Interval Multi-Objective Optimization for Low-Carbon Building Energy Management System Upon Deep Reinforcement Learning

To improve building energy efficiency and reduce the impact of scheduling uncertainty, a low-carbon optimization method for intelligent buildings based on deep reinforcement learning interval multi-objective optimization is proposed. Firstly, interval mathematics is used to model the multiple uncertainties in the system. Secondly, considering the system's carbon emissions and carbon transaction costs, optimize the system operation with the goal of the lowest comprehensive operating cost and the best user comfort. Thirdly, to solve the problem of interval multi-objective optimization, deep Q-network (DQN) and interval multi-objective particle swarm optimization (IMOPSO) are proposed for “offline training” and “online guidance”. The low-carbon optimization scheduling problem of intelligent buildings under multiple uncertain factors is efficiently solved. Case studies show that the proposed IMOPSO based on DQN can consider the system's low carbon, economical and user comfort, which effectively improves the system's ability to deal with uncertain factors.