ;

[0098] r 3,t and r 4,t These are the operating costs of thermal energy and electrical energy:

[0099] ;

[0100] .

[0101] (3) The soft actor-critic (SAC) algorithm is used as the core optimizer for solving the MDP;

[0102] The soft actor-critic (SAC) algorithm comprises the following core components and processes:

[0103] (i) The policy network (actors) is parameterized as Its optimization objective is to maximize the weighted sum of expected cumulative reward and policy entropy:

[0104] ;

[0105] Where α is a temperature parameter used to balance the reward term and the entropy term; It is the policy entropy, defined as:

[0106] ;

[0107] (ii) Two soft Q-function networks (commentator) Update by minimizing the soft Bellman residual:

[0108] ;

[0109] Where D is the experience replay buffer; γ is the discount factor; These are the target network parameters; It is the target soft state value function, calculated through the target Q-network and the policy network:

[0110] ;

[0111] (iii) The policy network (actors) is updated by minimizing the following objective function:

[0112] ;

[0113] In actual updates, the reparameterization technique is used to represent the action as a Gaussian distribution. To reduce variance;

[0114] (iv) The temperature parameter α is automatically adjusted by minimizing the following objective function:

[0115] ;

[0116] in, It is the target entropy, usually set to -dim(A) (a negative number of the action space dimension);

[0117] (v) Target Q-network parameters Slowly track the current Q network parameters using a soft update method:

[0118] ;

[0119] in The update step size for the target soft Q network.

[0120] (4) The key hyperparameter settings of the SAC algorithm include: learning rate λ Q , λ π , λ α =3×10 -4 Discount factor γ = 0.99, experience replay cache size |D| = 2 × 10 5 The target network update rate τ = 0.005, and the minimum batch size is 256.

[0121] (5) The constraints that the real-time control stage needs to meet include the electric power balance constraints, thermal power balance constraints, thermal comfort constraints and equipment operation boundary constraints defined in step S2. These constraints are reflected and guaranteed by the penalty term in the reward function and the action pruning method.

[0122] Compared with the prior art, the outstanding substantive features and significant progress of this invention are mainly reflected in the following three aspects:

[0123] (1) A two-stage collaborative optimization framework of "day-ahead scheduling-real-time control" is proposed: This framework uses the NSGA-II algorithm to perform day-ahead macro-load scheduling, generate an economical and efficient energy plan, and realize forward-looking energy planning; then, it uses the SAC algorithm for real-time fine control, which can quickly respond to uncertainties and ensure the real-time optimality of the system. The two stages are organically coordinated, which effectively overcomes the limitation of the single time scale optimization method of "paying attention to one thing but losing another", and realizes the optimization of global performance.

[0124] (2) A high-fidelity electro-thermal multi-energy flow coupling model was established and a true multi-objective collaborative optimization was achieved: The constructed model meticulously depicts the coupling relationship of the entire chain from energy generation, storage, conversion to consumption, providing a reliable physical basis for optimization. The optimization objectives cover multiple dimensions such as economic cost, renewable energy consumption and thermal comfort, and effectively balance multiple objectives through weighted reward functions, achieving a breakthrough in the overall benefits of the system.

[0125] (3) A high-fidelity joint simulation platform was constructed for verification, which significantly improved the reliability and persuasiveness of the results: Through the linkage of multiple software such as C++ (key component model) / Python (intelligent agent control) / EnergyPlus (high-precision simulation of building HVAC), high-precision modeling and closed-loop control simulation of complex home energy systems were realized, providing solid data support for the effectiveness and feasibility of the algorithm and laying a solid foundation for subsequent industrial applications.

[0126] This invention constructs a high-fidelity electro-thermal coupling system model and designs a two-stage collaborative optimization framework of "day-ahead scheduling-real-time control." It achieves multi-objective collaborative optimization of household electricity economy, efficient renewable energy consumption, and multi-regional thermal comfort, systematically solving the challenge of coordinating economic, energy efficiency, and comfort optimization of household wind-solar-storage systems under uncertainty. Under the dual uncertainties of renewable energy and load, it achieves a significant reduction in daily operating costs, a substantial increase in renewable energy self-consumption rate, and high-precision maintenance of thermal comfort in multiple regions. Ultimately, it achieves comprehensive optimization and synergistic improvement of energy efficiency, economy, and comfort, providing an effective technical solution for the refined and intelligent operation of smart buildings and household energy management. Attached Figure Description

[0127] Figure 1 This is the overall flowchart of the two-stage optimization method for home solar-wind storage systems that addresses the synergy between thermal comfort and energy efficiency, as proposed in this invention.

[0128] Figure 2 This is a flowchart illustrating the implementation of the two-stage collaborative optimization algorithm in this invention. Detailed Implementation

[0129] This invention addresses the practical needs of energy optimization management in residential homes by proposing a two-stage optimization method for residential wind-solar-storage systems that balances thermal comfort and energy efficiency. By constructing a high-fidelity electro-thermal coupling model and designing a two-stage collaborative optimization framework of "day-ahead scheduling-real-time control," the method takes into account the economic efficiency of household energy use, the efficient utilization of renewable energy, and the thermal comfort of multiple indoor areas, providing a systematic solution to address the dual uncertainties of renewable energy output and load demand.

[0130] The two-stage optimization method of this invention comprises three core steps, logically related as follows: first, constructing an accurate system model as the basis for optimization; second, defining the multi-objective optimization problem to be solved; and finally, designing an efficient two-stage solution framework. This method systematically solves the key issues throughout the entire process from accurate modeling and problem definition to efficient solution.

[0131] Figure 1A flowchart illustrating the overall implementation of the two-stage optimization method described in this invention is provided, clearly demonstrating the complete technical path from system model establishment and optimization problem construction to two-stage collaborative solution, comprising three sequentially progressive and closely related core steps. The following, in conjunction with... Figure 1 The specific embodiments of the present invention will be described in detail below.

[0132] S1. Construct a high-fidelity model of a home-based solar-powered energy storage system.

[0133] (1) This embodiment takes a residential building in Shandong, China as the application scenario and constructs an integrated model that includes photovoltaic, wind turbine, battery energy storage system, thermal storage tank, heat pump and multi-zone HVAC system. The model parameters are set based on the actual equipment specifications and the coupling relationship of electric-thermal multi-energy flow is accurately represented by mathematical formulas.

[0134] ① Output power P of the solar photovoltaic power generation model pv,t (KW) is determined by the actual light intensity G a,t (W / m 2 ) and photovoltaic panel operating temperature T a,t (°C) is jointly determined, specifically as follows:

[0135] ;

[0136] Where, N pv It is the number of photovoltaic panels, η pv It is the attenuation coefficient, P pv This refers to the rated power (kW) of the photovoltaic panel under standard operating conditions. pv It is the power temperature coefficient, T ref It is the standard ambient temperature (°C), G ref Standard solar irradiance (W / m²) 2 );

[0137] ② Output power P of the wind power generation model wt,t Wind speed V at the wheel hub w,t The decision is specifically expressed as follows:

[0138] ;

[0139] Among them, V in V ref V out These are the inlet velocity, rated velocity, and outlet velocity (m / s) of the fan, respectively. wt It is the rated power (KW) of the fan;

[0140] ③ State of Charge (SOC) of the battery energy storage model t Represented as:

[0141] ;

[0142] Where, η ess It refers to the energy utilization efficiency of battery energy storage systems, E ess It is the energy storage system capacity (kWh), Δt is the charging and discharging time, and P is the energy storage system capacity (kWh). ess,t It is the charging and discharging power (kW) of the energy storage system.

[0143] Its operation must meet power constraints and state of charge constraints:

[0144] ;

[0145] in, and These are the maximum allowable charging power and maximum discharging power (kW) of the battery energy storage system, and the State of Charge (SOC). max and SOC min These are the maximum and minimum values ​​of the state of charge;

[0146] ④ The residual heat Q of the thermal storage tank model tst,t Represented as:

[0147] ;

[0148] Among them, Q tst,t-1 It is the heat energy (kWh) stored in the thermal storage tank at the previous moment, α tst It is the heat exchange coefficient between the thermal storage tank and the environment, η tst It is the energy utilization efficiency of the thermal storage tank, P tst,t It is the heat storage or heat release power (KW) of the heat storage tank at time t;

[0149] Its operation must meet power constraints and heat constraints:

[0150] ;

[0151] in and These are the maximum allowable heat storage capacity and maximum heat release capacity (kW) of the thermal storage tank, respectively. and These are the maximum and minimum values ​​(kWh) of the remaining energy in the thermal storage tank, respectively.

[0152] ⑤ Heating or cooling power Q of the heat pump model hp,t Its power consumption P hp,t (kW) and coefficient of performance (COP) t Related, specifically:

[0153] ;

[0154] Where λ c1 , λc2 and λ c3 These are fitting parameters related to the physical characteristics of the heat pump.

[0155] ⑥ The electrical load model includes multiple electrical devices, and the operating power of the i-th electrical load at time t is... (KW) is represented as:

[0156] ;

[0157] in, This indicates the switching state of the i-th electrical load at time t (0 for off, 1 for on). This is the rated operating power (kW) of the load;

[0158] ⑦ A thermal dynamic model of a multi-zone HVAC system describes the temperature T of each zone at time t. i,t (°C), and the heat exchange Q between the building envelope and the outdoor environment. i,t out The heat exchange Q between adjacent regions i,t adj The cooling capacity Q supplied to the area by the multi-zone HVAC system i,t HVAC And the internal heat Q generated by people and equipment inside the room. i,t in The correlation is defined by the following system of equations:

[0159] ;

[0160] Where ρ is the air density (kg / m³), C ρ It is the specific heat capacity of air at constant pressure (kJ / (kg·K)), V i U is the volume (m³) of region i. trans1 and U trans2 These are the heat transfer coefficients (W / (m²·K)) between the area and the outdoors, and between the area and adjacent areas, respectively. ZONEi and A i,s These are the contact area between the region and the outside world, and the contact area with adjacent regions (m²), respectively. s It is the number of adjacent regions, T out,t and T s,t These represent the outdoor temperature at time t and the temperature of the adjacent area (°C), respectively, and m. i,t T is the air volume mass flow rate (kg / s) of supply area i. C It is the supply air temperature (°C), q l It is the infiltration air volume ratio (kg / s), Body m and Elec n These are the heat generation power (W) of personnel and equipment, respectively, and N. m and Nn These refer to the number of personnel and the number of equipment, respectively.

[0161] (2) The configuration parameters of the core equipment of the system are set based on a typical home energy system, as follows:

[0162] ① Solar photovoltaic: The installed capacity is 2.8kW, and its output model depends on the real-time light intensity and photovoltaic panel temperature.

[0163] ② Wind turbine: The rated power is 2kW. Its output characteristics are determined by the wind speed at the hub height and follow the standard cut-in-rated-cut-out wind speed power curve.

[0164] ③ Battery energy storage system: Rated energy capacity is 10kWh, maximum charge and discharge power is 3kW, and the state of charge operation range is set to 20% to 90%.

[0165] ④ Thermal storage tank: Rated heat capacity is 20kWh, maximum heat storage / release power is 4kW, and the heat storage range is set to 2 to 18kWh.

[0166] ⑤ Heat pump: Its coefficient of performance (COP) is a function of the outdoor temperature. In this embodiment, the COPs for cooling and heating modes are respectively fitted to quadratic polynomials of the outdoor temperature. The specific fitting coefficients need to be determined based on experimental data of the specific heat pump model.

[0167] (3) To enable closed-loop simulation verification of systems involving complex building thermal processes, a co-simulation platform was built. The platform consists of three core software modules that work together through standardized interfaces to form a complete simulation verification environment.

[0168] ① High-precision building and HVAC simulation module (EnergyPlus): This module is responsible for performing the highest-fidelity simulation of building thermal dynamics and multi-zone HVAC systems. First, a geometric model of the target residence is created using SketchUp and OpenStudio software, defining detailed parameters such as building envelope materials, internal thermal disturbances (personnel, equipment), and HVAC system topology. Then, the model is exported as a Functional Mock-up Unit (FMU) conforming to the Functional Mock-up Interface (FMI) standard. During simulation, the main control program calls this FMU through interfaces such as PyFMI to obtain high-precision data on temperature, humidity, and HVAC equipment energy consumption in each zone at the minute level.

[0169] ② Equipment and System Model Module (C++ Dynamic Link Library): To improve simulation efficiency, the mathematical models of the photovoltaic, wind turbine, battery, thermal storage tank, heat pump, and basic electrical load are implemented in C++ and compiled into a dynamic link library (.dll) while ensuring physical consistency. This module receives control commands from the main control program (such as battery charging and discharging power, heat pump start / stop status), and calculates the equipment state (such as SOC, remaining heat) and output at the next moment based on the physical model and the current state. The calculation speed is fast, making it suitable for integrating optimization algorithms to perform a large number of forward simulations.

[0170] ③ Optimized Control and Simulation Main Program (Python): This module serves as the brain and coordinator of the entire simulation platform and is developed using the Python language. Its main functions include:

[0171] (a) Simulation flow control: The co-simulation is advanced at fixed time steps (e.g., 15 minutes). At each time step, the current system state is obtained from the C++ module and the EnergyPlus FMU.

[0172] (b) Algorithm Integration: The "two-stage collaborative optimization framework" described in this invention is fully implemented. The multi-objective genetic algorithm (NSGA-II) for the day-ahead scheduling stage and the deep reinforcement learning algorithm (SAC) for the real-time control stage are both coded and implemented in this module.

[0173] (c) Data interaction and command issuance: Based on the decisions generated by the optimization algorithm (such as the start and stop time of the schedulable load, the battery power setting value, and the opening degree of the air valves in each area), corresponding control commands are generated and issued to the C++ device model and the HVAC system in the EnergyPlus FMU for execution.

[0174] (d) Data recording and analysis: Record all states, actions, costs and comfort data throughout the simulation process for subsequent performance evaluation and algorithm analysis.

[0175] S2. Construct a multi-objective collaborative optimization problem.

[0176] Building upon the high-fidelity model described in step S1, this step aims to construct a complete mathematical optimization problem. This problem clearly defines the core objective that the system must pursue and the physical and operational constraints that must be followed. This is the crucial bridge connecting the physical model and the optimization algorithm, ensuring that the optimization results not only meet user needs but also possess engineering feasibility.

[0177] (1) Define the multi-objective collaborative optimization objective.

[0178] Home energy management is essentially a multi-objective optimization problem that needs to balance economy, energy efficiency, and comfort. This invention defines the following three core optimization objectives to guide the system to achieve optimal overall performance.

[0179] ① Define the primary optimization objective with economy as its core, namely, minimizing the total daily operating cost C of the system. total Its composition consists of the cost of purchasing electricity from the grid, expressed as:

[0180] ;

[0181] Among them, P grid,t C is the power exchanged with the grid at time t (kW, positive for purchasing electricity, negative for selling electricity). grid,t It is the time-of-use electricity price (CNY / kWh) at time t;

[0182] ② Define a second optimization objective centered on energy efficiency, namely, maximizing the self-consumption rate of renewable energy, R. SCR It represents the proportion of renewable energy generation that meets the total system load, and its expression is:

[0183] ;

[0184] Among them, P pv-eload,t P pv-hp,t P wt-eload,t and P wt-hp,t These are the power (kW) supplied by the photovoltaic and wind turbines and the heat pump at time t, respectively.

[0185] ③ Define a third optimization objective centered on comfort, ensuring that the temperature T in each region i at time t is... i,t Maintain within the comfort zone:

[0186] ;

[0187] in, and These represent the upper and lower limits (°C) of the thermal comfort range.

[0188] (2) Define system operation constraints.

[0189] To ensure the physical feasibility of the optimization scheme and the safe operation of the equipment, the following system-level and device-level constraints must be defined.

[0190] ① Define the power balance constraints that the system must satisfy to ensure that the power supplied by all power sources is equal to the power of all loads at time t:

[0191] ;

[0192] in, This is the power consumption (kW) of the HVAC system fan;

[0193] ② Define the heat power balance constraints that the system must satisfy to ensure that the heat pump output power is balanced with the heat load and the power of the heat storage tank at time t:

[0194] ;

[0195] in, It is the power (kW) required by the HVAC system to meet the heat load;

[0196] ③ Define the start-stop time constraints for the electrical load:

[0197] ;

[0198] in, and The earliest start time and latest end time t of device i eload,i It is the device's on-time (i) and (h). eload,i It is the runtime of device i;

[0199] In summary, the multi-objective collaborative optimization problem constructed in step S2 is to find a set of control strategies that optimize both the economic objective (S2(1)①) and the energy efficiency objective (S2(1)②) while strictly ensuring the comfort objective (S2(1)③), under the premise of satisfying all the above system constraints (electrical power balance, thermal power balance, equipment operating boundary and thermal comfort constraints). This complex optimization problem will be solved efficiently through the two-stage collaborative optimization framework described in step S3 below.

[0200] S3. Design and solve a two-stage collaborative optimization framework:

[0201] To address the complex multi-objective collaborative optimization problem constructed in step S2, this invention creatively designs and implements a two-stage collaborative optimization framework of "day-ahead scheduling-real-time control." The core idea of ​​this framework is to decompose the global optimization problem into two sub-problems that are decoupled in time scale but functionally coordinated: First, in the day-ahead scheduling stage, a multi-objective evolutionary algorithm is used for macro-level load planning oriented towards economy and energy efficiency; subsequently, in the real-time control stage, a deep reinforcement learning algorithm is used for minute-level fine-grained control focused on maintaining thermal comfort and economy. The two stages achieve coordination through information exchange, thereby realizing multi-objective collaborative optimization of energy efficiency, economy, and comfort at the global level.

[0202] Figure 2This is a detailed flowchart illustrating the implementation of the "two-stage collaborative optimization algorithm" described in this invention. The diagram visually demonstrates the complete algorithm flow and collaborative relationship between the day-ahead scheduling and real-time control stages. The day-ahead scheduling stage provides the system with a globally optimal load operating time benchmark, while the real-time control stage precisely adjusts the energy storage and HVAC systems based on this benchmark.

[0203] (1) Day-ahead scheduling phase: Based on the renewable energy output and load forecasts for the next 24 hours, a multi-objective genetic algorithm (NSGA-II) is used, with economy and energy efficiency as the primary objectives, to optimize the start-up and shutdown times of dispatchable loads (such as electric vehicles, washing machines, etc.) and generate a macro-level day-ahead energy scheduling plan. This phase focuses on strategic energy planning with a time resolution of 1 hour. The specific implementation method is as follows:

[0204] ① The scheduling time scale is set to 24 hours, with 1 hour as the basic scheduling period, i.e., T=24, Δt=1h;

[0205] ② The optimization variable is the start-stop time series of each schedulable load:

[0206] ;

[0207] in, It is a binary variable representing whether the i-th schedulable load runs during time period t, N S This represents the total number of schedulable loads;

[0208] ③ The non-dominated sorting genetic algorithm (NSGA-II) with an elitist strategy is used to solve the optimization variables. The fitness function of this algorithm is the multi-objective collaborative optimization problem defined in step (2), that is, simultaneously optimizing the total daily operating cost C. total and renewable energy self-consumption rate R SCR ;

[0209] ④ The operating parameters settings for the NSGA-II algorithm include: initializing the population size N. p =100, maximum number of generations I max =200, crossover probability P c =0.8, mutation probability P m =0.1;

[0210] ⑤ The constraints that need to be met during the day-ahead scheduling phase include:

[0211] (i) Scheduled load runtime window constraint: The actual runtime of each scheduleable load i must fall within its allowed working time range. Inside:

[0212] (ii) System power balance constraints: The power balance constraints defined in step S2 (3) must be satisfied in each scheduling period t;

[0213] (iii) Equipment operating boundary constraints: The operating boundary constraints of the battery energy storage system, thermal storage tank and electrical load defined in steps S1 and S2 (6) must be satisfied in each scheduling period t;

[0214] ⑥ The NSGA-II algorithm continuously evolves the population through selection, crossover, and mutation genetic operations, and finally outputs a set of Pareto optimal solutions. Each solution represents a start-stop time scheme for a schedulable load. From this set of Pareto optimal solutions, the day-ahead scheduling plan is selected based on actual preferences and used as the macro-energy planning benchmark for the real-time control stage.

[0215] (2) Real-time control stage: Under the macro framework of day-ahead scheduling plan generation, the charging and discharging power P of the battery energy storage system is controlled based on the maximum entropy deep reinforcement learning algorithm (SAC). ess,t The heat storage and release power P of the heat storage tank tst,t and the change in air supply volume Δm in each area of ​​the HVAC system i,t Perform rolling optimization and fine-tuning at the minute level (e.g., 15 minutes). This stage focuses on tactical real-time response to address uncertainties such as prediction errors, with the core objective of ensuring thermal comfort and further reducing operating costs. The SAC algorithm encourages exploration through its maximum entropy principle, effectively learning the optimal stochastic strategy in complex dynamic environments.

[0216] The specific implementation method of the real-time control stage is as follows:

[0217] ① The control time scale is set to 15 minutes, that is, the control step size Δt = 0.25 h. At each control time k, the agent interacts with the environment once.

[0218] ② The real-time control problem is modeled as a Markov decision process (MDP), which includes the following elements:

[0219] (i) State space S: the state s at time k k ∈S includes renewable energy generation P pv,t and P wt,t Battery energy storage system state of charge (SOC) ess,t and the remaining heat Q of the thermal storage tank tst,t Electrical load power P eload,t Time-of-use electricity price C grid,t Indoor temperature T in each area i,t Outdoor ambient temperature T out,t Area occupancy status (Body) m and equipment heating power Elecn This can be expressed as a formula:

[0220] ;

[0221] (ii) Action space A: Action a at time k k ∈A includes the change in air supply volume Δm in each area. i,t Battery energy storage system charging and discharging power P ess,t And the heat storage and release power P of the thermal storage tank tst,t This can be expressed as a formula:

[0222] ;

[0223] (iii) Reward function R: reward r t This is a multi-objective weighted sum designed to guide the agent to simultaneously optimize thermal comfort, penalties for violating energy storage operation rules, and operating costs.

[0224] ;

[0225] Where r 1,t It is a penalty related to regional thermal comfort, when temperature T i,t Beyond the comfort zone Punishment will be imposed at that time:

[0226] ;

[0227] r 2,t This is a penalty for energy storage operations during non-economic periods, preventing battery energy storage and thermal storage tanks from discharging during off-peak electricity prices or charging during peak electricity prices, when the electricity price is lower than... It is off-peak electricity pricing, and the electricity price is higher than [previous price]. Peak electricity pricing:

[0228] ;

[0229] r 3,t and r 4,t These are the operating costs of thermal energy and electrical energy:

[0230] ;

[0231] .

[0232] ③ The Soft Actor-Critic (SAC) algorithm is used as the core optimizer for solving the MDP. This algorithm includes the following core components and processes:

[0233] (i) The policy network (actors) is parameterized as Its optimization objective is to maximize the weighted sum of expected cumulative reward and policy entropy:

[0234] ;

[0235] Where α is a temperature parameter used to balance the reward term and the entropy term; It is the policy entropy, defined as:

[0236] ;

[0237] (ii) Two soft Q-function networks (commentator) Update by minimizing the soft Bellman residual:

[0238] ;

[0239] Where D is the experience replay buffer; γ is the discount factor; These are the target network parameters; It is the target soft state value function, calculated through the target Q-network and the policy network:

[0240] ;

[0241] (iii) The policy network (actors) is updated by minimizing the following objective function:

[0242] ;

[0243] In actual updates, the reparameterization technique is used to represent the action as a Gaussian distribution. To reduce variance;

[0244] (iv) The temperature parameter α is automatically adjusted by minimizing the following objective function:

[0245] ;

[0246] in, It is the target entropy, usually set to −dim(A) (the negative of the action space dimension).

[0247] (v) Target Q-network parameters Slowly track the current Q network parameters using a soft update method:

[0248] ;

[0249] in The update step size for the target soft Q network.

[0250] ④ The key hyperparameter settings of the SAC algorithm include: learning rate λ Q , λ π , λ α =3×10 -4Discount factor γ = 0.99, experience replay cache size |D| = 2 × 10 5 The target network update rate τ = 0.005, and the minimum batch size is 256.

[0251] ⑤ The constraints that the real-time control stage needs to meet include the electric power balance constraints, thermal power balance constraints, thermal comfort constraints, and equipment operation boundary constraints defined in step S2. These constraints are reflected and guaranteed through the penalty term in the reward function and the action pruning method.

[0252] To rigorously verify the effectiveness, superiority, and engineering practical potential of the method described above in this invention, a high-fidelity C++ / Python-EnergyPlus co-simulation platform was constructed for comprehensive comparative experiments. The simulation environment was based on latitude and longitude: 36.1°N, 120.4°E. Measured meteorological data from the summer of 2023 (sourced from the NASA POWER database) and residential peak-valley time-of-use electricity pricing policies were used. Key equipment parameters in the system were set as follows: photovoltaic installed capacity 2.8kW, wind turbine installed capacity 2kW, battery energy storage capacity 10kWh, and thermal storage tank capacity 20kWh. To comprehensively evaluate performance, three comparative scenarios were set: Case 1 (fixed timetable + rule control), Case 2 (fixed timetable + SAC control), and Case 3 (the two-stage optimization method proposed in this invention).

[0253] The simulation results over a week show that the method of this invention (Case 3) significantly and consistently outperforms the comparative methods in all key performance indicators. In terms of economy, the average daily electricity cost of Case 3 is reduced to 6.04 CNY, a significant reduction of 50.3% and 33.5% compared to Case 1 (12.14 CNY) and Case 2 (9.08 CNY), respectively, with particularly noticeable cumulative cost savings. Regarding energy efficiency, its renewable energy self-consumption rate (SCR) reaches 72.3%, indicating a significantly enhanced local absorption capacity of renewable energy. In terms of comfort, Case 3 can maintain the temperature of each area within the set comfort range with high precision, with an average temperature violation rate as low as 5.1% throughout the week, and temperature fluctuations far less than in the comparative scenarios, providing users with a continuously stable thermal comfort environment. These data strongly demonstrate that this invention, through two-stage collaborative optimization, effectively mitigates source-load uncertainties while successfully achieving a synergistic improvement in energy efficiency, economy, and comfort.

Claims

1. A two-stage optimization method for a residential wind-solar-storage system that aims for synergy between thermal comfort and energy efficiency, characterized in that, Includes the following steps: S1. Construct a high-fidelity model of a residential solar-powered energy storage system: An integrated model was established, including photovoltaic, wind turbine, battery energy storage system, thermal storage tank, heat pump and multi-zone HVAC system, to accurately characterize the coupling relationship and operational constraints of electric and heat multi-energy flow in the entire chain of generation, storage, conversion and consumption, so as to provide an accurate physical basis for subsequent optimization; S2. Construct a multi-objective collaborative optimization problem: With the core optimization objectives of minimizing daily electricity costs, maximizing the self-consumption rate of renewable energy, and accurately maintaining thermal comfort in multiple regions, the system defines the power balance constraints, thermal power balance constraints, equipment operation boundary constraints, and comfort range constraints that the operation of a residential wind-solar-storage system must meet, forming a complete mathematical description of the optimization problem. S3. Design and solve a two-stage collaborative optimization framework: The global optimization problem is decomposed into two interconnected stages: day-ahead scheduling and real-time control, to achieve efficient solution.

2. The two-stage optimization method for a residential wind-solar-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 1, characterized in that, The high-fidelity model in step S1 specifically includes: (1) Output power P of the solar photovoltaic power generation model pv,t From the actual light intensity G a,t and photovoltaic panel operating temperature T a,t A joint decision, specifically expressed as: ； Where, N pv It is the number of photovoltaic panels, η pv It is the attenuation coefficient, P pv This refers to the rated power of the photovoltaic panel under standard conditions, k. pv It is the power temperature coefficient, T ref It is the standard ambient temperature, G ref This refers to standard solar irradiance. (2) The output power P of the wind power generation model wt,t is determined by the wind speed V at the hub w,t and is specifically expressed as: ； Among them, V in V ref V out These are the inlet velocity, rated velocity, and outlet velocity of the fan, respectively. wt This is the rated power of the fan; (3) State of charge (SOC) of the battery energy storage model t Represented as: ； Where, η ess It refers to the energy utilization efficiency of battery energy storage systems, E ess It is the capacity of the energy storage system, P ess,t It refers to the charging and discharging power of the energy storage system, SOC. t-1 It represents the state of charge of the battery energy storage model at the previous moment, and Δt is the length of each time step; (4) The residual heat Q of the thermal storage tank model tst,t Represented as: ； Among them, Q tst,t-1 It is the heat energy stored in the heat storage tank at the previous moment, α tst It is the heat exchange coefficient between the thermal storage tank and the environment, η tst It is the energy utilization efficiency of the thermal storage tank, P tst,t It is the heat storage or heat release power of the heat storage tank at time t; (5) Heating or cooling power Q of the heat pump model hp,t Its power consumption P hp,t and coefficient of performance (COP) t Related, specifically: ； Where λ c1 , λ c2 and λ c3 T is a fitting parameter related to the physical characteristics of a heat pump. out,t It is the ambient temperature when the heat pump is working; (6) The electrical load model includes multiple electrical devices, and the operating power of the i-th electrical load at time t is... Represented as: ； in, This indicates the switching state of the i-th electrical load at time t, where 0 represents off and 1 represents on; This is the rated operating power of the load; (7) The thermal dynamic model of the multi-zone HVAC system describes the temperature T of each zone at time t. i,t The heat exchange Q between the building envelope and the outdoor environment i,t out The heat exchange Q between adjacent regions i,t adj The cooling capacity Q supplied to the area by the multi-zone HVAC system i,t HVAC And the internal heat Q generated by people and equipment inside the room. i,t in The correlation is defined by the following system of equations: ； Where ρ is the air density, C ρ It is the specific heat capacity of air at constant pressure, V i A is the volume of region i. ZONEi and A i,s These are the contact area between the region and the outside world, and the contact area with adjacent regions, N. s It is the number of adjacent regions, T out,t and T s,t These are the outdoor temperature at time t and the temperature of the adjacent area, respectively. C It is the air supply temperature, q l It is the infiltration air volume ratio, Body m and Elec n These are the heat generation power of personnel and equipment, N. m and N n These refer to the number of personnel and equipment, U trans1 It is the heat transfer coefficient between the area and the outdoors, U trans2 It is the heat transfer coefficient between a region and its adjacent regions, T i,t T is the temperature of region i at time t. i,t+1 m is the temperature of region i at the previous moment. i,t It is the air supply volume for supply area i.

3. The two-stage optimization method for a residential solar-wind-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 2, is characterized in that... The state of charge (SOC) of the battery energy storage model t The operation must meet power constraints and state of charge constraints: ； in, and These are the maximum allowable charging power and maximum discharging power of the battery energy storage system, and SOC. max and SOC min These are the maximum and minimum values ​​of the state of charge.

4. The two-stage optimization method for a residential wind-solar-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 2, characterized in that, The residual heat Q of the thermal storage tank model tst,t The operation must meet power constraints and thermal constraints: ； in and These are the maximum allowable heat storage capacity and maximum heat release capacity of the thermal storage tank, respectively. and These are the maximum and minimum values ​​of the remaining energy in the thermal storage tank, respectively.

5. The two-stage optimization method for a residential solar-wind-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 1, characterized in that, The multi-objective collaborative optimization problem in step S2 is specifically defined as follows: (1) Define the first optimization objective with economy as the core, that is, minimize the total daily operating cost C of the system. total Its composition consists of the cost of purchasing electricity from the grid, expressed as: ； Among them, P grid,t C is the power exchanged with the power grid at time t. grid,t It is the time-of-use electricity price at time t; (2) Define the second optimization objective with energy efficiency as the core, namely, maximizing the self-consumption rate of renewable energy R. SCR It represents the proportion of renewable energy generation that meets the total system load, and its expression is: ； Among them, P pv-eload,t P pv-hp,t P wt-eload,t and P wt-hp,t These represent the power supplied by the photovoltaic and wind turbines to the electrical load and the power of the heat pump at time t, respectively. eload,t P is the power of the electrical load at time t. hp,t It is the heat pump power at time t; (3) Define the power balance constraints that the residential solar-wind-storage system must meet to ensure that the power supplied by all power sources is equal to the power of all loads at time t: ； Among them, P pv,t P is the photovoltaic power at time t. wt,t P is the power of the wind turbine at time t. grid,t P is the power purchased by the power grid at time t. ess,t P is the charging and discharging power of the battery energy storage system at time t. hp,t It is the heat pump power at time t. It is the power of the i-th load at time t. It refers to the power of the fans in a multi-zone HVAC system; (4) Define the heat power balance constraints that the residential solar-wind storage system must meet to ensure that the heat pump output power is balanced with the heat load and the power of the heat storage tank at time t: ； in, It is the power required by a multi-zone HVAC system to meet the heat load, P tst,t It is the thermal power provided by the thermal storage tank at time t; (5) Define a third optimization objective centered on comfort, ensuring that the temperature T in each region i at time t is constant. i,t Maintain within the comfort zone: ； in, and These represent the upper and lower bounds of the thermal comfort range, respectively. (6) Define the start-stop time constraints for electrical loads: ； in, and The earliest start time and latest end time t of device i eload,i It is the device i startup time, h eload,i It is the runtime of device i; The multi-objective collaborative optimization problem is to collaboratively optimize the first optimization objective, the second optimization objective, and the third optimization objective under the premise of satisfying the above-mentioned constraints on power balance, heat power balance, and power load start-up and shutdown time.

6. The two-stage optimization method for a residential wind-solar-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 1, characterized in that, The specific process of step S3 includes: (1) Day-ahead scheduling phase: Based on the renewable energy output and load forecast for the next 24 hours, a multi-objective genetic algorithm is used to optimize the start-up and shutdown times of the dispatchable loads with economy and energy efficiency as the primary objectives, and to generate a macro-level day-ahead energy scheduling plan; (2) Real-time control stage: Within the macro framework of the recently generated scheduling plan, the charging and discharging power P of the battery energy storage system is analyzed using a maximum entropy deep reinforcement learning algorithm. ess,t The heat storage and release power P of the heat storage tank tst,t And the change in air supply volume Δm in each area of ​​the multi-zone HVAC system. i,t Perform minute-level rolling optimization and fine-grained control; (3) The soft actor-critic algorithm is used as the core optimizer for solving the MDP; (4) The key hyperparameter settings of the SAC algorithm include: learning rate λ Q , λ π , λ α =3×10 -4 Discount factor γ = 0.99, experience replay cache size |D| = 2 × 10 5 The target network update rate τ = 0.005, and the minimum batch size is 256. (5) The constraints that the real-time control stage needs to meet include the electric power balance constraints, thermal power balance constraints, thermal comfort constraints and equipment operation boundary constraints defined in step S2. These constraints are reflected and guaranteed by the penalty term in the reward function and the action pruning method.

7. The two-stage optimization method for a residential solar-wind-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 6, characterized in that, The day-ahead scheduling phase focuses on strategic energy planning with a time resolution of 1 hour. The specific implementation method is as follows: (a) The scheduling time scale is set to 24 hours, with 1 hour as the basic scheduling period, i.e., T=24, Δt=1h; (b) The optimization variable is the start-stop time series of each schedulable load: ； in, It is a binary variable representing whether the i-th schedulable load runs during time period t, N S This represents the total number of schedulable loads; (c) The optimization variables are solved using a non-dominated sorting genetic algorithm with an elitist strategy. The fitness function of this algorithm is the multi-objective collaborative optimization problem defined in step S2, i.e., simultaneously optimizing the total daily operating cost C. total and renewable energy self-consumption rate R SCR ; (d) The running parameters of the non-dominated sorting genetic algorithm with elitist strategy include: initializing the population size N. p =100, maximum number of generations I max =200, crossover probability P c =0.8, mutation probability P m =0.1; (e) The constraints that the day-ahead scheduling phase must satisfy include: (i) Scheduled load runtime window constraint: The actual runtime of each scheduleable load i must fall within its allowed working time range. Inside: (ii) System power balance constraints: The power balance constraints defined in step S2 must be satisfied in each scheduling period t; (iii) Equipment operation boundary constraints: The operation boundary constraints of the battery energy storage system, thermal storage tank and electrical load defined in steps S1 and S2 must be satisfied in each scheduling period t; (f) The non-dominated sorting genetic algorithm with elite strategy continuously evolves the population through selection, crossover, and mutation genetic operations, and finally outputs a set of Pareto optimal solutions. Each solution represents a start-stop time scheme for a schedulable load. From the Pareto optimal solution set, the day-ahead scheduling plan is selected according to actual preferences as the macro-energy planning benchmark for the real-time control stage.

8. The two-stage optimization method for a residential wind-solar-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 6, characterized in that, The specific implementation method of the real-time control stage is as follows: (a) The control time scale is set to 15 minutes, i.e. the control step size Δt = 0.25 h. At each control time k, the agent interacts with the environment once. (b) The real-time control problem is modeled as a Markov decision process, which includes the following elements: (i) State space S: the state s at time k k ∈S includes renewable energy generation P pv,t and P wt,t Battery energy storage system state of charge (SOC) ess,t and the remaining heat Q of the thermal storage tank tst,t Electrical load power P eload,t Time-of-use electricity price C grid,t Indoor temperature T in each area i,t Outdoor ambient temperature T out,t Area occupancy status (Body) m and equipment heating power Elec n This can be expressed as a formula: ； (ii) Action space A: Action a at time k k ∈A includes the change in air supply volume Δm in each area. i,t Battery energy storage system charging and discharging power P ess,t And the heat storage and release power P of the thermal storage tank tst,t This can be expressed as a formula: ； (iii) Reward function R: reward r t This is a multi-objective weighted sum designed to guide the agent to simultaneously optimize thermal comfort, penalties for violating energy storage operation rules, and operating costs. ； Where r 1,t It is a penalty related to regional thermal comfort, when temperature T i,t Beyond the comfort zone Punishment will be imposed at that time: ； r 2,t This is a penalty for energy storage operations during non-economic periods, preventing battery energy storage and thermal storage tanks from discharging during off-peak electricity prices or charging during peak electricity prices, when the electricity price is lower than... It is off-peak electricity pricing, and the electricity price is higher than [previous price]. Peak electricity pricing: ； r 3,t and r 4,t These are the operating costs of thermal energy and electrical energy: ； 。 9. The two-stage optimization method for a residential wind-solar-storage system oriented towards the synergy of thermal comfort and energy efficiency as described in claim 6, characterized in that, The soft actor-critic algorithm comprises the following core components and processes: (i) The policy network is parameterized as Its optimization objective is to maximize the weighted sum of expected cumulative reward and policy entropy: ； Where α is a temperature parameter used to balance the reward term and the entropy term; It is the policy entropy, defined as: ； (ii) Two soft Q-function networks Update by minimizing the soft Bellman residual: ； Where D is the experience replay buffer; γ is the discount factor; These are the target network parameters; It is the target soft state value function, calculated through the target Q-network and the policy network: ； (iii) The policy network is updated by minimizing the following objective function: ； In actual updates, the reparameterization technique is used to represent the action as a Gaussian distribution. To reduce variance; (iv) The temperature parameter α is automatically adjusted by minimizing the following objective function: ； in, This is the target entropy, usually set to -dim(A); (v) Target Q-network parameters Slowly track the current Q network parameters using a soft update method: ； in The update step size for the target soft Q network.

Images