Research On Operation Strategy Of Ice Storage Air Conditioning System Based On Reinforcement Learning

The ice storage air conditioning system contains an energy storage system,which uses the surplus electric energy at night to store cold and assist chillers to supply cold during the day.Compared with ordinary air conditioning system,ice storage air conditioning system has higher initial investment.If the operating strategy is unreasonable,problems such as high operating costs and excessive energy losses will occur,and good economic and energysaving benefits cannot be achieved.Therefore,we need to study the operation strategy optimization method of ice storage air conditioning system.Some scholars have proposed model predictive control,particle swarm optimization,fuzzy PD control and other methods to optimize the operation strategy of ice storage air conditioning.These methods need to establish the thermal dynamic model of buildings.However,there are many factors affecting the thermal characteristics of buildings,which make the modeling process complex.Reinforcement learning is a data-driven method,which realizes adaptive learning and online optimization through trial-and-error mechanism,and avoids complex modeling process.This paper studies the optimization method of the operation strategy of ice storage air conditioning system based on reinforcement learning.Firstly,the power consumption models of the main energy consuming components in the ice storage air conditioning system in an operation cycle are established,which are the basis of calculating operation cost and analyzing power change.Then,on the premise of ensuring thermal comfort inside the shopping mall,the deep Q network algorithm is used to calculate hourly cooling ratio of chillers and ice storage tanks to reduce the operating cost.The experimental results show that the control method based on deep Q network can improve the indoor thermal comfort slightly.Compared with the rule-based control method and PID control method,the operation cost can be reduced by 7.43% and 8.12% respectively.Finally,the Markov decision process model of ice storage air conditioning system is established to smooth the power curve,and the operation strategy is optimized by using the deep deterministic strategy gradient algorithm.The experimental results show that the efficiency of the chillers are improved after optimization.Compared with the chillers priority control method and ice melting priority control method,the peak-valley difference of load can be reduced by 22.9% and 22.9% under 100% typical daily load,and by 28.8% and 25.1% under75% typical daily load.In general,the economic and energy-saving benefits of ice storage air conditioning system have been improved after optimization.