Steady state controller design for aero-engine based on reinforcement learning NNs

摘要

An aero-engine optimal steady state controller based on reinforcement learning neural networks (NNs) was proposed in this paper. The presented reinforcement learning NNs can achieve the optimal control objective by constructing two interconnected modules (i.e. action module and critic module). For the state variable models of small perturbation on steady operating points, the double-variable control of an aero-engine is accomplished by two similar backing propagation (BP) NNs. The simulation results show that the presented controller has the perfect performance with the smooth transition process. It not only has strong anti-interference ability and adaptability, but also has excellent robustness to the change of aero-engine model parameters.