Optimal Control Of Central Air Conditioning Based On Deep Reinforcement Learning

In recent years,the country has vigorously promoted the process of urbanization,and the number of high-rise buildings in the city is row by row,but the waste of energy consumption of buildings has always been a serious problem.Since the energy consumption of the central air-conditioning system is the highest proportion of the building energy consumption,the relevant control strategy has been established,but the energy consumption of the central air-conditioning in the next moment is often unknown,so that the control strategy cannot be adjusted in time The optimal state,and then can not better achieve the purpose of energy saving.Therefore,this paper uses deep reinforcement learning to predict the energy consumption of the central air conditioner at the next moment,and achieves energy saving of the air conditioning system through the optimal control strategy.In the optimization control of central air conditioning,this paper conducts research from the following three aspects based on deep reinforcement learning:(1)Aiming at the problem of the slow convergence speed of the classic DQN(Deep Q-Network)algorithm in the early stage of training,this chapter proposes a new DQN algorithm based on cooperative update of function approximation.The algorithm combines the linear function method on the basis of the classic DQN algorithm.In the initial stage of training,the linear function approximator is used to replace the behavior value function network in the neural network,and a rule for updating the policy value function is proposed,which is updated in collaboration with DQN.Value function parameters accelerate the parameter optimization of the neural network,thereby improving the convergence speed of the algorithm.The improved algorithm and DQN algorithm are applied to Cart Pole and Mountain Car problems.Experimental results show that the improved algorithm shortens the convergence time and has better convergence.(2)By analyzing the problems caused by the energy consumption of the central air conditioning system and the actual data collection,the relevant parameters of the central air conditioning system energy consumption prediction are established.Due to data loss and abnormality caused by power failure and crash during the data collection process,it is necessary to preprocess the actual data to avoid inaccurate prediction.Use reinforcement learning to model the energy consumption prediction problem of central air conditioning,and use the DQN-LF algorithm proposed in the first part to predict energy consumption.Finally,based on the central air-conditioning energy consumption data recorded by a college,the experiment was conducted.The experimental results show that the prediction method of energy consumption prediction method of central air conditioning system based on deep reinforcement learning has high prediction accuracy,which can provide guidance for the optimization strategy of central air conditioning system.(3)Aiming at the main part of the central air conditioning system,the cooling water system energy consumption of the central air conditioning system,this paper proposes an optimized control of the central air conditioning system based on deep reinforcement learning.First,analyze the central air-conditioning system and establish the main parameters of the optimal control,that is,the ratio of the flow of the cooling water system and the air volume of the cooling tower,establish the constraints of equipment operation according to its working state,and establish the goal of optimal control.When using deep reinforcement learning to solve the optimal control strategy at the next moment,because the energy consumption at the next moment is unknown,the fourth chapter DQN-LF central air conditioning energy consumption forecast is introduced to make up for the energy consumption at the next moment.The optimal control strategy is solved through deep reinforcement learning,and the central air conditioning control strategy is adjusted in time to achieve energy saving of the central air conditioning.