Contribution degree perceived power grid interaction type residential building collaborative operation optimization method

Abstract

The invention discloses a contribution degree perceived collaborative operation optimization method for a power grid interactive type high-efficiency residential building. The method comprises the following steps: (1) respectively modeling an operation cost minimization problem of the power grid interactive type high-efficiency residential building in a normal operation stage and a collaborative operation double-layer optimization problem in a power grid service stage; (2) proposing a neural network assisted hierarchical model prediction control method based on physical consistency to solve the operation cost minimization problem; (3) proposing a distributed collaborative operation optimization algorithm based on physical consistency neural network assisted hierarchical model predictive control guided binary search to solve the collaborative operation double-layer optimization problem; and (4) deploying the collaborative operation optimization method in an actual system. Compared with the prior art, the method can identify the contribution of each residential building participating in providing the power grid service and provide a basis for corresponding economic compensation, and can reduce the building energy cost and improve the power grid service capability on the premise of maintaining high user comfort.

Description

Technical Field

[0001] The present invention relates to a contribution-aware grid-interactive residential building collaborative operation optimization method, and belongs to the intersection field of residential building HVAC systems and artificial intelligence. Background Art

[0002] In recent years, the operation of the power grid is facing tremendous pressure caused by the growing peak power demand, limited transmission and distribution capacity, and large-scale renewable energy grid connection. In order to alleviate the above pressure, the grid-interactive efficient building was proposed by the U.S. Department of Energy in 2019. A grid-interactive efficient building is a high-efficiency building that can use smart technology and on-site distributed energy (such as rooftop photovoltaics, battery energy storage systems, and backup generators) to provide demand flexibility, while jointly optimizing energy costs, grid services, and user comfort. Compared with existing automatic demand response, grid-interactive efficient buildings have higher energy efficiency and load flexibility. Specifically, automatic demand response mainly achieves load flexibility by cutting or transferring loads according to the dispatch instructions of the grid or system operator, which is less helpful for improving system energy efficiency. Grid-interactive buildings can achieve higher energy efficiency and load flexibility through intelligent, comprehensive and autonomous management of distributed energy resources and building flexible loads. According to the analysis of the U.S. Department of Energy, from 2021 to 2040, grid-interactive efficient buildings are expected to provide cumulative benefits of US$100 billion to US$200 billion to the U.S. power system and reduce carbon emissions by 80 million tons per year. Therefore, its research has important economic and social value. At the same time, grid-interactive efficient buildings can provide various types of grid services, such as power generation energy services, power generation capacity services, emergency reserve services, frequency regulation services, and voltage support services. However, due to limited service capacity, a single residential building cannot provide grid services, so it is necessary to consider the coordination between multiple residential buildings. Specifically, through the load aggregator (e.g., a trusted third party) to coordinate the operation of multiple residential buildings, multiple buildings can efficiently provide grid services while minimizing the negative impact on building operating costs and their user needs (e.g., user thermal comfort). Since the economic incentives provided by the distribution system operator help reduce the operating costs of buildings, a win-win situation can be achieved between the buildings participating in the grid service and the distribution system operator. In summary, it is crucial to design an efficient method for the coordinated operation of grid-interactive residential buildings. However, it is challenging to achieve the coordinated operation optimization of multiple grid-interactive buildings due to the existence of uncertain parameters (such as outdoor temperature, electricity price, photovoltaic and base load), time coupling constraints (energy storage system level), spatial coupling constraints (the total demand of all residential buildings meets the power upper limit), nonlinear objective functions (such as the objective function is a nonlinear non-convex and non-separable function), the difficulty in establishing a clear and accurate building thermal dynamics model, and the difficulty in obtaining a large amount of operating data.

[0003] In view of the above challenges, the existing grid-interactive efficient building operation optimization methods are no longer applicable. Specifically, the existing operation optimization methods are mainly divided into model-based methods and learning-based methods. Model-based methods mainly include robust optimization, alternating vector multiplier method, and distributed model predictive control. The above methods require a clear building thermal dynamics model. However, since the indoor temperature of a building depends on many factors (such as building structure and materials, external environment (such as external temperature, humidity and solar radiation intensity), and internal heat gain from users and lighting systems), it is very difficult to establish an accurate and easy-to-control building thermal dynamics model. Although the existing learning-based operation optimization methods (such as soft actor-critic algorithm, multi-agent deep reinforcement learning, multi-agent actor-attention critic algorithm, and multi-agent hierarchical deep reinforcement learning) do not need to know a clear building thermal dynamics model and support real-time decision-making, the above methods require a large number of interactions with the actual environment or establish a high-fidelity simulation environment. The deployment cost and threshold of the method are high, which is not conducive to implementation. In addition, existing methods ignore the identification of changes in electricity demand for each grid-interactive building due to its participation in providing grid services, and thus fail to calculate the economic compensation that each grid-interactive efficient building should receive, which is not conducive to the successful deployment of grid-interactive efficient buildings in practice.

[0004] In summary, the existing grid-interactive building operation optimization methods have shortcomings in overcoming the above challenges. It is urgent to study a new contribution-aware method for collaborative operation optimization of grid-interactive buildings to optimize the operating costs of grid-interactive buildings and provide efficient grid services while maintaining high user comfort. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a contribution-aware grid-interactive residential building collaborative operation optimization method, which aims to make up for the shortcomings of the existing operation optimization methods that do not consider the contribution of grid-interactive buildings in participating in grid services and make corresponding economic compensation, and to solve the technical problems that the existing model-based methods require a clear indoor environment dynamic model and the existing learning-based methods require a large amount of operation data.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] The present invention provides a contribution-aware grid-interactive residential building collaborative operation optimization method, comprising:

[0008] (1) Model the operation cost minimization problem of grid-interactive efficient residential buildings in the normal operation phase and the two-level optimization problem of coordinated operation in the grid service phase respectively;

[0009] (2) A physically consistent neural network-assisted hierarchical model predictive control method is proposed to solve the above-mentioned operation cost minimization problem;

[0010] (3) A distributed collaborative operation optimization algorithm based on physically consistent neural network-assisted hierarchical model predictive control guided binary search is proposed to solve the above-mentioned collaborative operation two-level optimization problem;

[0011] (4) Deploy the proposed collaborative operation optimization method in an actual system.

[0012] Optionally, the modeled grid-interactive high-efficiency residential building operation cost minimization problem during normal operation mainly includes decision variables, constraints and objective functions.

[0013] The decision variable χ includes the energy storage system input power c of residential building j managed by load aggregator i in time slot t. i,j,t , energy storage system output power d i,j,t 、HVAC system input power e i,j,t and the electricity demand of residential buildings P i,j,t .

[0014] The constraints include:

[0015]

[0016] c i,j,t ·d i,j,t =0, (5)

[0017] P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (6)

[0018]

[0019] Among them: B i,j,t represents the storage capacity of the energy storage system of residential building j in load aggregator i at time slot t; σ c,i,j and σ d,i,j They represent the input efficiency and output efficiency of the energy storage system respectively; and They represent the minimum storage capacity and maximum storage capacity of the energy storage system respectively; and They represent the maximum input power and maximum output power of the energy storage system respectively; Δt represents the time slot length; r i,j,t represents the photovoltaic power output at time slot t, b i,j,t represents the fixed load of time slot t; T i,j,trepresents the indoor temperature of residential building j in load aggregator i at time slot t, T t out represents the outdoor temperature at time slot t, ξ i,j,t represents the indoor thermal disturbance of residential building j in load aggregator i at time slot t, represents the indoor thermal dynamics of residential building j in load aggregator i; represents the maximum HVAC system input power of residential building j in load aggregator i; It represents the upper limit of electricity demand of residential building j in load aggregator i during the grid service stage.

[0020] The objective function is:

[0021]

[0022] in: represents the expectation operator, acting on uncertain system parameters (such as fixed load b i,j,t , Photovoltaic output i,j,t , outdoor temperature T t out and buying and selling electricity prices, etc.); C i,j,1,t and C i,j,2,t They represent the operating cost of residential building j and the loss cost of energy storage system in load aggregator i at time slot t respectively, and the calculation formula is as follows:

[0023]

[0024] C i,j,2,t =ψ(c i,j,t +d i,j,t ),

[0025] Where: s t and w t They represent the purchase price and the selling price of electricity respectively; ψ represents the depreciation coefficient of the energy storage system.

[0026] Optionally, the established two-layer optimization problem for the coordinated operation of grid-interactive efficient residential buildings in the grid service stage mainly includes decision variables, constraints and objective functions.

[0027] The decision variable is the peak power demand limit of each building in the grid service phase t time slot ρ t .

[0028] The constraints include:

[0029] 0≤ρ t ≤ρ max , (12)

[0030]

[0031] c i,j,t ·d i,j,t =0, (20)

[0032] P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (twenty one)

[0033]

[0034] Where: max Indicates the maximum peak power demand limit; represents the optimal power demand of load aggregator i at time slot t; P limit,t represents the sum of the power caps of all residential buildings in time slot t; represents the optimal power demand of each building at time slot t; X1 represents the initial decision quantity before the proposed binary search; α represents a weight coefficient.

[0035] The objective function is:

[0036]

[0037] Where: Γ(·) represents the economic compensation related to the change in electricity demand of all grid-interactive residential buildings. In the present invention, It is the economic compensation coefficient.

[0038] Optionally, the operation optimization method based on physically consistent neural network assisted hierarchical model predictive control is designed as follows:

[0039] The proposed grid-interactive high-efficiency residential building operation cost minimization problem in the normal operation phase is transformed into the upper indoor comfort temperature deviation minimization sub-problem and the lower operation cost minimization sub-problem;

[0040] The upper-level model predictive control method assisted by the indoor thermal dynamic model based on the physical consistency neural network is deployed to solve the indoor comfort temperature deviation minimization problem and obtain the upper-level optimal decision in the t time slot.

[0041] The optimal decision of the upper-level model predictive control method is input into the lower-level model predictive control, and the lower-level model predictive control method is deployed to solve the sub-problem of minimizing the operating cost, and the optimal decision of the lower level in time slot t is obtained.

[0042] Optionally, the proposed upper indoor comfort temperature deviation minimization sub-problem is as follows:

[0043]

[0044]

[0045] Where: X represents the prediction length, t′ represents the corresponding time slot within the prediction length, C i,j,3,t′ Indicates the indoor comfort temperature deviation value; It is a dynamic model of indoor thermal comfort based on physical consistency neural network, which is mainly composed of black box module and physical module. The black box module is composed of a deep neural network, whose input is outdoor light intensity and output is indoor temperature disturbance T caused by light intensity. b,i,j,t′+1 ; The physical module is composed of the indoor temperature T of the current time slot i,j,t′ , the outdoor temperature at the current time slot Current time slot HVAC system input power e i,j,t′ It can be represented by a clear physical linear model, namely Where: γ and ν represent the learnable parameters that ensure the physical consistency of the neural network.

[0046] Optionally, the proposed lower layer operation cost minimization sub-problem is as follows:

[0047]

[0048] c i,j,t′ ·d i,j,t′ =0, (32)

[0049] P i,j,t′ +r i,j,t′ +d i,j,t′ =b i,j,t′ +e i,j,t′ +c i,j,t′ , (33)

[0050]

[0051] Where: C i,j,1,t′ and C i,j,2,t′ They respectively represent the operating cost of each building and the loss cost of the energy storage system in the time slot t′ within the prediction length.

[0052] Optionally, the deployed upper model predictive control method mainly uses the gradient descent method to solve the optimal HVAC system decision for the current time slot and the future X-1 time slots, that is, The specific solution process is as follows: First, set the current iteration step to k = 0 and the maximum iteration step to k. max :

[0053] (1.1) Predictive control output decision based on upper-level model The proposed physical embedded neural network predicts

[0054] (1.2) Calculate the indoor comfort deviation between the current time slot and the next X-1 time slots

[0055] (1.3) Based on and The gradient is calculated according to the following formula:

[0056]

[0057] (1.4) Based on the obtained gradient, the gradient optimizer optimizes the upper-level decision with gradient minimization as the update direction. The specific calculation formula is as follows:

[0058]

[0059] in: and are the exponential moving average of the gradient and the exponential moving average of the squared gradient, respectively; β and For updating and Hyperparameters of and They are and bias-corrected estimates of ; express The function is used to update the decision of the current iteration step; ζ and ∈ are the learning rate and hyperparameters respectively.

[0060] (1.5) If the current And k<k max , k=k+1 and jump to iterate again; otherwise, the iteration ends and the optimal HVAC system decision is output

[0061] The deployed lower-level model predictive control method mainly introduces four auxiliary variables θ i,j,t′ ,κ i,j,t′ ,I c,t′ ,I d,t′ Transform (P4) into the following mixed integer linear programming problem (P5) to solve the optimal energy storage charging and discharging decision for the current time slot and the next X-1 time slots, namely:

[0062]

[0063] st(28)-(29),(43)

[0064]

[0065] I c,t′ +Id,t′ ≤1, (46)

[0066] I c,t′ ,I d,t′ ∈{0,1},(47)

[0067]

[0068] θ i,j,t ≥0, (50)

[0069] κ i,j,t ≥0, (51)

[0070] in: make and P i,j,t′ =θ i,j,t′ -κ i,j,t′ and |P i,j,t′ |=θ i,j,t′ +κ i,j,t′ ; The decision variable of (P5) is c i,j,t′ ,d i,j,t′ ,θ i,j,t′ ,κ i,j,t′ ,I c,t′ ,I d,t′ .

[0071] Optionally, the distributed collaborative operation optimization algorithm based on hierarchical model predictive control assisted binary search proposed in the power grid service stage is designed as follows:

[0072] If the current operation time slot t is in the grid service stage, firstly, the initial decision X1 of the current time slot is obtained according to the operation optimization method based on hierarchical model predictive control; then the upper limit of the current power demand peak limit is set to ρ u =ρ max , the current lower limit of the peak power demand is ρ l =ρ min , is a threshold. And the current iteration step k=0 is less than the maximum number of iterations k max , iterate according to the following steps to solve the optimal peak power demand limit and optimal decision

[0073] (2.1) Let the current peak power demand limit be ρ t =(ρ u +ρ l ) / 2;

[0074] (2.2) The distributed system operator willt Transmitted to all load aggregators i;

[0075] (2.3) Load aggregator will t Transmitted to all grid-interactive and efficient residential buildingsj;

[0076] (2.4) According to the rule-based decision adjustment method, all buildings adjust X1 to obtain the optimal decision

[0077] (2.5) All buildings will be optimally decided Transmitted to the corresponding load aggregator;

[0078] (2.6) Distributed system operators obtain optimal power demand

[0079] (2.7) If the current Then l =ρ t On the contrary, ρ u =ρ t ;

[0080] (2.8) If ρ u -ρ l >ω and k<k max , return to (8.1) and continue iteration, k = k + 1; otherwise, end the iteration and output the current optimal ρ t and optimal decision

[0081] Optionally, the proposed rule-based decision adjustment method is designed as follows: (3.1) If and but

[0082] (3.2) If and but

[0083] (3.3) If and but (3.4) If and and but

[0084] (3.5) If and and but

[0085] Optionally, the actual operation system of the deployment of the proposed collaborative operation optimization method includes a grid-interactive efficient residential building operation module and a real-time decision-making module based on hierarchical model predictive control assisted binary search.

[0086] The grid-interactive high-efficiency residential building operation module is responsible for receiving the optimal decision output by the real-time decision module and executing it in each grid-interactive high-efficiency residential building.

[0087] The real-time decision-making module based on hierarchical model predictive control assisted binary search includes an information acquisition submodule, an upper-level model predictive control real-time decision-making submodule, a lower-level model predictive control real-time decision-making submodule, a binary search collaborative operation optimization submodule and an optimal control strategy output submodule.

[0088] The information collection submodule is responsible for collecting the latest status of each component after each grid-interactive efficient residential building executes the decision, such as the indoor temperature of the building and the real-time storage capacity of the energy storage system. Then, it is sent to the upper and lower model predictive control real-time decision submodules respectively.

[0089] The upper-layer model predictive control real-time decision submodule reads the indoor temperature status sent by the information acquisition submodule. Then, the operation optimization method is used to solve the optimal HVAC system decision. Finally, the optimal HVAC system decision is sent to the lower-layer model predictive control real-time decision submodule.

[0090] The lower-layer model predictive control real-time decision submodule reads the energy storage system state issued by the information acquisition submodule and the optimal HVAC system decision issued by the upper-layer model predictive control real-time decision submodule. Then, the optimal energy storage system decision is solved using the operation optimization method. Finally, the optimal HVAC system decision and the energy storage system decision are transmitted to the binary search collaborative operation optimization module.

[0091] The binary search collaborative operation optimization submodule first determines whether the current time slot requires each grid-interactive efficient residential building to participate in grid services. If it is necessary to participate in grid services, the collaborative operation optimization method is used to further solve the optimal decision, and the latest decision is sent to the optimal control strategy output submodule; if it is not necessary to participate in grid services in the current time slot, the optimal decision of the upper and lower model predictive control real-time decision submodules is sent to the optimal control strategy output submodule.

[0092] The optimal control strategy output submodule reads the decision issued by the binary search collaborative operation optimization submodule and sends it to the grid-interactive efficient residential building operation module.

[0093] In general, the technical solution of the present invention can identify the contribution of each residential building in providing grid services and provide a basis for corresponding economic compensation, which can reduce building energy costs and improve grid service capabilities while maintaining high user comfort. At the same time, compared with the existing technology, it can achieve the following beneficial effects:

[0094] (1) Compared with the existing model-based methods, the proposed method does not need to know the explicit building thermodynamic model. By constructing a building thermodynamic model based on a physical consistency neural network, it accurately captures the indoor thermal dynamics and deploys it in the proposed operation optimization algorithm to assist the upper-level model predictive control to obtain the optimal decision and maintain a comfortable indoor environment.

[0095] (2) Compared with existing learning-based methods, the method proposed in the present invention does not require a large number of interactions with the actual environment and has a low deployment threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 It is a flow chart of the contribution-aware grid-interactive residential building collaborative operation optimization method proposed in the present invention;

[0097] Figure 2 is a comparison chart of the total operating costs of the method of the present invention and other methods;

[0098] Figure 3 is a comparison chart of average temperature deviations of the method of the present invention and other methods;

[0099] Figure 4 It is a comparison diagram of power limit offset between the method of the present invention and other methods. DETAILED DESCRIPTION

[0100] The present invention will be further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0101] Embodiment 1:

[0102] like Figure 1 As shown, the design flow chart of the contribution-aware grid-interactive residential building collaborative operation optimization method provided by the present invention includes the following steps:

[0103] 1. Model the operation cost minimization problem of grid-interactive efficient residential buildings in the normal operation phase and the two-level optimization problem of coordinated operation in the grid service phase respectively;

[0104] 1.1. In this embodiment, the problem of minimizing the operating cost of a grid-interactive high-efficiency residential building during the normal operation phase includes decision variables, constraints, and an objective function;

[0105] The decision variables χ include the energy storage system input power c of residential building j managed by load aggregator i in time slot t i,j,t , energy storage system output power d i,j,t 、HVAC system input power e i,j,t and the electricity demand of residential buildings P i,j,t .

[0106] The constraints include:

[0107]

[0108] c i,j,t ·d i,j,t =0, (5)

[0109] P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (6)

[0110]

[0111] Among them: B i,j,t represents the storage capacity of the energy storage system of residential building j in load aggregator i at time slot t; σ c,i,j and σ d,i,j They represent the input efficiency and output efficiency of the energy storage system respectively; and They represent the minimum storage capacity and maximum storage capacity of the energy storage system respectively; and They represent the maximum input power and maximum output power of the energy storage system respectively; Δt represents the time slot length; r i,j,t represents the photovoltaic power output at time slot t, b i,j,t represents the fixed load of time slot t; T i,j,t represents the indoor temperature of residential building j in load aggregator i at time slot t, T t out represents the outdoor temperature at time slot t, ξ i,j,t shows the indoor thermal disturbance of residential building j in load aggregator i at time slot t, represents the indoor thermal dynamics of residential building j in load aggregator i; represents the maximum HVAC system input power of residential building j in load aggregator i; It represents the upper limit of electricity demand of residential building j in load aggregator i during the grid service stage.

[0112] The objective function is:

[0113]

[0114] in: represents the expectation operator, acting on uncertain system parameters (such as fixed load b i,j,t , Photovoltaic output i,j,t , outdoor temperature T t out and buying and selling electricity prices, etc.); C i,j,1,t and C i,j,2,t They represent the operating cost of residential building j and the loss cost of energy storage system in load aggregator i at time slot t respectively, and the calculation formula is as follows:

[0115]

[0116] C i,j,2,t =ψ(c i,j,t +d i,j,t ),

[0117] Where: s t and w t They represent the purchase price and the selling price of electricity respectively; ψ represents the depreciation coefficient of the energy storage system.

[0118] 1.2. In this embodiment, the two-level optimization problem of the coordinated operation of grid-interactive residential buildings in the grid service stage includes decision variables, constraints and objective functions;

[0119] The decision variable is the peak power demand limit ρ at time slot t during the grid service phase t .

[0120] The constraints include:

[0121] 0≤ρ t ≤ρ max , (12)

[0122]

[0123]

[0124] c i,j,t ·d i,j,t =0, (20)

[0125] P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (twenty one)

[0126]

[0127] Where: maxIndicates the maximum peak power demand limit; represents the optimal power demand of load aggregator i at time slot t; P limit,t represents the sum of the power caps of all residential buildings in time slot t; represents the optimal power demand of each building in time slot t; X1 represents the initial decision amount before the proposed binary search; α represents a weight coefficient;

[0128] The objective function is:

[0129]

[0130] Where: Γ(·) represents the economic compensation related to the change in electricity demand of all grid-interactive residential buildings. In the embodiment, It is the economic compensation coefficient.

[0131] 2. A physically consistent neural network-assisted hierarchical model predictive control method is proposed to solve the above-mentioned operation cost minimization problem;

[0132] In this embodiment, the specific design is as follows:

[0133] 2.1. The proposed grid-interactive high-efficiency residential building operation cost minimization problem in the normal operation phase is transformed into the upper indoor comfort temperature deviation minimization sub-problem and the lower operation cost minimization sub-problem;

[0134] 2.1.1. The upper indoor comfort temperature deviation minimization sub-problem is as follows:

[0135]

[0136] Where: X represents the prediction length, t′ represents the corresponding time slot within the prediction length, C i,j,3,t′ Indicates the indoor comfort temperature deviation value; It is a dynamic model of indoor thermal comfort based on physical consistency neural network, which is mainly composed of black box module and physical module. The black box module is composed of a deep neural network, whose input is outdoor light intensity and output is indoor temperature disturbance T caused by light intensity. b,i,j,t′+1 ; The physical module is composed of the indoor temperature T of the current time slot i,j,t′ , the outdoor temperature at the current time slot Current time slot HVAC system input power e i,j,t′ It can be represented by a clear physical linear model, namely Where: γ and ν represent the learnable parameters that ensure the physical consistency of the neural network.

[0137] 2.1.2. The sub-problem of minimizing the lower-level operation cost is as follows:

[0138]

[0139] c i,j,t′ ·d i,j,t′ =0, (32)

[0140] P i,j,t′ +r i,j,t′ +d i,j,t′ =b i,j,t′ +e i,j,t′ +c i,j,t′ , (33)

[0141]

[0142] Where: C i,j,1,t′ and C i,j,2,t′ They respectively represent the operating cost of each building and the loss cost of the energy storage system in the time slot t′ within the prediction length.

[0143] 2.2. Deploy the upper-level model predictive control method assisted by the indoor thermal dynamic model based on the physical consistency neural network to solve the problem of minimizing the indoor comfort temperature deviation, and use the gradient descent method to solve the optimal HVAC system decision for the current time slot and the future X-1 time slots, that is, The specific solution process is as follows: First, set the current iteration step to k = 0 and the maximum iteration step to k. max :

[0144] 2.2.1. Predictive control output decision based on upper-level model The proposed physical embedded neural network predicts

[0145] 2.2.2. Calculate the indoor comfort deviation between the current time slot and the next X-1 time slots

[0146] 2.2.3 Based on and The gradient is calculated according to the following formula:

[0147]

[0148] 2.2.4. Based on the obtained gradient, the gradient optimizer optimizes the upper-level decision by minimizing the gradient. The specific calculation formula is as follows:

[0149]

[0150] in: and are the exponential moving average of the gradient and the exponential moving average of the squared gradient, respectively; β and For updating and Hyperparameters of and They are and bias-corrected estimates of ; express The function is used to update the decision of the current iteration step; ζ and ∈ are the learning rate and hyperparameters respectively.

[0151] 2.2.4, if the current And k<k max , k=k+1 and jump to iterate again; otherwise, the iteration ends and the optimal HVAC system decision is output

[0152] 2.3. Input the optimal decision of the upper-level model predictive control method into the lower-level model predictive control, and deploy the lower-level model predictive control method to solve the sub-problem of minimizing the operating cost. By introducing four auxiliary variables θ i,j,t′ ,κ i,j,t′ ,I c,t′ ,I d,t′ Transform (P4) into the following mixed integer linear programming problem (P5) to solve the optimal energy storage charging and discharging decision for the current time slot and the next X-1 time slots, namely:

[0153]

[0154] st(28)-(29),(43)

[0155]

[0156] I c,t′ +I d,t′ ≤1, (46)

[0157] I c,t′ ,I d,t′ ∈{0,1},(47)

[0158]

[0159] θ i,j,t ≥0, (50)

[0160] κ i,j,t ≥0, (51)

[0161] in: make and P i,j,t′ =θ i,j,t′ -κ i,j,t′ and |P i,j,t′ |=θ i,j,t′ +κ i,j,t′; The decision variable of (P5) is c i,j,t′ ,d i,j,t′ ,θ i,j,t′ ,κ i,j,t′ ,I c,t′ ,I d,t′ .

[0162] 3. A distributed collaborative operation optimization algorithm based on physically consistent neural network-assisted hierarchical model predictive control guided binary search is proposed to solve the above-mentioned collaborative operation two-level optimization problem;

[0163] In this embodiment, if the current operation time slot t is in the grid service stage, firstly, the initial decision X1 of the current time slot is obtained according to the operation optimization method based on physical consistency neural network assisted hierarchical model predictive control; then the upper limit of the current power demand peak limit is set to ρ u =ρ max , the current lower limit of the peak power demand is ρ l =ρ min , is a threshold. And the current iteration step k=0 is less than the maximum number of iterations k max , iterate according to the following steps to solve the optimal peak power demand limit and optimal decision

[0164] 3.1. Let the current peak power demand limit be ρ t =(ρ u +ρ l ) / 2;

[0165] 3.2. Distributed system operators will t Transmitted to all load aggregators i;

[0166] 3.3、Load aggregator will t Transmitted to all grid-interactive and efficient residential buildingsj;

[0167] 3.4. According to the following rule-based decision adjustment method, all buildings are adjusted X1 to obtain the optimal decision

[0168] make

[0169] 3.4.1 If and but

[0170] 3.4.2 If and but

[0171] 3.4.3 If and but

[0172] 3.4.4 If and and but

[0173] 3.4.5 If and and but

[0174] 3.5. All buildings will be optimally decided Transmitted to the corresponding load aggregator;

[0175] 3.6 Distributed system operators obtain optimal power demand

[0176] 3.7. If the current Then l =ρ t On the contrary, ρ u =ρ t ;

[0177] 3.8 If And k<k max , return to (8.1) and continue iteration, k = k + 1; otherwise, end the iteration and output the current optimal ρ t and optimal decision

[0178] 4. Deploy the proposed collaborative operation optimization method in the actual system.

[0179] In this embodiment, the actual operation system includes a grid-interactive high-efficiency residential building operation module and a real-time decision-making module based on hierarchical model predictive control assisted binary search.

[0180] The grid-interactive efficient residential building operation module is responsible for receiving the optimal decision output by the real-time decision-making module and executing it in each grid-interactive efficient residential building.

[0181] The real-time decision-making module based on hierarchical model predictive control assisted binary search includes an information acquisition submodule, an upper-level model predictive control real-time decision-making submodule, a lower-level model predictive control real-time decision-making submodule, a binary search collaborative operation optimization submodule and an optimal control strategy output submodule.

[0182] The information collection submodule is responsible for collecting the latest status of each component after each grid-interactive efficient residential building executes the decision, such as the indoor temperature of the building and the real-time storage capacity of the energy storage system. Then, it is sent to the upper and lower model predictive control real-time decision submodules respectively.

[0183] The upper-layer model predictive control real-time decision submodule reads the indoor temperature status sent by the information acquisition submodule. Then, the operation optimization method is used to solve the optimal HVAC system decision. Finally, the optimal HVAC system decision is sent to the lower-layer model predictive control real-time decision submodule.

[0184] The lower-layer model predictive control real-time decision submodule reads the energy storage system state issued by the information acquisition submodule and the optimal HVAC system decision issued by the upper-layer model predictive control real-time decision submodule. Then, the optimal energy storage system decision is solved using the operation optimization method. Finally, the optimal HVAC system decision and the energy storage system decision are transmitted to the binary search collaborative operation optimization module.

[0185] The binary search collaborative operation optimization submodule first determines whether the current time slot requires each grid-interactive efficient residential building to participate in grid services. If it is necessary to participate in grid services, the collaborative operation optimization method is used to further solve the optimal decision, and the latest decision is sent to the optimal control strategy output submodule; if it is not necessary to participate in grid services in the current time slot, the optimal decision of the upper and lower model predictive control real-time decision submodules is sent to the optimal control strategy output submodule.

[0186] The optimal control strategy output submodule reads the decision issued by the binary search collaborative operation optimization submodule and sends it to the grid-interactive efficient residential building operation module.

[0187] In order to demonstrate the effectiveness of the method of the present invention, four comparative schemes were introduced.

[0188] Solution 1 adopts a rule-based HVAC system operation strategy. In cooling mode, when the indoor temperature is higher than the upper limit of the comfortable temperature range, the input power of the HVAC system is set to the maximum; when the indoor temperature is lower than the lower limit of the comfortable temperature range, the HVAC system is turned off, that is, the input power is set to 0; when the indoor temperature is within the comfortable temperature range, the HVAC system input power remains unchanged.

[0189] Scheme 2 uses a deep reinforcement learning algorithm based on proximal policy optimization to train each residential building agent to optimize the input power decision of the HVAC system and the charging and discharging power decision of the energy storage system. Specifically, the state of each residential building agent in Scheme 2 includes outdoor temperature, indoor temperature, light intensity and fixed load; the action design is the input power of the HVAC system and the charging and discharging power of the energy storage system; the reward function consists of three items: operating cost, depreciation cost of the energy storage system, and indoor comfort temperature deviation.

[0190] Solution 3 uses a deep reinforcement learning algorithm based on double-delayed deep deterministic policy gradient. The state, action and reward function design of each agent in this solution are the same as those in Solution 2. It is worth noting that none of the above three solutions consider the coordinated operation in the power grid service stage.

[0191] Scheme 4 adopts the physical consistency neural network assisted hierarchical model predictive control method proposed in the method of the present invention to optimize the operation of the grid-interactive residential building system during the normal operation stage, but does not use the collaborative algorithm based on the binary search assisted by the above-mentioned operation optimization method proposed in the method of the present invention to optimize the operation of the grid service stage.

[0192] Figure 2-Figure 4 is a performance comparison chart of the method of the present invention and the comparative scheme, and Table 1 is a summary of the performance indicators of all schemes.

[0193] Table 1 Summary of performance indicators of the method of the present invention and other methods

[0194] Comparison plan Total operating cost ($) Average temperature deviation (℃) Average power exceeds the limit (%) Economic compensation (kW) Solution 1 1380.084 0.451 34.474 0 Solution 2 1269.773 0 8.429 0 Option 3 1106.086 0.002 7.981 0 Option 4 843.466 0 41.595 0 Method of the present invention 803.690 0.042 6.724 453.463

[0195] like Figure 2 As shown, compared with all other solutions, the method of the present invention can reduce the total operating cost by 4.716%-41.765% (assuming that the economic compensation for each household is 0.1 US dollars per kilowatt), which has a high operating cost saving potential and economic benefits. Figure 3 As shown, the average temperature deviation of the method of the present invention is better than that of the comparative scheme 1, and can reduce the average temperature deviation by 90.687%. Although the average temperature deviation of schemes 2, 3 and 4 is slightly better than that of the method of the present invention, the method of the present invention and the above three schemes can control the average temperature deviation below 0.1°C, which also meets the user's demand for indoor thermal comfort. Moreover, the operation method adopted in scheme 4 is the method proposed in the method of the present invention in the normal operation stage. Without considering the grid service, the average temperature deviation can be controlled to 0, which also illustrates the effectiveness of the method of the present invention in meeting the comfort needs of indoor users. Figure 4 As shown, compared with all the comparative schemes, the method of the present invention can reduce the power limit over-limit by 15.750%-83.835%, and better meet the service demand of the power grid.

Claims

1. A contribution-aware grid-interactive residential building collaborative operation optimization method, characterized in that: The collaborative operation optimization method comprises the following steps: Step 1: Considering the economic compensation when grid-interactive efficient residential buildings provide grid services, the operation cost minimization problem of grid-interactive efficient residential buildings in the normal operation stage and the two-layer optimization problem of grid-interactive efficient residential buildings coordinated operation in the grid service stage are modeled respectively; Step 2: Propose an operation optimization method based on physically consistent neural network assisted hierarchical model predictive control to solve the problem of minimizing the operation cost of grid-interactive high-efficiency residential buildings during normal operation; Step 3: Propose a distributed collaborative operation optimization algorithm based on physically consistent neural network-assisted hierarchical model predictive control guided binary search to solve the grid-interactive efficient residential building collaborative operation two-level optimization problem in the grid service stage; Step 4: Deploy the proposed collaborative operation method in the actual system of grid-interactive efficient residential buildings.

2. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 1, characterized in that: The operation cost minimization problem of the grid-interactive high-efficiency residential building in the normal operation stage modeled in step 1 includes decision variables, constraints and objective functions; Decision variables in the operating cost minimization problem The energy storage system input power c of residential building j managed by load aggregator i in time slot t i,j,t , energy storage system output power d i,j,t 、HVAC system input power e i,j,t and the electricity demand of residential buildings P i,j,t ; The constraints in the running cost minimization problem include: c i,j,t ·d i,j,t =0, (5) P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (6) Among them: B i,j,t represents the storage capacity of the energy storage system of residential building j in load aggregator i at time slot t; σ c,i,j and σ d,i,j They represent the input efficiency and output efficiency of the energy storage system respectively; and They represent the minimum storage capacity and maximum storage capacity of the energy storage system respectively; and They represent the maximum input power and maximum output power of the energy storage system respectively; Δt represents the time slot length; r i,j,t represents the photovoltaic power output at time slot t, b i,j,t represents the fixed load of time slot t; T i,j,t represents the indoor temperature of residential building j in load aggregator i at time slot t, T t out represents the outdoor temperature at time slot t, ξ i,j,t represents the indoor thermal disturbance of residential building j in load aggregator i at time slot t, represents the indoor thermal dynamics of residential building j in load aggregator i; represents the maximum HVAC system input power of residential building j in load aggregator i; represents the upper limit of electricity demand of residential building j in load aggregator i during the grid service phase; The objective function in the running cost minimization problem is: in: represents the expectation operator acting on the uncertain system parameters, including the fixed load b i,j,t , Photovoltaic output i,j,t , outdoor temperature T t out and buying and selling electricity prices; C i,j,1,t and C i,j,2,t They represent the operating cost of residential building j in load aggregator i at time slot t and the loss cost of the energy storage system respectively.

3. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 2, characterized in that: The two-level optimization problem of the coordinated operation of the grid-interactive efficient residential buildings in the grid service stage established in step 1 includes decision variables, constraints and objective functions; The decision variables in the collaborative operation bi-level optimization problem are the peak power demand limit of each building in the grid service phase t time slot ρ t ; The constraints in the co-operating bi-level optimization problem include: 0≤ρ t ≤ρ max , (12) c i,j,t ·d i,j,t =0, (20) P i,j,t +r i,j,t +d i,j,t =b i,j,t +e i,j,t +c i,j,t , (21) Where: max Indicates the maximum peak power demand limit; represents the optimal power demand of load aggregator i at time slot t; P limit,t represents the sum of the power caps of all residential buildings in time slot t; represents the optimal power demand of each building in time slot t; represents the initial decision amount before the proposed binary search; α represents a weight coefficient; The objective function in the collaborative two-level optimization problem is: Where: Γ(·) represents the economic compensation associated with the change in electricity demand of all grid-interactive residential buildings.

4. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 3, characterized in that: The operation optimization method based on physical consistency neural network assisted hierarchical model predictive control proposed in step 2 is designed as follows: (4.1) The proposed grid-interactive high-efficiency residential building operation cost minimization problem in the normal operation phase is transformed into an upper-layer indoor comfort temperature deviation minimization sub-problem and a lower-layer operation cost minimization sub-problem; (4.2) Deploy the upper-level model predictive control method assisted by the indoor thermal dynamic model based on the physical consistency neural network to solve the indoor comfort temperature deviation minimization problem and obtain the upper-level optimal decision for time slot t (4.3) The optimal decision of the upper-level model predictive control method is input into the lower-level model predictive control, and the lower-level model predictive control method is deployed to solve the sub-problem of minimizing the operating cost, and the optimal decision of the lower level in time slot t is obtained.

5. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 4, characterized in that: The upper indoor comfort temperature deviation minimization subproblem proposed in step 4.1 is as follows: Where: X represents the prediction length, t′ represents the corresponding time slot within the prediction length, C i,j,3,t′ Indicates the indoor comfort temperature deviation value; It is a dynamic model of indoor thermal comfort based on physical consistency neural network, which consists of a black box module and a physical module. The black box module consists of a deep neural network, whose input is outdoor light intensity and output is the indoor temperature disturbance ξ caused by light intensity. i,j,t′ ; The physical module consists of the indoor temperature T of the current time slot i,j,t′ , the outdoor temperature at the current time slot HVAC system input power e in the current time slot i,j,t′ composition.

6. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 5, characterized in that: The sub-problem of minimizing the lower-level operation cost proposed in step 4.1 is as follows: c i,j,t′ ·d i,j,t′ =0, (32) P i,j,t′ +r i,j,t′ +d i,j,t′ =b i,j,t′ +e i,j,t′ +c i,j,t′ , (33) Where: C i,j,1,t′ and C i,j,2,t′ They respectively represent the operating cost of each building and the loss cost of the energy storage system in the time slot t′ within the prediction length.

7. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 6, characterized in that: The upper model predictive control method deployed in step 4.2 uses the gradient descent method to solve the optimal HVAC system decision for the current time slot and the next X-1 time slots, that is, The lower-level model predictive control method deployed in step 4.3 solves the optimal energy storage charging and discharging decision for the current time slot and the next X-1 time slots by converting the proposed (P4) into a mixed integer linear programming problem, that is, 8. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 7, characterized in that: The distributed collaborative operation optimization algorithm based on hierarchical model predictive control assisted binary search proposed in step 3 in the power grid service stage is designed as follows: If the current operation time slot t is in the grid service stage, firstly, the initial decision of the current time slot is obtained according to the operation optimization method based on hierarchical model predictive control Then set the current peak power demand limit as ρ u =ρ max , the current lower limit of the peak power demand is ρ l =ρ min , is a threshold; when And the current iteration step k=0 is less than the maximum number of iterations k max , iterate according to the following steps to solve the optimal peak power demand limit and optimal decision (8.1) Let the current peak power demand limit be ρ t =(ρ u +ρ l ) / 2; (8.2) The distributed system operator will t Transmitted to all load aggregators i; (8.3) The load aggregator will t Transmitted to all grid-interactive and efficient residential buildingsj; (8.4) According to the rule-based decision adjustment method, all building adjustments Get the best decision (8.5) All buildings will be optimally decided Transmitted to the corresponding load aggregator; (8.6) Distributed system operators obtain optimal power demand (8.7) If the current Then l =ρ t On the contrary, ρ u =ρ t ; (8.8) If And k<k max , return to (8.1) and continue iteration, k = k + 1; otherwise, end the iteration and output the current optimal ρ t and optimal decision 9. A contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 8, characterized in that: The rule-based decision adjustment method proposed in step 8.4 is designed as follows: (9.1) If and but (9.2) If and but (9.3) If and but (9.4) If and and but (9.5) If and and but 10. The contribution-aware grid-interactive residential building collaborative operation optimization method according to claim 1, characterized in that: The actual operation system in which the proposed collaborative operation optimization method is deployed in step 4 includes a grid-interactive efficient residential building operation module and a real-time decision-making module based on hierarchical model predictive control assisted binary search; The grid-interactive efficient residential building operation module is responsible for receiving the optimal decision output by the real-time decision module and executing it in each grid-interactive efficient residential building; The real-time decision-making module based on hierarchical model predictive control assisted binary search includes an information acquisition submodule, an upper-layer model predictive control real-time decision-making submodule, a lower-layer model predictive control real-time decision-making submodule, a binary search collaborative operation optimization submodule, and an optimal control strategy output submodule; The information collection submodule is responsible for collecting the latest status of each component after each grid-interactive efficient residential building executes the decision, including the indoor temperature of the building and the real-time storage capacity of the energy storage system; then, it is sent to the upper and lower model predictive control real-time decision submodules respectively; The upper-layer model predictive control real-time decision submodule reads the indoor temperature state sent by the information acquisition submodule; then, the operation optimization method is used to solve the optimal HVAC system decision; finally, the optimal HVAC system decision is sent to the lower-layer model predictive control real-time decision submodule; The lower-layer model predictive control real-time decision submodule reads the energy storage system state issued by the information acquisition submodule and the optimal HVAC system decision issued by the upper-layer model predictive control real-time decision submodule. Then, the optimal energy storage system decision is solved using the operation optimization method; finally, the optimal HVAC system decision and the energy storage system decision are transmitted to the binary search collaborative operation optimization module; The binary search collaborative operation optimization submodule first determines whether the current time slot requires each grid-interactive efficient residential building to participate in grid service; if it is necessary to participate in grid service, the collaborative operation optimization method is used to further solve the optimal decision, and the latest decision is sent to the optimal control strategy output submodule; if it is not necessary to participate in grid service in the current time slot, the optimal decision of the upper and lower model predictive control real-time decision submodules is sent to the optimal control strategy output submodule; The optimal control strategy output submodule reads the decision issued by the binary search collaborative operation optimization submodule and sends it to the grid-interactive efficient residential building operation module.

Images