Energy-saving Control of Multi-split Air Conditioning Based on Multi-agent Reinforcement Learning
The multi-split air conditioning system is the main trend in the development of air conditioning,and research on its energy-saving control has gradually become a hot topic in the field of air conditioning.However,traditional energy-saving control methods face significant challenges in nonlinear feature representation,high generalization mechanism modeling,and precise self-adjustment of equipment parameters.To address this,a multi-agent deep reinforcement learning-based energy-saving control method for multi-split air conditioning is proposed.Firstly,the energy-saving control parameters of the multi-split air conditioning are selected to achieve parameter decoupling and dimensionality reduction.A highly generalized operational model for multi-split air conditioning is constructed using deep neural networks to reduce the training time of the multi-agent reinforcement learning model.Then,a multi-agent collaborative intelligent energy-saving control method for multi-split air conditioning is proposed,designing the state space,action space,and reward function of the agents.The agents are trained in a centralized manner and executed in a distributed manner to optimize the air conditioning control strategy.Meanwhile,environmental initialization conditions are designed to increase environmental diversity.Finally,extensive experiments are conducted on an air conditioning operation historical data set from three regions and two seasons.The results show that the proposed method can effectively control indoor temperature,with an increase in energy efficiency ratio of about 18％.
multi-split air conditioningprecise self-adjustment of parametersmulti-agent reinforcement learningdeep neural networksenergy-saving control
万方数据
维普
