Abstract: An energy-efficient control method for air conditioners based on adaptive network constraints and Q-learning is proposed to reduce the high energy consumption of air conditioners under traditional integral differential control. First, the reward matrix is constructed by using the expert system and the Reward function, and the air conditioner operating parameters corresponding to its elements are divided into data sets A and B. Next, the Radial Basis Function (RBF) neural network model is initialized, the constructed network is trained using dataset A as training data, and the network constraint model is validated using dataset B until the accuracy rate reaches more than 90%. Finally, the network constraint model is combined with the Q-learning algorithm to realize the optimal strategy selection for air conditioning energy saving. Experiments show that the algorithm is able to achieve energy saving effect compared to traditional air conditioning control logic without sacrificing user comfort.

Key words: Expert system, Radial basis neural network, Reinforcement learning, Comfortable energy saving