In order to solve the problem that the current fan coil units control method only takes indoor temperature as a single control object and ignores humidity, an office building in Beijing with fan coil units and fresh air system was studied. To obtain a better joint control satisfaction rate of indoor temperature and relative humidity, a reinforcement learning control method based on action intervention was proposed for regulating the air supply volume of fan coil units. A reinforcement learning algorithm was deployed using TensorFlow, a building energy system simulation model was built in TRNSYS, and the proposed algorithm was trained, tested and evaluated by using a self-developed TRNSYS-Python co-simulation platform. The results show that the proposed control method can improve the joint control satisfaction rate of indoor temperature and relative humidity by at least 9.5% compared with the traditional on-off control and rule-based control. It is concluded that the proposed method is valuable in engineering application and provides a new research idea for improving indoor thermal comfort in buildings.