Aero-Engine Model Predictive Control Based on Deep Reinforcement Learning
Aiming at the problem that the traditional model predictive control is difficult to ensure the control ac-curacy in the case of large model mismatch such as sudden change of operating conditions and large value faults,an improved aero-engine model predictive control method based on deep reinforcement learning is proposed.Firstly,the regulation effect of MPC on model mismatch is evaluated,and the impact of different types and degrees of model mis-match on MPC is analyzed.Secondly,a model predictive control method based on deep reinforcement learning(DRL-MPC)is proposed.Through deep reinforcement learning,the model deviation between the prediction model and the actual system is corrected to improve the prediction accuracy of the prediction model.Finally,the cost function of MPC is reconstructed,and an aero-engine improved MPC controller based on deep reinforcement learning is estab-lished.The traditional MPC method is compared with the proposed method.The simulation results show that the pro-posed method can effectively reduce the sensitivity of the MPC controller to the model mismatch without increasing the computational cost in the case of large model mismatch.
NETL
NSTL
万方数据
