Neural Network Control Systems For Aeroengines Based On Genetic Algorithms

Aeroengine is a complicated, time varying, non-linear multivariable control plant with uncertainty over full flight envelope. The control algorithm design using traditional control theory depends on the accuracy model of the plant which is difficult to obtain. The control algorithms for aeroengines based on artificial neural networks and genetic algorithms are studied in this paper.First, the neural network based PID controller for multivariable engine control system is designed. The multivariable control system consists of two subsystems: the rotor speed control subsystem and pressure ratio control subsystem. The parameters of the PID controller are modulated by two three-layer feed-forward sub-networks for each subsystems. The multivariable control is realized. The weights of each sub-network are updated adaptively utilizing BP algorithm. Simulation results show that the control system has the propriety of adaptability, reduced coupling and satisfactory dynamic and static performances.Then, an adaptive control system of engine is developed with recurrent neural network. The recurrent neural network is used to identify the engine model. The on-line adaptive law is used to adjust the weights of the neural network. Numerical simulation results show the effectiveness of the method and self-learning ability of the system in the full envelope.The PID controller design method with genetic algorithm is studied. The parameters of the PID controller was optimized by genetic algorithms within a certain range. The parameter of the controller is obtained automatically.Finally, the rotor speed control system for aeroengine is proposed using genetic algorithms and neural networks. The weights of the feedforward neural network are optimized using genetic algorithms. Simulation results indicates a satisfying capability of the aeroengine control systems in full envelope.