Abstract: In view of the equipment safety problems caused by the high operation temperature of the indoor substations and the disturbing noise problems caused by relevant heat dissipation measures, this paper proposes a method for optimizing the design parameters of indoor substation air inlets based on finite element simulation and deep reinforcement learning to obtain the optimal ventilation and heat dissipation effects. Firstly, the temperature field, fluid field and sound field of the indoor substations are modeled and simulated with the finite element analysis method. Then, based on a large number of simulation data, the convolutional neural network is used to establish the prediction model of temperature and noise. Finally, considering the noise constraint, the maximum entropy reinforcement learning framework based SAC algorithm is used to optimize the design parameters of the air inlets with the goal of minimizing the indoor temperature of the substation. The research results show that the optimized air inlet design scheme can effectively reduce the indoor temperature in the substation, and at the same time make the noise meet the requirements of national regulations.