Direct Thrust Inverse Control of Aero-Engine Based on Deep Neural Network

Abstract

Abstract A novel thrust control method based on inverse control is proposed to improve engine response ability. The On Line Sliding Window Deep Neural Network (OL SW DNN) is proposed and adopted as inverse mapping model modeling method of inverse control. The OL SW DNN has deeper layer structure, which makes the inverse mapping model have stronger fitting capacity for nonlinear object than traditional NN. Moreover, due to adopt on-line learning modeling method, the proposed adaptive control method can obtain desired thrust whether engine degrades or not. The comparison simulations of the traditional control method based on PID and the proposed control method are carried out. Compared with the traditional control method, the proposed control method can obtain desired thrust when the engine degradation occurs, but also has fast response ability (the acceleration times for engine thrust increase to 95 % thrust of acceleration object decreases by 1.35 seconds).