Study On Aero-engine Control Based On Neural Network

Aero-engine was a controlled object with such features as complicated working condition. For such controlled object, classical feedback control theory and optimal control theory can't guarantee optimal performance. In this paper, a new type ofmethod------the PED control based on neural network is studied. It is used in theaero-engine transient control. The work of this thesis is as follows:They were presented in this thesis: one is a multivanable decoupling PID control system based on DRNN(diagonal recurrent neural network), and the other is a multivanable decoupling PID control system based on QDRNN (quasi-diagonal recurrent neural network). It was designed for aeroengine control by studying DRNN/QDRNN and GMD (gradient decent method). It was illustrated the structure of the controller and the theory of the algorithm. The result of computer simulation shows that a multivariable decoupling PID control system based on DRNN has fine performance of decoupling capability; on the other hand, QDRNN, which based on DRNN, has better performance of decoupling capability than DRNN with also simple structure. Especially, all training algorithm of DRNN may be utilized in QDRNN only by little transformation.