Reliable and economical operation optimization method for hydrogen-containing building energy system in off-grid mode

Abstract

The invention discloses a reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode, and the method comprises the following steps: (1) building an operation optimization problem of the hydrogen-containing building energy system in the off-grid mode on the premise of maintaining an indoor comfortable environment; (2) proposing a building indoor environment dynamic model based on a physical embedded neural network; (3) proposing a hydrogen-containing building energy system operation optimization method based on physical embedded neural network assisted hierarchical model predictive control to solve the modeling problem; and (4) deploying the hydrogen-containing building energy system operation optimization method in an actual operation system. Compared with an existing method, the indoor user comfort control under the three energy demands of electricity, heat and cold is considered, the operation cost of the system can be effectively reduced, and the reliability of the system is improved.

Description

Technical Field

[0001] The present invention relates to a reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode, and belongs to the intersection field of building energy systems and artificial intelligence. Background Art

[0002] Due to the intermittent and volatile characteristics of renewable energy (such as wind power and photovoltaics), the large-scale grid connection of renewable energy will have a negative impact on the efficiency, reliability, stability and safety of energy system operation. In this context, hydrogen-containing energy systems have received widespread attention. Specifically, the hydrogen energy storage system in the hydrogen-containing energy system can convert excess renewable energy into hydrogen for storage, and can consume hydrogen for energy supply when energy is scarce. As a clean energy, hydrogen energy has many advantages in developing energy storage systems, such as flexible storage and transportation methods, large storage capacity and zero pollution. Moreover, the efficient coordination of hydrogen energy storage systems and other energy storage systems in hydrogen-containing energy systems will further improve energy efficiency and promote the use of renewable energy. Therefore, it is of great significance to design efficient operation algorithms for hydrogen-containing energy systems.

[0003] In existing research, many methods have been proposed to optimize the operation of hydrogen-containing energy systems, mainly including model-based methods (such as genetic algorithms, model predictive control, particle swarm optimization, Benders decomposition, and robust optimization) and learning-based methods (such as deep deterministic policy gradient and multi-agent double-delay deterministic policy gradient). At the same time, hydrogen-containing building energy systems have received widespread attention due to the huge energy consumption of buildings and the high demand flexibility they can provide. Many operation optimization methods have been proposed for hydrogen-containing building energy systems, such as multi-agent deep deterministic policy gradient, meta-heuristic multi-objective optimization, and Lyapunov optimization. Although the above methods have made some progress, they ignore the user comfort control of hydrogen-containing building energy systems under the three demands of electricity, heat, and cooling. In fact, efficient indoor user comfort control will reduce energy demand, thereby reducing the operating cost of hydrogen-containing building energy systems and improving system reliability. Therefore, it is of great significance to study the operation optimization of hydrogen-containing building energy systems considering indoor user comfort control. However, existing model-based methods require knowledge of an explicit dynamics model of the indoor environment, and existing learning-based methods require a large number of interactions with the real environment or an accurately simulated environment, which leads to high training costs and deployment barriers.

[0004] In summary, the existing operation optimization methods of hydrogen-containing building energy systems have shortcomings, and it is urgent to study a new method for reliable and economical operation optimization of hydrogen-containing building energy systems in off-grid mode, so as to reduce the operating cost of hydrogen-containing building energy systems and improve system reliability while maintaining a comfortable indoor environment. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode, which aims to achieve the joint optimization of the operating cost, system reliability and user comfort of the hydrogen-containing building energy system, to make up for the deficiency of the existing operation optimization method in not considering indoor comfort control, and to solve the technical problems that the existing model-based method requires a clear indoor environmental dynamics model and the existing learning-based method requires a large number of interactions with the real environment or an accurately simulated environment.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention provides a reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode, comprising:

[0008] (1) Modeling the operation optimization of hydrogen-containing building energy systems in off-grid mode while maintaining a comfortable indoor environment;

[0009] (2) Modeling the dynamics of indoor environments based on physically embedded neural networks;

[0010] (3) An operation optimization method based on physically embedded neural network-assisted hierarchical model predictive control is proposed to solve the above modeling problem;

[0011] (4) Deploy the proposed hydrogen-containing building energy system operation optimization method in the actual operation system;

[0012] Furthermore, the operation optimization problem of the hydrogen-containing building energy system in off-grid mode includes decision variables, constraints and objective functions.

[0013] The decision variable a t Including the building HVAC system air supply rate v in the time slot t t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the actual photovoltaic output power P pv,t , electrolyzer input power P el,t , fuel cell output power P fc,t , adsorption refrigerator input heat energy Q ac,t , the excess heat energy of the fuel cell Q w,t , ground source heat pump heating / cooling input power P gsh,t / P gsc,t , electric boiler input power P eb,t , hot water tank input / output thermal power q tc,t / q td,t , cold water tank input / output cooling power in c cc,t / ccd,t And the amount of hydrogen purchased m t .

[0014] The constraints include: building HVAC system operation constraints, photovoltaic operation constraints, hydrogen energy storage system operation constraints, hot water tank operation constraints, cold water tank operation constraints, electric boiler operation constraints, ground source heat pump operation constraints, adsorption chiller operation constraints and energy balance constraints.

[0015] The operating constraints of the building HVAC system are:

[0016]

[0017] T min ≤T t in ≤T max ,(3)

[0018]

[0019]

[0020] Where: T t in , W t in as well as They represent the indoor temperature, indoor humidity ratio and indoor relative humidity at time slot t respectively. The relative indoor humidity is a function of indoor temperature and indoor humidity. t out and W t out They represent the outdoor temperature and outdoor humidity ratio at time slot t respectively; e t represents the light intensity at time slot t; ξ t and ψ t represent the indoor thermal disturbance and humidity disturbance related to light, users and lighting respectively; and Respectively represent the dynamic model of indoor temperature and indoor humidity; T min and T max Respectively represent the lower and upper limits of indoor comfort temperature; and They represent the lower and upper limits of indoor comfortable relative humidity respectively; and They represent the indoor high temperature and high humidity discomfort boundary and high temperature and low humidity discomfort boundary respectively. The specific calculation formula is as follows: Where: β ht ,β hh ,δ ht ,δ lh ,γrepresent the coefficients of modeling the above boundaries; vmin and v max Respectively represent the minimum and maximum value of air supply speed adjustment; d min and d max Respectively represent the minimum and maximum values ​​of the air valve position adjustment; T ca,t and T ha,t represent the cooling temperature and heating temperature in time slot t, T ma,t represents the mixed air temperature at time slot t; W ca,t and W ha,t They represent the dehumidification humidity ratio and humidification humidity ratio at time slot t, respectively, and W ma,t represents the mixed air humidity ratio at time slot t, Indicates the maximum humidification humidity ratio; η fan Indicates the air supply fan coefficient; η wv is the latent heat of evaporation of water, η hum Indicates the efficiency of the humidification coil; η pa , η cc and COP cc They represent the specific heat capacity of air, cooling coil efficiency, and cooling coil performance coefficient respectively; Indicates the building HVAC operation time slot, and Represent summer and winter respectively; Δt represents the time slot length.

[0021] The photovoltaic operation constraints are:

[0022]

[0023] in: represents the maximum photovoltaic power generation in time slot t, and h pv They represent the photovoltaic power generation efficiency and illumination area in time slot t respectively; P pv,t Indicates the actual photovoltaic utilization in time slot t.

[0024] The operating constraints of the hydrogen energy storage system are:

[0025]

[0026] 0≤H t ≤H max ,(16)

[0027] P el,t ·P fc,t =0, (17)

[0028]

[0029] 0≤m t ≤m max ,(20)

[0030] Q fc,t =ω e2h ω hr P fc,t Δt, (21)

[0031] Where: H t represents the storage capacity of the hydrogen tank at time slot t; ω el and ω fc They represent the electrolyzer electricity-to-hydrogen conversion efficiency and the fuel cell hydrogen-to-electricity conversion efficiency respectively; P el,t and P fc,t Respectively represent the electrolyzer input power and fuel cell output power at time slot t; m t represents the amount of hydrogen purchased in time slot t; H max Indicates the maximum storage capacity of the hydrogen tank; and Respectively represent the maximum input power of the electrolyzer and the maximum output power of the fuel cell; m max Indicates the maximum amount of hydrogen purchased; Q fc,t represents the heat generated by the fuel cell at time slot t, ω e2h and ω hr They represent the fuel cell electrical-to-thermal conversion efficiency and heat recovery efficiency respectively.

[0032] The hot water tank operation constraints are:

[0033]

[0034]

[0035] q tc,t ·q td,t =0, (26)

[0036] Where: M hwt,t represents the storage capacity of the hot water tank at time slot t; ω tc and ω td They represent the thermal power input efficiency and output efficiency of the hot water tank respectively; q tc,t and q td,t denote the heat input power and heat output power in time slot t respectively; Indicates the maximum storage capacity of the hot water tank; and They represent the maximum input thermal power and maximum output thermal power of the hot water tank respectively.

[0037] The cold water tank operation constraints are:

[0038]

[0039] c cc,t ·c cd,t =0, (31)

[0040] Among them: U cwt,t represents the storage capacity of the cold water tank at time slot t; ω cc and ω cd They represent the cold water tank cooling power input efficiency and output efficiency respectively; c cc,t and c cd,t They represent the cold input power and cold output power in time slot t respectively; Indicates the maximum storage capacity of the cold water tank; and They represent the maximum input cooling power and maximum output cooling power of the cold water tank respectively.

[0041] The electric boiler operation constraints are:

[0042] Q eb,t =ω eb P eb,t Δt, (32)

[0043]

[0044] Where: Q eb,t and P eb,t They represent the output heat energy and input power of the electric boiler in time slot t respectively; ω eb Indicates the electricity-heat conversion efficiency of electric boilers; Indicates the maximum input power of the electric boiler.

[0045] The ground source heat pump furnace operation constraints are:

[0046] Q gs,t =ω gsh P gsh,t Δt, (34)

[0047] C gs,t =ω gsc P gsc,t Δt, (35)

[0048]

[0049] 0≤Q w,t ≤Q fc,t ,(40)

[0050] Q w,t ·C gs,t Q gs,t =0, (41)

[0051] Where: Q gs,t and C gs,t They represent the heat output and cold output of the ground source heat pump at time slot t respectively; P gsh,t and P gsc,t They represent the heating power input and cooling power input of the ground source heat pump at time slot t respectively; ωgsh and ω gsc They represent the electricity-heat conversion efficiency and electricity-cold conversion efficiency of the ground source heat pump respectively; Indicates the maximum input power of the ground source heat pump; G gw,t represents the storage capacity of the geothermal well at time slot t, Indicates the maximum storage capacity of the geothermal well; Q w,t Represents the thermal energy that a fuel cell injects into a geothermal well.

[0052] The adsorption chiller operation constraints are:

[0053] C ac,t =ω acc Q ac,t , (42)

[0054] P ac,t =ω ace C ac,t , (43)

[0055] 0≤Q ac,t ≤Q fc,t ,(44)

[0056] C ac,t =ω acc Q ac,t , (45)

[0057] P ac,t =ω ace C ac,t , (46)

[0058] Where: Q ac,t , C ac,t and P ac,t They represent the heat input, cold output and required power of the adsorption refrigerator in time slot t respectively; ω acc and ω ace They represent the heat-to-cold conversion efficiency and the cold-to-electricity conversion efficiency of the adsorption refrigerator respectively.

[0059] The energy balance constraint is:

[0060] E load,t +E hvac,t +(P el,t +P eb,t +P gsh,t +P gsc,t +P ac,t )Δt=(P pv,t +P fc,t )Δt+E loss,t , (47)

[0061] Q load,t +Q hvac,t +Qac,t +q tc,t Δt=Q eb,t +Q gs,t +Q fc,t +Q loss,t +q td,t Δt, (48)

[0062] C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (49)

[0063] C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (50)

[0064] 0≤E loss,t ≤E load,t +E hvac,t , (51)

[0065] 0≤Q loss,t ≤Q load,t +Q hvac,t , (52)

[0066] 0≤C loss,t ≤C load,t +C hvac,t , (53) where: E load,t , Q load,t and C load,t They represent the electricity, heat and cooling energy requirements caused by the operation of non-building HVAC systems in time slot t; E loss,t , Q loss,t and C loss,t They represent the electrical energy, thermal energy and cold energy losses in time slot t respectively.

[0067] The objective function is:

[0068]

[0069] Among them: F 1,t 、F 2,t 、F 3,t and F 4,t They represent the photovoltaic reduction cost, hydrogen energy system operation cost, hydrogen purchase cost and energy demand loss in time slot t respectively; L t Indicates the total time slot of operation; represents the expectation operator acting on uncertain system parameters (such as PV output, outdoor temperature, electricity / heating / cooling energy demand) and stochastic control decisions (i.e., v t ,d t ,T t ,w t ,P pv,t ,P el,t ,P fc,t ,Q ac,t ,m t ,P gsh,t ,P gsc,t ,P eb,t ,q tc,t ,q td,t ,c cc,t ,c cd,t ,Q w,t ); α pv Represents the photovoltaic cost reduction coefficient; α h2e and α c2e represents two coefficients; and represents the cost of operating, starting up, and shutting down electrolyzers and fuel cells; and Indicates the flags of operation, startup and shutdown; λ indicates the price of buying hydrogen;

[0070] Furthermore, the indoor environment dynamics model based on physical embedded neural network specifically includes two modules, namely black box module and physical module. The black box module is composed of a deep neural network, and the input is the outdoor light intensity e t , the output is the indoor temperature disturbance T caused by the light intensity b,t+1 and the indoor humidity ratio disturbance W b,t+1 ; The input of the physical module is the indoor temperature T t in , outdoor temperature T t out , indoor humidity ratio W t in , outdoor humidity ratio W t out , air supply rate v t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the output is the indoor temperature T which can be described by clear physical laws p,t+1 and indoor humidity ratio W p,t+1 Furthermore, the proposed indoor environment dynamics model based on physical embedded neural network is obtained by minimizing the loss function training method, and the designed reward function is as follows:

[0071]

[0072] in: and Respectively and The true value of and Respectively and The differential value with respect to time slot t.

[0073] Furthermore, the operation optimization method based on physical embedded neural network assisted hierarchical model predictive control is designed as follows:

[0074] (4.1) The proposed hydrogen-containing building energy system operation optimization problem is transformed into an upper-level indoor comfort deviation minimization subproblem and a lower-level operation cost and system reliability joint optimization subproblem;

[0075] Specifically, the remodeled upper floor indoor comfort deviation minimization subproblem is as follows:

[0076]

[0077] st(6)-(12),(62)

[0078] Among them: F 5,t′ represents the indoor comfort deviation penalty in time slot t′; t′ represents the corresponding time slot in the prediction length; h represents the length of the prediction time slot; Represents a physically embedded neural network model of indoor environment dynamics.

[0079] Specifically, the remodeled lower-level operating cost and system reliability joint optimization sub-problem is as follows:

[0080]

[0081] st(13)-(50),(64)

[0082] Where: α loss Represents a coefficient.

[0083] (4.2) If the current operation time slot t is the operation time slot of the building HVAC system, the upper-level model predictive control algorithm assisted by the indoor environmental dynamics model based on the physical embedded neural network is deployed to solve the indoor comfort deviation minimization problem and obtain the optimal upper-level decision for time slot t. in: represents the optimal upper-level decision; and They represent the optimal air supply rate, optimal air valve position, optimal air supply temperature, and optimal air supply humidity ratio. And calculate the various energy requirements of the building HVAC system; conversely, the HVAC system has no energy requirements;

[0084] Specifically, the deployed physical embedded neural network-assisted upper-level model predictive control algorithm uses the gradient descent method to solve the optimal upper-level decision for the current time slot and the next h-1 time slots, namely: The specific steps of using the gradient descent method to solve the optimal upper-level decision are as follows: first, set the gradient descent iteration target Current number of iterations k = 0, maximum number of iterations k max ; If the current time slot t is in the HVAC system operation time slot The upper-level algorithm reads the current state And output the upper-level decision a of the current time slot and the next h-1 time slots upp,t ,a upp,t+1 ,…,a upp,t+h-1 .like And k<k max , then enter the iteration:

[0085] (4.2a) Based on the upper decision a upp,t ,a upp,t+1 ,…,a upp,t+h-1 and The proposed physical embedded neural network predicts upp,t+1 ,…,s upp,t+h ;

[0086] (4.2b) Calculate the indoor comfort deviation between the current time slot and the next h-1 time slots

[0087] (4.2c) Based on with a upp,t ,a upp,t+1 ,…,a upp,t+h-1 , the gradient solver calculates the gradient;

[0088] (4.2d) Based on the obtained gradient, the gradient optimizer optimizes the upper-level decision by minimizing the gradient as the update direction; (4.2e) If the current And k<k max , k=k+1 and jump to (4.2a) to iterate again; otherwise, the iteration ends and the optimal upper decision is output

[0089] (4.3) The various energy demands of the building HVAC system obtained by the upper-level model predictive control algorithm are input into the lower-level model predictive control algorithm. The lower-level model predictive control algorithm solves the joint optimization sub-problem of operating cost and system reliability to obtain the optimal decision of the lower level in time slot t in: represents the optimal lower-level decision; and They respectively represent the optimal photovoltaic power generation utilization, the optimal electrolyzer input power, the optimal fuel cell output power, the optimal adsorption refrigerator input heat energy, the optimal hydrogen purchase amount, the optimal ground source heat pump input power, the optimal electric boiler input power, the optimal hot tank heat absorption power, the optimal hot tank heat release power, the optimal cold tank cold absorption power, the optimal cold tank cold release power and the optimal fuel cell waste heat injection into the geothermal well.

[0090] Specifically, the deployed lower-level model predictive control method mainly solves the optimal lower-level decision for the current time slot and the next X-1 time slots by converting the proposed (P3) into a mixed integer linear programming problem (P4), that is, The mixed integer linear programming problem (P4) is specifically as follows:

[0091]

[0092] st(13)-(50)(66)

[0093]

[0094]

[0095] 0≤Q w,t′ ≤Q fc,t′ K wh,t′ , (75)

[0096] K el,t′ +K fc,t′ ≤1, (76)

[0097] K gsh,t′ +K gsc,t′ +K wh,t′ ≤1, (77)

[0098] K tc,t′ +K td,t′ ≤1, (78)

[0099] K cc,t′ +K cd,t′ ≤1, (79)

[0100] K el,t′ ,K fc,t′ ,K tc,t′ ,K td,t′ ,K cc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ ∈{0,1} (80) where: K el,t′ ,K fc,t′ ,K tc,t′ ,K td,t′ ,Kcc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ is an auxiliary variable.

[0101] (4.4) Execute the optimal decision of the upper layer Optimal decision making In the operating environment of the hydrogen-containing building energy system, the environment feedbacks the next time slot state s upp,t+1 ,s low,t+1 . Let t = t + 1, if t < L t , then jump to 4.1 to continue running; otherwise, the running ends;

[0102] Furthermore, the actual operation system for deploying the proposed operation optimization method includes an operation environment module for a hydrogen-containing building energy system and a real-time decision-making module based on physical embedded neural network-assisted hierarchical model predictive control.

[0103] The hydrogen-containing building energy system operating environment module is responsible for receiving the optimal decision output by the real-time decision module and executing it on the hydrogen-containing building energy system.

[0104] The real-time decision-making module based on physical embedded neural network assisted hierarchical model predictive control 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 and an optimal control strategy output submodule.

[0105] The information collection submodule is responsible for collecting the dynamic changes of each component after the hydrogen-containing building energy system executes the decision, such as the indoor temperature of the building, indoor relative humidity, indoor humidity ratio, real-time storage capacity of hydrogen tanks / hot water tanks / cold water tanks / geothermal wells, etc. Then, the latest status of each component is sent to the upper and lower model predictive control real-time decision submodules respectively.

[0106] The upper-level model predictive control real-time decision submodule reads the indoor environmental status of the building, i.e., the indoor temperature, indoor relative humidity, and indoor humidity ratio, issued by the information acquisition submodule. Then, the optimal upper-level decision is solved using the operation optimization method, and the multiple energy requirements of the building HVAC system are calculated. Finally, the optimal upper-level decision and the multiple energy requirements of the building HVAC system are sent to the optimal control strategy output submodule and the lower-level model predictive control real-time decision submodule, respectively.

[0107] The lower-level model predictive control real-time decision submodule reads the energy storage system status (i.e., the real-time storage volume of hydrogen tanks / hot water tanks / cold water tanks / geothermal wells) issued by the information acquisition submodule and the various energy requirements of the HVAC system issued by the upper-level model predictive control real-time decision submodule. Then, the optimal lower-level decision is solved using the operation optimization method. Finally, the optimal lower-level decision is transmitted to the optimal control strategy output submodule.

[0108] The optimal strategy output module reads the optimal decisions issued by the upper and lower model predictive control real-time decision submodules, respectively, and then sends the optimal upper and lower decisions to the hydrogen-containing building energy system operating environment module.

[0109] Beneficial effects: A reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode. Compared with the prior art, the present invention takes into account the indoor user comfort control of a hydrogen-containing building energy system in an off-grid mode under three load demands of electric energy, thermal energy, and cold energy, and the beneficial effects achieved by the present invention are as follows:

[0110] (1) Compared with the existing model-based methods, the method proposed in the present invention does not need to know a clear indoor environment dynamic model. By constructing an indoor environment dynamic model based on a physical embedded neural network, it accurately captures the indoor thermal dynamics and humidity 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.

[0111] (2) Compared with existing learning-based methods, the method proposed in the present invention does not require a large number of interactions with the real environment or a high-precision simulation environment, and has the advantages of low training cost and easy deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1 It is a flow chart of the reliable and economical operation optimization method of the hydrogen-containing building energy system in the off-grid mode proposed by the present invention;

[0113] Figure 2 It is a comparison chart of the operating costs of the method of the present invention and other methods;

[0114] Figure 3 is a comparison chart of average system energy demand losses between the method of the present invention and other methods;

[0115] Figure 4 is a comparison diagram of the average indoor environment comfort deviation between the method of the present invention and other methods;

[0116] Figure 5 It is a summary diagram of performance indicators of the method of the present invention and other methods. DETAILED DESCRIPTION

[0117] 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.

[0118] Embodiment 1:

[0119] like Figure 1 As shown, the design flow chart of the reliable and economical operation optimization method of the hydrogen-containing building energy system in the off-grid mode provided by the present invention includes the following steps:

[0120] Step 1: Model the operation optimization problem of hydrogen-containing building energy system in off-grid mode while maintaining indoor comfortable environment;

[0121] Step 2: Modeling the indoor environment dynamics model based on physical embedded neural network;

[0122] Step 3: Propose an operation optimization method based on physically embedded neural network assisted hierarchical model predictive control to solve the above modeling problem;

[0123] Step 4: Deploy the proposed hydrogen-containing building energy system operation optimization method in the actual operation system;

[0124] In the above step 1, the operation optimization problem of the hydrogen-containing building energy system in off-grid mode includes decision variables, constraints and objective functions.

[0125] The decision variable a t Including the building HVAC system air supply rate v in the time slot t t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the actual photovoltaic output power P pv,t , electrolyzer input power P el,t , fuel cell output power P fc,t , adsorption refrigerator input heat energy Q ac,t , the excess heat energy of the fuel cell Q w,t , ground source heat pump heating / cooling input power P gsh,t / P gsc,t , electric boiler input power P eb,t , hot water tank input / output thermal power q tc,t / q td,t , cold water tank input / output cooling power in c cc,t / c cd,t And the amount of hydrogen purchased m t .

[0126] The constraints include: building HVAC system operation constraints, photovoltaic operation constraints, hydrogen energy storage system operation constraints, hot water tank operation constraints, cold water tank operation constraints, electric boiler operation constraints, ground source heat pump operation constraints, adsorption chiller operation constraints and energy balance constraints.

[0127] The operating constraints of the building HVAC system are:

[0128]

[0129] T min ≤T t in ≤T max ,(3)

[0130]

[0131]

[0132] Where: T t in , W t in as well as They represent the indoor temperature, indoor humidity ratio and indoor relative humidity at time slot t respectively. The relative indoor humidity is a function of indoor temperature and indoor humidity. t out and W t out They represent the outdoor temperature and outdoor humidity ratio at time slot t respectively; e t represents the light intensity at time slot t; ξ t and ψ t represent the indoor thermal disturbance and humidity disturbance related to light, users and lighting respectively; and Respectively represent the dynamic model of indoor temperature and indoor humidity; T min and T max Respectively represent the lower and upper limits of indoor comfort temperature; and They represent the lower and upper limits of indoor comfortable relative humidity respectively; and They represent the indoor high temperature and high humidity discomfort boundary and high temperature and low humidity discomfort boundary respectively. The specific calculation formula is as follows: Where: β ht ,β hh ,δ ht ,δ lh ,γrepresent the coefficients of modeling the above boundaries; v min and v max Respectively represent the minimum and maximum value of air supply speed adjustment; dmin and d max Respectively represent the minimum and maximum values ​​of the air valve position adjustment; T ca,t and T ha,t represent the cooling temperature and heating temperature in time slot t, T ma,t represents the mixed air temperature at time slot t; W ca,t and W ha,t They represent the dehumidification humidity ratio and humidification humidity ratio at time slot t, respectively, and W ma,t represents the mixed air humidity ratio at time slot t, Indicates the maximum humidification humidity ratio; η fan Indicates the air supply fan coefficient; η wv is the latent heat of evaporation of water, η hum Indicates the efficiency of the humidification coil; η pa , η cc and COP cc They represent the specific heat capacity of air, cooling coil efficiency, and cooling coil performance coefficient respectively; Indicates the building HVAC operation time slot, and Represent summer and winter respectively; Δt represents the time slot length.

[0133] The photovoltaic operation constraints are:

[0134]

[0135]

[0136] in: represents the maximum photovoltaic power generation in time slot t, and h pv They represent the photovoltaic power generation efficiency and illumination area in time slot t respectively; P pv,t Indicates the actual photovoltaic utilization in time slot t.

[0137] The operating constraints of the hydrogen energy storage system are:

[0138]

[0139] 0≤H t ≤H max , (16)

[0140] P el,t ·P fc,t =0, (17)

[0141]

[0142] 0≤m t ≤m max , (20)

[0143] Q fc,t =ω e2h ω hr P fc,t Δt, (21)

[0144] Where: H t represents the storage capacity of the hydrogen tank at time slot t; ω el and ω fc They represent the electrolyzer electricity-to-hydrogen conversion efficiency and the fuel cell hydrogen-to-electricity conversion efficiency respectively; P el,t and P fc,t Respectively represent the electrolyzer input power and fuel cell output power at time slot t; m t represents the amount of hydrogen purchased in time slot t; H max Indicates the maximum storage capacity of the hydrogen tank; and Respectively represent the maximum input power of the electrolyzer and the maximum output power of the fuel cell; m max Indicates the maximum amount of hydrogen purchased; Q fc,t represents the heat generated by the fuel cell at time slot t, ω e2h and ω hr They represent the fuel cell electrical-to-thermal conversion efficiency and heat recovery efficiency respectively.

[0145] The hot water tank operation constraints are:

[0146]

[0147] q tc,t ·q td,t =0, (26)

[0148] Where: M hwt,t represents the storage capacity of the hot water tank at time slot t; ω tc and ω td They represent the thermal power input efficiency and output efficiency of the hot water tank respectively; q tc,t and q td,t denote the heat input power and heat output power in time slot t respectively; Indicates the maximum storage capacity of the hot water tank; and They represent the maximum input thermal power and maximum output thermal power of the hot water tank respectively.

[0149] The cold water tank operation constraints are:

[0150]

[0151] c cc,t ·c cd,t =0, (31)

[0152] Among them: U cwt,t represents the storage capacity of the cold water tank at time slot t; ωcc and ω cd They represent the cold water tank cooling power input efficiency and output efficiency respectively; c cc,t and c cd,t They represent the cold input power and cold output power in time slot t respectively; Indicates the maximum storage capacity of the cold water tank; and They represent the maximum input cooling power and maximum output cooling power of the cold water tank respectively.

[0153] The electric boiler operation constraints are:

[0154] Q eb,t =ω eb P eb,t Δt, (32)

[0155]

[0156] Where: Q eb,t and P eb,t They represent the output heat energy and input power of the electric boiler in time slot t respectively; ω eb Indicates the electricity-heat conversion efficiency of electric boilers; Indicates the maximum input power of the electric boiler.

[0157] The ground source heat pump furnace operation constraints are:

[0158] Q gs,t =ω gsh P gsh,t Δt, (34)

[0159] C gs,t =ω gsc P gsc,t Δt, (35)

[0160]

[0161]

[0162] 0≤Q w,t ≤Q fc,t ,(40)

[0163] Q w,t ·C gs,t Q gs,t =0, (41)

[0164] Where: Q gs,t and C gs,t They represent the heat output and cold output of the ground source heat pump at time slot t respectively; P gsh,t and P gsc,t They represent the heating power input and cooling power input of the ground source heat pump at time slot t respectively; ωgsh and ω gsc They represent the electricity-heat conversion efficiency and electricity-cold conversion efficiency of the ground source heat pump respectively; Indicates the maximum input power of the ground source heat pump; G gw,t represents the storage capacity of the geothermal well at time slot t, Indicates the maximum storage capacity of the geothermal well; Q w,t Represents the thermal energy that a fuel cell injects into a geothermal well.

[0165] The adsorption chiller operation constraints are:

[0166] C ac,t =ω acc Q ac,t , (42)

[0167] P ac,t =ω ace C ac,t , (43)

[0168] 0≤Q ac,t ≤Q fc,t ,(44)

[0169] C ac,t =ω acc Q ac,t , (45)

[0170] P ac,t =ω ace C ac,t , (46)

[0171] Where: Q ac,t , C ac,t and P ac,t They represent the heat input, cold output and required power of the adsorption refrigerator in time slot t respectively; ω acc and ω ace They represent the heat-to-cold conversion efficiency and the cold-to-electricity conversion efficiency of the adsorption refrigerator respectively.

[0172] The energy balance constraint is:

[0173] E load,t +E hvac,t +(P el,t +P eb,t +P gsh,t +P gsc,t +P ac,t )Δt=(P pv,t +P fc,t )Δt+E loss,t , (47)

[0174] Q load,t +Q hvac,t +Qac,t +q tc,t Δt=Q eb,t +Q gs,t +Q fc,t +Q loss,t +q td,t Δt, (48)

[0175] C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (49)

[0176] C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (50)

[0177] 0≤E loss,t ≤E load,t +E hvac,t ,(51)

[0178] 0≤Q loss,t ≤Q load,t +Q hvac,t ,(52)

[0179] 0≤C loss,t ≤C load,t +C hvac,t ,(53)Where: E load,t , Q load,t and C load,t They represent the electricity, heat and cooling energy requirements caused by the operation of non-building HVAC systems in time slot t; E loss,t , Q loss,t and C loss,t They represent the electrical energy, thermal energy and cold energy losses in time slot t respectively.

[0180] The objective function is:

[0181]

[0182] Among them: F 1,t , F 2,t , F 3,t and F 4,t They represent the photovoltaic reduction cost, hydrogen energy system operation cost, hydrogen purchase cost and energy demand loss in time slot t respectively; L t Indicates the total time slot of operation; represents the expectation operator acting on uncertain system parameters (such as PV output, outdoor temperature, electricity / heating / cooling energy demand) and stochastic control decisions (i.e., v t ,d t ,T t ,w t ,P pv,t ,P el,t ,P fc,t ,Q ac,t ,m t ,P gsh,t ,P gsc,t ,P eb,t ,q tc,t ,q td,t ,c cc,t ,c cd,t ,Q w,t ); α pv Represents the photovoltaic cost reduction coefficient; α h2e and α c2e represents two coefficients; and represents the cost of operating, starting up, and shutting down electrolyzers and fuel cells; and Indicates the flags of operation, startup and shutdown; λ indicates the price of buying hydrogen;

[0183] In the above step 2, the indoor environment dynamics model based on physical embedded neural network is constructed, which specifically includes two modules, namely black box module and physical module. The black box module is composed of a deep neural network, and the input is the outdoor light intensity e t , the output is the indoor temperature disturbance T caused by the light intensity b,t+1 and the indoor humidity ratio disturbance W b,t+1 ; The input of the physical module is the indoor temperature T t in , outdoor temperature T t out , indoor humidity ratio W t in , outdoor humidity ratio W t out , air supply rate v t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the output is the indoor temperature T which can be described by clear physical laws p,t+1 and indoor humidity ratio W p,t+1 Furthermore, the proposed indoor environment dynamics model based on physical embedded neural network is obtained by minimizing the loss function training method, and the designed reward function is as follows:

[0184]

[0185] in: and Respectively and The true value of and Respectively and The differential value with respect to time slot t.

[0186] In the above step 3, the operation optimization method based on physical embedded neural network assisted hierarchical model predictive control is designed as follows:

[0187] (4.1) The proposed hydrogen-containing building energy system operation optimization problem is transformed into an upper-level indoor comfort deviation minimization subproblem and a lower-level operation cost and system reliability joint optimization subproblem;

[0188] Specifically, the remodeled upper floor indoor comfort deviation minimization subproblem is as follows:

[0189]

[0190] st(6)-(12),(62)

[0191] Among them: F 5,t′ represents the indoor comfort deviation penalty in time slot t′; t′ represents the corresponding time slot in the prediction length; h represents the length of the prediction time slot; Represents a physically embedded neural network model of indoor environment dynamics.

[0192] Specifically, the remodeled lower-level operating cost and system reliability joint optimization sub-problem is as follows:

[0193]

[0194] st(13)-(50),(64)

[0195] Where: α loss Represents a coefficient.

[0196] (4.2) If the current operation time slot t is the operation time slot of the building HVAC system, the upper-level model predictive control algorithm assisted by the indoor environmental dynamics model based on the physical embedded neural network is deployed to solve the indoor comfort deviation minimization problem and obtain the optimal upper-level decision for time slot t. in: represents the optimal upper-level decision; and They represent the optimal air supply rate, optimal air valve position, optimal air supply temperature, and optimal air supply humidity ratio. And calculate the various energy requirements of the building HVAC system; conversely, the HVAC system has no energy requirements;

[0197] Specifically, the deployed physical embedded neural network-assisted upper-level model predictive control algorithm uses the gradient descent method to solve the optimal upper-level decision for the current time slot and the next h-1 time slots, namely: The specific steps of using the gradient descent method to solve the optimal upper-level decision are as follows: first, set the gradient descent iteration target Current number of iterations k = 0, maximum number of iterations k max ; If the current time slot t is in the HVAC system operation time slot The upper-level algorithm reads the current state And output the upper-level decision a of the current time slot and the next h-1 time slots upp,t ,a upp,t+1 ,…,a upp,t+h-1 .like And k<k max , then enter the iteration:

[0198] (4.2a) Based on the upper decision a upp,t ,a upp,t+1 ,…,a upp,t+h-1 and The proposed physical embedded neural network predicts upp,t+1 ,…,s upp,t+h ;

[0199] (4.2b) Calculate the indoor comfort deviation between the current time slot and the next h-1 time slots

[0200] (4.2c) Based on with a upp,t ,a upp,t+1 ,…,a upp,t+h-1 , the gradient solver calculates the gradient;

[0201] (4.2d) Based on the obtained gradient, the gradient optimizer optimizes the upper-level decision by minimizing the gradient as the update direction; (4.2e) If the current And k<k max , k=k+1 and jump to 4.2a to iterate again; otherwise, the iteration ends and the optimal upper decision is output

[0202] (4.3) The various energy demands of the building HVAC system obtained by the upper-level model predictive control algorithm are input into the lower-level model predictive control algorithm. The lower-level model predictive control algorithm solves the joint optimization sub-problem of operating cost and system reliability to obtain the optimal decision of the lower level in time slot t in: represents the optimal lower-level decision; and They represent the optimal photovoltaic power generation utilization, the optimal electrolyzer input power, the optimal fuel cell output power, the optimal adsorption refrigerator input heat energy, the optimal hydrogen purchase amount, the optimal ground source heat pump input power, the optimal electric boiler input power, the optimal hot tank heat absorption power, the optimal hot tank heat release power, the optimal cold tank cold absorption power, the optimal cold tank cold release power, and the optimal fuel cell waste heat injection power into the geothermal well. ;

[0203] Specifically, the deployed lower-level model predictive control method mainly solves the optimal lower-level decision for the current time slot and the next X-1 time slots by converting the proposed (P3) into a mixed integer linear programming problem (P4), that is, The mixed integer linear programming problem (P4) is specifically as follows:

[0204]

[0205] st(13)-(50)(66)

[0206]

[0207]

[0208] 0≤Q w,t′ ≤Q fc,t′ K wh,t′ ,(75)

[0209] K el,t′ +K fc,t′ ≤1,(76)

[0210] K gsh,t′ +K gsc,t′ +K wh,t′ ≤1,(77)

[0211] K tc,t′ +K td,t′ ≤1,(78)

[0212] K cc,t′ +K cd,t′ ≤1,(79)

[0213] K el,t′ ,K fc,t′ ,K tc,t′ ,K td,t′ ,K cc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ ∈{0,1}(80)where: K el,t′ ,Kfc,t′ ,K tc,t′ ,K td,t′ ,K cc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ is an auxiliary variable.

[0214] (4.4) Execute the optimal decision of the upper layer Optimal decision making In the operating environment of the hydrogen-containing building energy system, the environment feedbacks the next time slot state s upp,t+1 ,s low,t+1 . Let t = t + 1, if t < L t , then jump to 4.1 to continue running; otherwise, the running ends;

[0215] In the above step 4, the actual operation system for deploying the proposed operation optimization method includes an operation environment module of a hydrogen-containing building energy system and a real-time decision-making module based on physical embedded neural network-assisted hierarchical model predictive control.

[0216] The hydrogen-containing building energy system operating environment module is responsible for receiving the optimal decision output by the real-time decision module and executing it on the hydrogen-containing building energy system.

[0217] The real-time decision-making module based on physical embedded neural network assisted hierarchical model predictive control 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 and an optimal control strategy output submodule.

[0218] The information collection submodule is responsible for collecting the dynamic changes of each component after the hydrogen-containing building energy system executes the decision, such as the indoor temperature of the building, indoor relative humidity, indoor humidity ratio, real-time storage capacity of hydrogen tanks / hot water tanks / cold water tanks / geothermal wells, etc. Then, the latest status of each component is sent to the upper and lower model predictive control real-time decision submodules respectively.

[0219] The upper-level model predictive control real-time decision submodule reads the indoor environmental status of the building, i.e., the indoor temperature, indoor relative humidity, and indoor humidity ratio, issued by the information acquisition submodule. Then, the optimal upper-level decision is solved using the operation optimization method, and the multiple energy requirements of the building HVAC system are calculated. Finally, the optimal upper-level decision and the multiple energy requirements of the building HVAC system are sent to the optimal control strategy output submodule and the lower-level model predictive control real-time decision submodule, respectively.

[0220] The lower-level model predictive control real-time decision submodule reads the energy storage system status (i.e., the real-time storage volume of hydrogen tanks / hot water tanks / cold water tanks / geothermal wells) issued by the information acquisition submodule and the various energy requirements of the HVAC system issued by the upper-level model predictive control real-time decision submodule. Then, the optimal lower-level decision is solved using the operation optimization method. Finally, the optimal lower-level decision is transmitted to the optimal control strategy output submodule.

[0221] The optimal strategy output module reads the optimal decisions issued by the upper and lower model predictive control real-time decision submodules, respectively, and then sends the optimal upper and lower decisions to the hydrogen-containing building energy system operating environment module.

[0222] In order to demonstrate the effectiveness of the method of the present invention, four comparative schemes were introduced.

[0223] Scheme 1 adopts a rule-based commercial building HVAC system operation strategy and a rule-based hydrogen-containing energy system operation strategy. In the rule-based commercial building HVAC system operation strategy, the comfort control of indoor users is achieved by adjusting the air supply rate, and the air valve position, air supply temperature, and air supply humidity control are all set to fixed values. Specifically, in cooling mode, when the indoor temperature is higher than the upper limit of the comfortable temperature range, the air supply rate 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 air supply rate of the HVAC system is set to the minimum; when the indoor temperature is within the comfortable temperature range, the air supply rate of the HVAC system remains unchanged. In the rule-based hydrogen-containing energy system operation strategy, other control components (such as electrolyzers, fuel cells, adsorption chillers, and ground source heat pumps, etc.) except the air handling units inside the building are operated according to the energy supply and demand relationship at the current moment.

[0224] Solution 2 uses a deep reinforcement learning algorithm based on proximal policy optimization to optimize the operation of the building HVAC control system and uses the rule-based hydrogen energy system operation strategy in Solution 1 to control other components. Specifically, two agents trained based on proximal policy optimization are used to optimize the control of air supply rate, air valve position, air supply temperature and air supply humidity in cooling mode and heating mode respectively.

[0225] Solution 3 uses a deep reinforcement learning algorithm based on proximal strategy optimization to optimize the operation of the building HVAC control system and a model predictive control algorithm to optimize the operation of other control components except the air handling unit inside the building. Specifically, the model predictive control method used in this solution is the same as the lower-level model predictive control method in the method of the present invention.

[0226] Scheme 4 uses a greedy algorithm to optimize the operation of the hydrogen-containing building energy system, which is equivalent to the method of the present invention when the prediction interval is 1, that is, this scheme only makes decisions based on the current time slot status.

[0227] Figure 2-Figure 4 is a performance comparison diagram of the method of the present invention and the comparative scheme, Figure 5 is a summary of the performance indicators of all solutions. Figure 2 The figure shows a comparison of the operating costs of the method of the present invention and other methods. Specifically, compared with other comparative schemes, the method of the present invention can reduce the operating costs by 11.19%-62.10%, and has better economic benefits.

[0228] like Figure 3 The figure shows a comparison of the average system energy demand loss of the method of the present invention and other methods. Specifically, compared with other solutions, the method of the present invention can reduce the average energy demand loss by 13.06%-82.65%, and has stronger system reliability.

[0229] like Figure 4 The figure shows the average indoor environment comfort deviation comparison diagram of the method of the present invention and other methods. Specifically, compared with other solutions, the method of the present invention can reduce the average indoor comfort deviation by 82.83%-99.39%, and can provide a more comfortable indoor environment.

Claims

1. A reliable and economical operation optimization method for a hydrogen-containing building energy system in an off-grid mode, characterized in that: The operation optimization method comprises the following steps: Step 1: Model the operation optimization problem of hydrogen-containing building energy system in off-grid mode while maintaining indoor comfortable environment; Step 2: Modeling the indoor environment dynamics model based on physical embedded neural network; Step 3: Propose an operation optimization method based on physically embedded neural network assisted hierarchical model predictive control to solve the above modeling problem; Step 4: Deploy the proposed hydrogen-containing building energy system operation optimization method in the actual operation system.

2. The method for optimizing the reliable and economic operation of a hydrogen-containing building energy system in an off-grid mode according to claim 1, characterized in that: The operation optimization problem of the hydrogen-containing building energy system in the off-grid mode modeled in step 1 includes decision variables, constraints and objective functions; The decision variable a t Including the building HVAC system air supply rate v in the time slot t t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the actual photovoltaic output power P pv,t , electrolyzer input power P el,t , fuel cell output power P fc,t , adsorption refrigerator input heat energy Q ac,t , the excess heat energy of the fuel cell Q w,t , ground source heat pump heating / cooling input power P gsh,t / P gsc,t , electric boiler input power P eb,t , hot water tank input / output thermal power q tc,t / q td,t , cold water tank input / output cooling power in c cc,t / c cd,t And the amount of hydrogen purchased m t ; The constraints include: building HVAC system operation constraints, photovoltaic operation constraints, hydrogen energy storage system operation constraints, hot water tank operation constraints, cold water tank operation constraints, electric boiler operation constraints, ground source heat pump operation constraints, adsorption chiller operation constraints and energy balance constraints; The operating constraints of the building HVAC system are: T min ≤T t in ≤T max ,(3) Where: T t in , W t in as well as They represent the indoor temperature, indoor humidity ratio and indoor relative humidity at time slot t respectively; T t out and W t out They represent the outdoor temperature and outdoor humidity ratio at time slot t respectively; e t represents the light intensity at time slot t; ξ t and ψ t represent the indoor thermal disturbance and humidity disturbance related to light, users and lighting respectively; and Respectively represent the dynamic model of indoor temperature and indoor humidity; T min and T max Respectively represent the lower and upper limits of indoor comfort temperature; and Respectively represent the lower and upper limits of indoor comfortable relative humidity; B hh,t and B hl,t They represent the indoor high temperature and high humidity discomfort boundary and the high temperature and low humidity discomfort boundary respectively; v min and v max Respectively represent the minimum and maximum value of air supply speed adjustment; d min and d max Respectively represent the minimum and maximum values ​​of the air valve position adjustment; T ca,t and T ha,t represent the cooling temperature and heating temperature in time slot t, T ma,t represents the mixed air temperature at time slot t; W ca,t and W ha,t They represent the dehumidification humidity ratio and humidification humidity ratio at time slot t, respectively, and W ma,t represents the mixed air humidity ratio at time slot t, Indicates the maximum humidification humidity ratio; η fan Indicates the air supply fan coefficient; η wv is the latent heat of evaporation of water, η hum Indicates the efficiency of the humidification coil; η pa , η cc and COP cc They represent the specific heat capacity of air, cooling coil efficiency, and cooling coil performance coefficient respectively; Indicates the building HVAC operation time slot, and represent summer and winter respectively; Δt represents the time slot length; The photovoltaic operation constraints are: in: represents the maximum photovoltaic power generation in time slot t, and h pv They represent the photovoltaic power generation efficiency and illumination area in time slot t respectively; P pv,t Indicates the actual utilization of photovoltaic power in time slot t; The operating constraints of the hydrogen energy storage system are: 0≤H t ≤H max , (16) P el,t ·P fc,t =0, (17) 0≤m t ≤m max , (20) Q fc,t =ω e2h ω hr P fc,t Δt, (21) Where: H t represents the storage capacity of the hydrogen tank at time slot t; ω el and ω fc They represent the electrolyzer electricity-to-hydrogen conversion efficiency and the fuel cell hydrogen-to-electricity conversion efficiency respectively; P el,t and P fc,t Respectively represent the electrolyzer input power and fuel cell output power at time slot t; m t represents the amount of hydrogen purchased in time slot t; H max Indicates the maximum storage capacity of the hydrogen tank; and Respectively represent the maximum input power of the electrolyzer and the maximum output power of the fuel cell; m max Indicates the maximum amount of hydrogen purchased; Q fc,t represents the heat generated by the fuel cell at time slot t, ω e2h and ω hr represent the fuel cell electrical-to-thermal conversion efficiency and heat recovery efficiency respectively; The hot water tank operation constraints are: q tc,t ·q td,t =0, (26) Where: M hwt,t represents the storage capacity of the hot water tank at time slot t; ω tc and ω td They represent the thermal power input efficiency and output efficiency of the hot water tank respectively; q tc,t and q td,t denote the heat input power and heat output power in time slot t respectively; Indicates the maximum storage capacity of the hot water tank; and They represent the maximum input thermal power and the maximum output thermal power of the hot water tank respectively; The cold water tank operation constraints are: c cc,t ·c cd,t =0, (31) Among them: U cwt,t represents the storage capacity of the cold water tank at time slot t; ω cc and ω cd They represent the cold water tank cooling power input efficiency and output efficiency respectively; c cc,t and c cd,t They represent the cold input power and cold output power in time slot t respectively; Indicates the maximum storage capacity of the cold water tank; and They represent the maximum input cooling power and the maximum output cooling power of the cold water tank respectively; The electric boiler operation constraints are: Q eb,t =ω eb P eb,t Δt, (32) Where: Q eb,t and P eb,t They represent the output heat energy and input power of the electric boiler in time slot t respectively; ω eb Indicates the electricity-heat conversion efficiency of electric boilers; Indicates the maximum input power of the electric boiler; The ground source heat pump furnace operation constraints are: Q gs,t =ω gsh P gsh,t Δt, (34) C gs,t =ω gsc P gsc,t Δt, (35) 0≤Q w,t ≤Q fc,t , (40) Q w,t ·C gs,t ·Q gs,t =0, (41) Where: Q gs,t and C gs,t They represent the heat output and cold output of the ground source heat pump at time slot t respectively; P gsh,t and P gsc,t They represent the heating power input and cooling power input of the ground source heat pump at time slot t respectively; ω gsh and ω gsc They represent the electricity-heat conversion efficiency and electricity-cold conversion efficiency of the ground source heat pump respectively; Indicates the maximum input power of the ground source heat pump; G gw,t represents the storage capacity of the geothermal well at time slot t, Indicates the maximum storage capacity of the geothermal well; Q w,t represents the heat energy injected into the geothermal well by the fuel cell; The adsorption chiller operation constraints are: C ac,t =ω acc Q ac,t , (42) P ac,t =ω ace C ac,t , (43) 0≤Q ac,t ≤Q fc,t , (44) C ac,t =ω acc Q ac,t , (45) P ac,t =ω ace C ac,t , (46) Where: Q ac,t , C ac,t and P ac,t They represent the heat input, cold output and required power of the adsorption refrigerator in time slot t respectively; ω acc and ω ace They represent the heat-to-cold conversion efficiency and the cold-to-electricity conversion efficiency of the adsorption chiller respectively; The energy balance constraint is: E load,t +E hvac,t +(P el,t +P eb,t +P gsh,t +P gsc,t +P ac,t )Δt=(P pv,t +P fc,t )Δt+E loss,t , (47) Q load,t +Q hvac,t +Q ac,t +q tc,t Δt=Q eb,t +Q gs,t +Q fc,t +Q loss,t +q td,t Δt, (48) C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (49) C load,t +C hvac,t +c cc,t Δt=C ac,t +C gs,t +C loss,t +c cd,t Δt, (50) 0≤E loss,t ≤E load,t +E hvac,t , (51) 0≤Q loss,t ≤Q load,t +Q hvac,t , (52) 0≤C loss,t ≤C load,t +C hvac,t , (53) Where: E load,t , Q load,t and C load,t They represent the electricity, heat and cooling energy requirements caused by the operation of non-building HVAC systems in time slot t; E loss,t , Q loss,t and C loss,t They represent the electric energy, heat energy and cold energy loss in time slot t respectively; The objective function is: Among them: F 1,t , F 2,t , F 3,t and F 4,t They represent the photovoltaic reduction cost, hydrogen energy system operation cost, hydrogen purchase cost and energy demand loss in time slot t respectively; L t Indicates the total time slot of operation; represents the expectation operator, which acts on uncertain system parameters and random control decisions. The uncertain system parameters include photovoltaic output, outdoor temperature, and electricity / heating / cooling energy demand. The random control decisions include: v t ,d t ,T t ,w t ,P pv,t ,P el,t ,P fc,t ,Q ac,t ,m t ,P gsh,t ,P gsc,t ,P eb,t ,q tc,t ,q td,t ,c cc,t ,c cd,t ,Q w,t 。 3. The reliable and economical operation optimization method of a hydrogen-containing building energy system in an off-grid mode according to claim 1 is characterized in that: The indoor environment dynamics model based on physical embedded neural network built in step 2 includes two modules, namely black box module and physical module; the black box module is composed of a deep neural network, and the input is the outdoor light intensity e t , the output is the indoor temperature disturbance T caused by the light intensity b,t+1 and the indoor humidity ratio disturbance W b,t+1 ; The input of the physical module is the indoor temperature T t in , outdoor temperature T t out , indoor humidity ratio W t in , outdoor humidity ratio W t out , air supply rate v t , air valve position d t , air supply temperature T t And the air supply humidity ratio w t , the output is the room temperature T described by clear physical laws p,t+1 and indoor humidity ratio W p,t+1 ; The proposed indoor environment dynamics model based on physical embedded neural network is obtained by minimizing the loss function training method. The designed reward function is as follows: in: and Respectively and The true value of and Respectively and The differential value with respect to time slot t.

4. The method for optimizing the reliable and economic operation of a hydrogen-containing building energy system in an off-grid mode according to claim 1, characterized in that: The operation optimization method based on physical embedded neural network assisted hierarchical model predictive control proposed in step 3 is designed as follows: (4.1) The proposed hydrogen-containing building energy system operation optimization problem is transformed into an upper-level indoor comfort deviation minimization subproblem and a lower-level operation cost and system reliability joint optimization subproblem; (4.2) If the current operation time slot t is the operation time slot of the building HVAC system, the upper-level model predictive control algorithm assisted by the indoor environmental dynamics model based on the physical embedded neural network is deployed to solve the indoor comfort deviation minimization problem and obtain the optimal upper-level decision for time slot t. in: represents the optimal upper-level decision; and They represent the optimal air supply rate, optimal air valve position, optimal air supply temperature and optimal air supply humidity ratio respectively, and calculate the various energy requirements of the building HVAC system; otherwise, the HVAC system has no energy requirements; (4.3) The various energy demands of the building HVAC system obtained by the upper-level model predictive control algorithm are input into the lower-level model predictive control algorithm. The lower-level model predictive control algorithm solves the joint optimization sub-problem of operating cost and system reliability to obtain the optimal decision of the lower level in time slot t in: represents the optimal lower-level decision; and They represent respectively the optimal photovoltaic power generation utilization, the optimal electrolyzer input power, the optimal fuel cell output power, the optimal adsorption refrigerator input heat energy, the optimal hydrogen purchase amount, the optimal ground source heat pump input power, the optimal electric boiler input power, the optimal hot tank heat absorption power, the optimal hot tank heat release power, the optimal cold tank cold absorption power, the optimal cold tank cold release power and the optimal fuel cell waste heat injection power into the geothermal well; (4.4) Execute the optimal decision of the upper layer Optimal decision making In the operating environment of the hydrogen-containing building energy system, the environment feedbacks the next time slot state s upp,t+1 ,s low,t+1 ; Let t = t + 1, if t < L t , it will jump to 4.1 to continue running; otherwise, it will end the running.

5. The method for optimizing the reliable and economic operation of a hydrogen-containing building energy system in an off-grid mode according to claim 4, characterized in that: The upper indoor comfort deviation minimization subproblem remodeled in step 4.1 is as follows: st(6)-(12),(57) Among them: F 5,t′ represents the indoor comfort deviation penalty in time slot t′; t′ represents the corresponding time slot in the prediction length; h represents the length of the prediction time slot; Represents a physically embedded neural network model of indoor environment dynamics.

6. The method for optimizing the reliable and economic operation of a hydrogen-containing building energy system in an off-grid mode according to claim 4, characterized in that: The lower-level operating cost and system reliability joint optimization subproblem remodeled in step 4.1 is as follows: st(13)-(50),(59) Where: α loss Represents a coefficient.

7. The operation optimization method based on physical embedded neural network assisted hierarchical model predictive control according to claim 4 is characterized in that: The upper-level model predictive control algorithm assisted by the physical embedded neural network deployed in step 4.2 uses the gradient descent method to solve the optimal upper-level decision for the current time slot and the next h-1 time slots, that is, The specific steps of using the gradient descent method to solve the optimal upper-level decision are as follows: first, set the gradient descent iteration target Current number of iterations k = 0, maximum number of iterations k max ; If the current time slot t is in the HVAC system operation time slot The upper-level algorithm reads the current state And output the upper-level decision a of the current time slot and the next h-1 time slots upp,t ,a upp,t+1 ,…,a upp,t+h-1 ;like And k<k max , then enter the iteration: (7.1) Based on the upper decision a upp,t ,a upp,t+1 ,…,a upp,t+h-1 and The proposed physical embedded neural network predicts upp,t+1 ,…,s upp,t+h ; (7.2) Calculate the indoor comfort deviation between the current time slot and the next h-1 time slots (7.3) Based on with a upp,t ,a upp,t+1 ,…,a upp,t+h-1 , the gradient solver calculates the gradient; (7.4) Based on the obtained gradient, the gradient optimizer optimizes the upper-level decision by minimizing the gradient as the update direction; (7.5) If the current And k<k max , k=k+1 and jump to 7.1 to iterate again; otherwise, the iteration ends and the optimal upper-level decision is output 8. The operation optimization method based on physical embedded neural network assisted hierarchical model predictive control according to claim 4 is characterized in that: The lower-level model predictive control method deployed in step 4.3 mainly solves the optimal lower-level decision for the current time slot and the next X-1 time slots by converting the proposed (P3) into a mixed integer linear programming problem (P4), that is, The mixed integer linear programming problem (P4) is specifically as follows: st(13)-(50)(61) 0≤Q w,t′ ≤Q fc,t′ K wh,t′ , (70) K el,t′ +K fc,t′ ≤1, (71) K gsh,t′ +K gsc,t′ +K wh,t′ ≤1, (72) K tc,t′ +K td,t′ ≤1, (73) K cc,t′ +K cd,t′ ≤1, (74) K el,t′ ,K fc,t′ ,K tc,t′ ,K td,t′ ,K cc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ ∈{0,1} (75) Where: K el,t′ ,K fc,t′ ,K tc,t′ ,K td,t′ ,K cc,t′ ,K cd,t′ ,K gsh,t′ ,K gsc,t′ ,K wh,t′ is an auxiliary variable.

9. The method for optimizing the reliable and economic operation of a hydrogen-containing building energy system in an off-grid mode according to claim 1, characterized in that: The actual operation system in which the proposed operation optimization method is deployed in step 4 includes an operation environment module of a hydrogen-containing building energy system and a real-time decision-making module based on physical embedded neural network-assisted hierarchical model predictive control; The hydrogen-containing building energy system operating environment module is responsible for receiving the optimal decision output by the real-time decision module and executing it on the hydrogen-containing building energy system; The real-time decision-making module based on physical embedded neural network assisted hierarchical model predictive control 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, and an optimal control strategy output submodule; The information collection submodule is responsible for collecting the dynamic changes of each component after the hydrogen-containing building energy system executes the decision, including the indoor temperature, indoor relative humidity, indoor humidity ratio, and real-time storage capacity of hydrogen tanks / hot water tanks / cold water tanks / geothermal wells; then, the latest status of each component is sent to the upper and lower model predictive control real-time decision submodules respectively; The upper-level model predictive control real-time decision-making submodule reads the indoor environmental status of the building, i.e., indoor temperature, indoor relative humidity, and indoor humidity ratio, sent by the information acquisition submodule; then, the optimal upper-level decision is solved using the operation optimization method, and the multiple energy requirements of the building HVAC system are calculated; finally, the optimal upper-level decision and the multiple energy requirements of the building HVAC system are sent to the optimal control strategy output submodule and the lower-level model predictive control real-time decision-making submodule respectively; The lower-level model predictive control real-time decision submodule reads the energy storage system status issued by the information acquisition submodule and the various energy requirements of the HVAC system issued by the upper-level model predictive control real-time decision submodule, where the energy storage system status includes the real-time storage capacity of the hydrogen tank / hot water tank / cold water tank / geothermal well; then, the optimal lower-level decision is solved using the operation optimization method; finally, the optimal lower-level decision is transmitted to the optimal control strategy output submodule; The optimal strategy output module reads the optimal decisions issued by the upper and lower model predictive control real-time decision submodules respectively; then, the optimal upper decision and the optimal lower decision are sent to the hydrogen-containing building energy system operating environment module.

Images