Intelligent Control Of Variable Cycle Engine Based On Dynamic Neural Network

With the development of science and technology,the new generation of variable cycle engine has become the focus of research in the field of aero-engine because it combines the advantages of turbofan engine and turbojet engine.Due to the increasing number of adjustable components in variable cycle engine,the control variables increase,the coupling degree between circuits increases,and the complexity of control system increases.Based on the project of "XX engine basic problem research" of a ministry,an intelligent control method of variable cycle engine based on dynamic neural network is proposed for a variable cycle engine,and a simulation verification is carried out based on the mechanism model of a variable cycle engine.Specific contents include:1.Research on component-level model of variable cycle engine.Through analysis of variable cycle engine research status at home and abroad,at the component level,within the framework of the modeling method based on variable cycle engine variable geometry characteristics and external bypass steady-state mapping relationship,the relationship between the variable geometry parts performance,based on power balance equation,energy balance equation and flow balance equation,a certain kind of variable cycle engine has been clear about the nonlinear component level modeling method,and has carried on the simulation of engine performance,dynamic neural network controller to lay the foundation for subsequent design.2.Design of dynamic neural network based on grey relational analysis.In view of the large structure of neural network,the approximation effect is good,but the ability of overfitting and network generalization is poor.The structure of the network is too small,and the learning ability is weak in training,which may lead to the problem of insufficient training accuracy.The structure growth algorithm based on error and the structure pruning algorithm based on grey correlation analysis method are proposed,and the hidden layer neurons of the neural network are adjusted layer by layer,so as to realize the structural optimization of the dynamic neural network.3.Design an intelligent controller for a variable cycle engine based on dynamic neural network,and conduct simulation verification.Aiming at the problem of variable cycle engine with many control variables and strong coupling,according to the engine control plan,the steady-state controller design based on structure growth based on error and structure pruningalgorithm based on grey correlation analysis is realized.Simulation results show that the proposed control method can solve the above problems well,and the control error is within0.5% without overshoot.Compared with the fixed-structure neural network controller,the network structure of the proposed method is more compact,the steady-state error is smaller,and the response speed is faster.4.Development of intelligent control simulation software for variable cycle engine.On the basis of determining the software development requirements and development plans,the system software program design and key technology analysis as well as software function development are carried out.The data collection of the original model is realized under the condition set by the user,and the neural network training sample is constructed.Under the neural network training method selected by the user,the training error and the change of hidden layer neurons are displayed.The trained dynamic neural network or fixed structure neural network is packaged into a variable cycle engine intelligent controller to realize the variable cycle engine intelligent control and to demonstrate the controller's tracking effect on control instructions and the change of control variables.