Abstract: In order to fully tap and effectively utilize the energy saving and emission reduction potential of buildings, a low-carbon optimal dispatch method of smart buildings based on interval multi-objective optimization with deep reinforcement learning is proposed. Firstly, the unit parameters, building temperature and multiple uncertainties of source and load in the system are modeled by using interval number and other methods. Secondly, taking into account the system carbon emission and carbon trading mechanism, the system operation is optimized with the goal of the lowest comprehensive operation cost and the best user comfort.In order to solve the interval multi-objective optimization problem, a new multi-objective optimization algorithm is proposed,which combines deep Q network of deep reinforcement learning and interval multi-objective particle swarm optimization algorithm,and performs “offline training” and “online guidance” to efficiently solve the low-carbon optimization dispatch problem of smart buildings under multiple uncertainties. The case results show that the proposed interval multi-objective optimization algorithm with deep reinforcement learning and dispatch model can take into account the low carbon, economy and user comfort of the system,and effectively improve the ability of the system to deal with multiple uncertainties.