Research On Control System Of Turboshaft Engine Based On Component Method Modeling And Neural Network

Turboshaft aero-engine technology has always been a strategic focus for every air power since its birth,representing core technology and comprehensive strength.As one of the most cutting-edge technologies,great attention is drawn to Full Authority Digital Control Technology in academic circles and industry.Meanwhile,the artificial intelligence is put into practice in turboshaft aero-engine technology so widely that tremendous achievement is obtained currently.Based on it,this paper proposes the mathematic component model of turboshaft engine and takes neural network control technology into consideration.First of all,taking a certain type of turboshaft engine as the basis a component-level non-linear mathematic model is established under the MATLAB/Simulink environment.In terms of this method,the engine model is comprised of six relatively independent sub-components with interrelation,such as air inlet,compressor,combustor,gas turbine,power turbine and exhaust nozzle.And these components are in harmony with each other based on the aerothermodynamic law.Further the steady-state and dynamic-state model are established with a numerical solution to guarantee the stability and validity.This component-level model lays the foundation of the further neural-network control technology research.Secondly,with the power turbine single-loop PID control method,BP neural-network,optimized by genetic algorithm(GA),is introduced to intelligently set the parameters of PID controller such as k _p、k _i、k_d.Ahead of the simulation,the measured data is preprocessed with the Kalman filter.The simulation results indicate that both steady-state and dynamic-state requirements for the normal operation of the engine are satisfied with a steady rotate speed of power turbine but abnormal overshoot and gross smooth of some parameter graph requires being improved further.Finally,gas turbine rotate speed PID controller is introduced into the power turbine single PID control loop to constitute cascade control loop with the NARX(Nonlinear Auto Regressive with eXogenous inputs)neural-network.Meanwhile,information fusion of oil supply with the wavelet transform(WT)method is introduced to smooth the graphs and lower the overshoot.The simulation results indicate that NARX-PID controller achieves the desired effect indeed and enhances the sensitivity of model except the effect of transition-state with the WT.Both the BP-GA single-loop and the NARX with WT cascade PID method satisfy the steady-state and dynamic-state control requirements for the engine with their own advantages in terms of smooth and overshoot.In general,NARX neural network is the better.