Real-time optimization control method for central air-conditioning system based on proxy model

Abstract

The invention relates to a real-time optimization control method for a central air-conditioning system based on an agent model. The machine learning algorithm is adopted to construct the agent model used for predicting the optimal control strategy result, the real-time operation condition is input into the agent model in the actual application process, the optimal control strategy result under the operation condition is predicted, and compared with a traditional control method based on model optimization, the method has the advantages that the operation efficiency is improved, and the cost is reduced. According to the method, the calculation complexity is reduced, the calculation efficiency of the optimal control strategy is remarkably improved, and the real-time optimization control requirement of the central air conditioning system is met. The mathematical optimization model of the central air-conditioning system is established based on the equipment composition and the topological structure of the target central air-conditioning system as well as the optimization target and the constraint condition during model training, so that a large number of optimal control strategies corresponding to the operation condition sample are generated through the randomly generated operation condition; and forming a training data set for agent model training, so that the trained agent model can accurately predict an optimal control strategy.

Description

Technical Field

[0001] The present invention relates to a real-time optimization control method for a central air-conditioning system based on an agent model, which is applicable to the field of operation control of an integrated energy system. Background Art

[0002] The central air-conditioning system is the most important energy-consuming system in the operation stage of public buildings, and its energy consumption accounts for more than 50% of the total energy consumption in the operation stage of buildings. Research on the optimization control method of the central air-conditioning system is of great significance for meeting the cooling load requirements of building users and reducing the operating energy consumption of the system. It is a powerful tool for achieving low-carbon energy conservation in public buildings.

[0003] Central air-conditioning systems have the characteristics of high nonlinearity, complex topological structure, and strong thermal coupling. Currently, this field generally adopts the method of solving the system optimization model to calculate the optimal control strategy of the system.

[0004] For example, the Chinese invention patent with patent publication number CN115577828A, a group control method for air-conditioning and refrigeration station systems based on data-driven modeling and optimization, uses a differential evolution algorithm to solve the air-conditioning and refrigeration station model to achieve the optimal cooling capacity of the chiller, and builds an optimization database for operation and maintenance personnel to query; the Chinese invention patent with patent publication number CN115829093A, a data-driven optimization method for setting the chilled water outlet temperature, uses a fish school algorithm to solve the indoor temperature and electric energy prediction model, and achieves the optimal chilled water outlet temperature of the chiller.

[0005] However, although the above methods have high optimization accuracy, their computational complexity and optimization time cost are usually high, and in practical applications, they are often difficult to meet the real-time optimization control requirements of the system. How to reduce the computational complexity and improve the optimization efficiency while ensuring the optimization accuracy of the control strategy to achieve efficient real-time control of the central air-conditioning system needs further research. Summary of the invention

[0006] The technical problem to be solved by the present invention is: in view of the above-mentioned problems, a real-time optimization control method for a central air-conditioning system based on an agent model is provided.

[0007] The technical solution adopted by the present invention is: a real-time optimization control method for a central air-conditioning system based on an agent model, comprising: Obtain the real-time operating conditions of the target central air-conditioning system, input the operating conditions into the trained proxy model, and predict the optimal control strategy results under the operating conditions; The training of the proxy model includes: Determine an operating condition set of a target central air-conditioning system and a set of control parameters to be optimized for each device in the system; the operating condition set includes a numerical range of multiple types of operating condition parameters, and the set of control parameters to be optimized includes a numerical range of multiple types of device operating parameters for each device in the system; Based on the equipment composition and topological structure of the target central air-conditioning system, as well as the optimization objectives and constraints, a mathematical optimization model of the central air-conditioning system is established; the optimization objectives include minimizing the total operating energy consumption of the system, and the constraints include energy conservation constraints; Based on the numerical range of each parameter in the target central air-conditioning system operating condition set, a plurality of operating condition samples are randomly generated, and the mathematical optimization model under each operating condition sample is solved to obtain the optimal control strategy corresponding to the parameters in the control parameter set to be optimized under each operating condition sample, and a training data set is constructed based on the operating condition samples and the corresponding optimal control strategy; Based on the training data set, a proxy model is trained, which is built using a machine learning algorithm.

[0008] The operating condition parameter categories in the operating condition set include terminal cooling load, outdoor temperature and outdoor humidity; the parameter categories in the control parameter set to be optimized include set value parameters of each device of the target central air-conditioning system.

[0009] The set value parameters of each device of the target central air-conditioning system include the chilled water outlet temperature of the chiller, the operating frequency of the chilled water pump, the operating frequency of the cooling water pump, and the operating frequency of the cooling tower fan.

[0010] The constraints include: The energy conservation constraint between the cooling water side and the chilled water side, that is, the total heat rejection of the cooling tower should be equal to the sum of the cooling load and the operating power of each chiller; The energy conservation constraint between cooling capacity and cooling load, that is, the cooling load should be equal to the sum of the cooling capacity of each chiller; Energy conservation constraints for the chilled water loop; Energy conservation constraints for the cooling water loop.

[0011] The optimization algorithms for solving the mathematical optimization model include genetic algorithm, particle swarm algorithm and grey wolf algorithm and their variants.

[0012] The machine learning algorithms used to build the proxy model include extreme gradient boosting, recursive neural network, artificial neural network, and support vector machine regression.

[0013] A storage medium stores a computer program that can be executed by a processor, and when the computer program is executed, the steps of the real-time optimization control method of the central air-conditioning system are implemented.

[0014] A central air conditioning system real-time optimization control device comprises a memory and a processor. The memory stores a computer program executable by the processor. When the computer program is executed, the steps of the central air conditioning system real-time optimization control method are implemented.

[0015] The beneficial effects of the present invention are as follows: the present invention adopts a machine learning algorithm to construct an agent model for predicting the results of the optimal control strategy. In actual application, the real-time operating conditions are input into the agent model to predict the results of the optimal control strategy under the operating conditions. Compared with the traditional control method based on model optimization, the present invention reduces the computational complexity and significantly improves the computational efficiency of the optimal control strategy, thus meeting the real-time optimization control needs of the central air-conditioning system.

[0016] During model training, the present invention establishes a mathematical optimization model of the central air-conditioning system based on the equipment composition and topological structure of the target central air-conditioning system, as well as the optimization objectives and constraints, thereby generating a large number of optimal control strategies corresponding to the operating conditions through randomly generated operating conditions, forming a training data set for agent model training, overcoming the technical problem that the target central air-conditioning system has a small amount of data and cannot be effectively trained, so that the trained agent model can accurately predict the optimal control strategy.

[0017] The model training in the present invention can be performed on a dedicated model training device, and only the trained proxy model needs to be deployed on the specific controller. Compared with the traditional method of deploying a mathematical optimization model in the controller, the computing power requirements for the controller are low, the controller overhead is small, and the computing efficiency is extremely high, which can significantly improve the real-time control performance of the system.

[0018] The present invention selects terminal cooling load, outdoor temperature and outdoor humidity as inputs for model prediction, wherein the terminal cooling load mainly affects the operation of the chilled water side, and has an important influence on the outlet water temperature set point of the chiller and the frequency of the chilled water pump; the outdoor temperature and outdoor humidity mainly affect the operation of the cooling water side, and have an important influence on the heat removal performance of the cooling tower, and further affect the frequency of the cooling water pump. Therefore, selecting the terminal cooling load, outdoor temperature and outdoor humidity as inputs helps the proxy model to accurately predict the optimal control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of the structure of the target central air-conditioning system in an embodiment of the present invention.

[0020] Figure 2 4 is a flow chart of training the proxy model in an embodiment of the present invention.

[0021] Figure 3 The optimization accuracy comparison of the optimal control strategy under different methods in the embodiments of the present invention is shown in FIG.

[0022] Figure 4The figure is a comparison of the calculation speed of the optimal control strategy under different methods in the embodiments of the present invention. DETAILED DESCRIPTION

[0023] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0024] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0025] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0026] Embodiment 1: Figure 1 : is a schematic diagram of the structure of the target central air-conditioning system in this embodiment, which includes three chillers, three chilled water pumps, three cooling water pumps and three cooling towers. On the chilled water side, the low-temperature chilled water generated by the chiller is transmitted to the terminal for cooling, and the chilled water return water is returned to the chiller through the chilled water pump for circulation; on the cooling water side, the high-temperature cooling water return water is transmitted to the cooling tower through the cooling water pump, and returns to the chiller after being cooled by the cooling tower.

[0027] This embodiment is a real-time optimization control method for a central air conditioning system based on an agent model, which specifically includes the following steps: A. Obtain the real-time operating conditions of the target central air-conditioning system, including the terminal cooling load , Outdoor temperature , Outdoor humidity Specific values ​​of parameters such as; B. Input the operating conditions obtained in step A into the trained proxy model, and use the proxy model to predict the optimal control strategy results under the operating conditions.

[0028] The optimal control strategy in this case includes the chilled water outlet temperature of each chiller (where subscripts 1-3 correspond to the three chillers in this example), the operating frequency of each chilled water pump , the operating frequency of each cooling water pump , fan operating frequency of each cooling tower Specific values ​​of other parameters.

[0029] In this embodiment, the proxy model is used to predict the optimal control strategy based on the operating conditions, such as Figure 2 As shown, the training of the proxy model includes: S100: Determine a set of operating conditions of a target central air-conditioning system and a set of control parameters to be optimized for each device in the system.

[0030] In this embodiment, the operating condition set includes the terminal cooling load , Outdoor temperature and outdoor humidity The numerical range of other operating parameters.

[0031] Operation condition collection As shown below: (1)

[0032] In this example, the set of control parameters to be optimized includes the chilled water outlet temperature of each chiller Operating frequency of each chilled water pump , the operating frequency of each cooling water pump , fan operating frequency of each cooling tower The numerical range of equipment operating parameters.

[0033] The set of control parameters to be optimized As shown below: (2)

[0034] S200: Based on the equipment composition and topological structure of the target central air-conditioning system, as well as the optimization objectives and constraints, a mathematical optimization model of the central air-conditioning system is established.

[0035] In this embodiment, the optimization objective function is to minimize the total operating energy consumption of the system; the constraints include the energy conservation constraints between the cooling water side and the chilled water side, the energy conservation constraints between the cooling capacity and the cooling load, the energy conservation constraints of the chilled water loop and the energy conservation constraints of the cooling water loop, and the numerical range constraints of the operating parameters of each device in the system.

[0036] The mathematical optimization model of the target central air-conditioning system in this example includes: (3) (4) (5) (6) (7) in, , , and They represent the number of chillers, chilled water pumps, cooling water pumps and cooling towers, and are all 3 in this embodiment. , , and Respectively represent The operating energy consumption of each chiller, chilled water pump, cooling water pump and cooling tower, is the terminal cooling load, For the The heat rejection of a cooling tower, For the The cooling capacity of each chiller, is the specific heat capacity of water at constant pressure, and Respectively represent The chilled water and cooling water flow of each loop, and Respectively represent The supply and return temperatures of chilled water in each chiller. and Respectively represent The outlet water temperature and inlet water temperature of the cooling tower.

[0037] In this optimization model, formula (3) represents the optimization objective function, i.e., minimizing the total operating energy consumption of the system; formula (4) represents the energy conservation constraint between the cooling water side and the chilled water side, i.e., the total heat rejection of the cooling tower should be equal to the sum of the cooling load and the operating power of each chiller; formula (5) represents the energy conservation constraint between the cooling capacity and the cooling load, i.e., the cooling load should be equal to the sum of the cooling capacity of each chiller; formula (6) represents the energy conservation constraint of the chilled water loop; formula (7) represents the energy conservation constraint of the cooling water loop.

[0038] In this embodiment, the energy consumption model of the chiller is as follows: (8) in, is the coefficient of performance of the chiller, is the rated cooling capacity, is the rated performance coefficient. The energy consumption model of the chilled water pump and the cooling water pump is shown in the following formula: (9) in, is the coefficient of performance of the pump, is the pump efficiency, is the rated frequency of the water pump. The energy consumption model of the cooling tower is shown as follows: (10) in, is the cooling tower performance coefficient, is the rated power of the cooling tower.

[0039] S300. Randomly generate multiple operating condition samples based on the numerical range of each parameter in the target central air-conditioning system operating condition set, and solve the mathematical optimization model under each operating condition sample to obtain the optimal control strategy corresponding to the control parameter set to be optimized under each operating condition sample, and construct a training data set based on the operating condition samples and the corresponding optimal control strategy.

[0040] S310, terminal cooling load in this embodiment The range is [500, 4000] kW, outdoor temperature The range is [15, 40] ℃, and the outdoor humidity The range is [30, 90] %, and random sampling is performed from the above operating condition range to generate 8000 sets of operating condition samples. , as shown below: (11) in, Indicates operating condition samples; For the The terminal cooling load in the operating condition samples; For the The outdoor temperature in the operating condition samples; For the The outdoor humidity in the operating condition samples.

[0041] S320, for each operating condition sample , substitute it into the mathematical optimization model and solve it to obtain the optimal control strategy under this condition , as shown below: (12) in, Indicates The optimal control strategy corresponding to the operating condition samples; The superscript i in the above formula indicates the A sample of operating conditions.

[0042] In this embodiment, the optimization algorithms for solving the mathematical optimization model of the target central air-conditioning system include, but are not limited to, genetic algorithms, particle swarm algorithms, grey wolf algorithms and other heuristic algorithms and their variants.

[0043] In some specific embodiments, a particle swarm algorithm is used to solve the mathematical optimization model, and the specific steps are as follows: S321. Set the particle swarm size , inertia weight , acceleration factor , , maximum number of iterations and iteration counter ; S322, the terminal cooling load , Outdoor temperature and outdoor humidity Bring in the optimization model and complete model initialization; S323, randomly initialize the position vector of each particle in the search space and the velocity vector , as shown below: (13) (14) In this embodiment, The search space is [6, 10]℃, , and The search space is [20, 50] Hz; S324, each particle is substituted into the optimization model according to the control parameters represented by its position variable to calculate the corresponding fitness value. The calculation formula is as follows: (15) in, To optimize the objective function value of the model under this set of control parameters, In order to optimize the violation degree of each constraint of the model under this set of control parameters, is the penalty coefficient. By adding the penalty term of constraint violation on the basis of the original objective function value, it is ensured that the particles can meet the model constraints when searching for the optimal solution; S325. Update the velocity vector of each particle , as shown below: (16)

[0044] in, and is a random number in the interval [0, 1], For particles The position vector corresponding to the minimum value of the objective function during the iteration process is is the position vector corresponding to the minimum objective function value of all particles in the iteration process; S326. Update the position vector of each particle , as shown below: (17) S327, if the number of iterations reaches Or within a certain number of consecutive iterations If both remain unchanged, the algorithm stops and The control parameters represented are output to obtain the optimal . Otherwise, return to step S323.

[0045] S330, with samples of various operating conditions is the input, and the corresponding optimal control strategy As output, a training dataset for the proxy model is constructed.

[0046] S400, training an agent model based on a training data set, where the agent model is constructed using a machine learning algorithm.

[0047] The machine learning algorithms used to build the proxy model in this example include, but are not limited to, extreme gradient boosting, recursive neural networks, artificial neural networks, support vector machine regression, etc.

[0048] In some specific embodiments, an artificial neural network algorithm is used to construct an agent model for optimal control strategy prediction. The specific steps include: S410, setting network structure parameters, as shown in the following table. Randomly initialize the network weights and bias parameters according to the structure parameters; Table 1 Artificial neural network structure parameters

[0049] S420, dividing the data set into a training set and a validation set in a ratio of 7:3, which are used for training the proxy model and verifying the model accuracy respectively; S430, using minimum and maximum standardization to uniformly convert the data into the interval [0, 1], as shown in the following formula: (18) in, is the original sample data, including input data and output data, and are the maximum and minimum values ​​of this type of data, respectively. The data after standardization operation.

[0050] S440. Weights and biases for neural networks Perform batch updates as shown below: (19) in, is the learning rate, which is 0.01 in this embodiment. is the batch size, which is 256 in this embodiment. for Relative to the gradient function of the loss function, the loss function in this embodiment is the root mean square error , as shown below: (20) in, is the sample size, and The samples are No. The actual and predicted values ​​of the output parameters are repeated until the algorithm converges.

[0051] In this embodiment, the real-time operating conditions of the system , the optimal control strategy results under this condition using the agent model Make a prediction as shown below: (twenty one) in, and are the weights and biases of the first hidden layer, and are the weights and biases of the second hidden layer, is the activation function.

[0052] Comparison of the calculation time results of the optimal control strategy under different methods Figure 3 As shown in the figure. The traditional model-based optimization method needs to solve the system optimization model every time the optimal control strategy is calculated. Since the central air-conditioning system optimization model has many parameters to be optimized and the nonlinearity between the parameters is high, the time required for model solution is relatively high, and the average calculation time reaches 176.6 s. In contrast, this method does not need to solve the system optimization model, and can directly predict the optimal control strategy through the proxy model. The average calculation time is only 0.1 s, which is 99.9% lower than the traditional method.

[0053] Comparison of the operating energy consumption results of the central air-conditioning system under different methods Figure 4 As shown in the figure, it can be seen that the prediction result of the optimal control strategy of this method comes from model optimization. Compared with the traditional control rule-based method (the chilled water outlet temperature of the chiller is always 7°C, and the frequency of the water pump and cooling tower fan is always the industrial frequency), the optimal control strategy obtained has a better optimization effect, and its daily operating energy consumption is reduced from 11149.8 kWh to 9695.2 kWh, and the energy saving rate reaches 13.0%.

[0054] In summary, the method proposed in the present invention realizes the rapid and accurate calculation of the system's optimal control strategy by constructing a control strategy agent model, can realize real-time optimization control of the central air-conditioning system while ensuring the optimization accuracy of the control strategy, and can be effectively applied to the efficient and real-time operation control scenarios of the central air-conditioning system.

[0055] Embodiment 2: This embodiment is a storage medium on which a computer program that can be executed by a processor is stored. When the computer program is executed, the steps of the real-time optimization control method of the central air-conditioning system in Embodiment 1 are implemented.

[0056] Embodiment 3: A central air conditioning system real-time optimization control device comprises a memory and a processor, wherein the memory stores a computer program that can be executed by the processor, and when the computer program is executed, the steps of the central air conditioning system real-time optimization control method of embodiment 1 are implemented.

Claims

1. A real-time optimization control method for a central air-conditioning system based on an agent model, characterized in that: include: Obtain the real-time operating conditions of the target central air-conditioning system, input the operating conditions into the trained proxy model, and predict the optimal control strategy results under the operating conditions; The training of the proxy model includes: Determine an operating condition set of a target central air-conditioning system and a set of control parameters to be optimized for each device in the system; the operating condition set includes a numerical range of multiple types of operating condition parameters, and the set of control parameters to be optimized includes a numerical range of multiple types of device operating parameters for each device in the system; Based on the equipment composition and topological structure of the target central air-conditioning system, as well as the optimization objectives and constraints, a mathematical optimization model of the central air-conditioning system is established; the optimization objectives include minimizing the total operating energy consumption of the system, and the constraints include energy conservation constraints; Based on the numerical range of each parameter in the target central air-conditioning system operating condition set, a plurality of operating condition samples are randomly generated, and the mathematical optimization model under each operating condition sample is solved to obtain the optimal control strategy corresponding to the parameters in the control parameter set to be optimized under each operating condition sample, and a training data set is constructed based on the operating condition samples and the corresponding optimal control strategy; Based on the training data set, train a proxy model, which is built using a machine learning algorithm; The operating condition parameter categories in the operating condition set include terminal cooling load, outdoor temperature and outdoor humidity; the parameter categories in the control parameter set to be optimized include set value parameters of each device of the target central air-conditioning system.

2. The real-time optimization control method for a central air-conditioning system based on an agent model according to claim 1, characterized in that: The set value parameters of each device of the target central air-conditioning system include the chilled water outlet temperature of the chiller, the operating frequency of the chilled water pump, the operating frequency of the cooling water pump, and the operating frequency of the cooling tower fan.

3. The real-time optimization control method for a central air-conditioning system based on an agent model according to claim 1 is characterized in that: The constraints include: The energy conservation constraint between the cooling water side and the chilled water side, that is, the total heat rejection of the cooling tower should be equal to the sum of the cooling load and the operating power of each chiller; The energy conservation constraint between cooling capacity and cooling load, that is, the cooling load should be equal to the sum of the cooling capacity of each chiller; Energy conservation constraints for the chilled water loop; Energy conservation constraints for the cooling water loop.

4. The real-time optimization control method for a central air-conditioning system based on an agent model according to claim 1, characterized in that: The optimization algorithms for solving the mathematical optimization model include genetic algorithm, particle swarm algorithm and grey wolf algorithm and their variants.

5. The real-time optimization control method for a central air-conditioning system based on an agent model according to claim 1, characterized in that: The machine learning algorithms used to build the proxy model include extreme gradient boosting, recursive neural network, artificial neural network, and support vector machine regression.

6. A storage medium having stored thereon a computer program executable by a processor, characterized in that: When the computer program is executed, the steps of the real-time optimization control method for the central air-conditioning system described in any one of claims 1 to 5 are implemented.

7. A central air conditioning system real-time optimization control device, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor, characterized in that: When the computer program is executed, the steps of the real-time optimization control method for the central air-conditioning system described in any one of claims 1 to 5 are implemented.

Images