Intelligent Optimization Control for Air Starting Systems Based on Deep Reinforcement Learning

In this paper, an intelligent optimization control method based on deep reinforcement learning is proposed for air starting systems of altitude test facility. A deep Q-network (DQN) controller with self-learning capability is employed to effectively improve the dynamic and steady-state performance of the air starting system under pressure and flow disturbances. Key design methods including state space selection, action output and reward function design are studied and given. Simulation results show that compared with PID control, the designed controller can realize no overshoot adjustment of the intake pressure under constant or variable operating conditions, and the adjustment time is shorter. The results verify the rapidity, stability and robustness of the proposed intelligent optimization control method.