研究了一种基于神经网络的航空发动机全包线PID控制器设计方法。首先在全包线内选定若干典型工况点,并通过遗传算法离线优化PID控制器参数。然后通过BP神经网络训练建立飞行高度和马赫数与PID参数的非线性映射关系,亦即建立起基于神经网络的航空发动机全包线最优PID控制器。最后,将该控制器应用于某型涡扇发动机稳态控制和飞行过程控制仿真,与原控制器比较,控制效果获得有效改善。

An aero-engine optimal PID control scheme was presented, which was suitable for the engine whole envelope with a neural network approximation. Firstly, the optimal PID controller parameters were designed by genetic algorithm (GA) for a certain turbofan engine simplified real-time simulation model around some design-points. Then according to these optimized PID controller parameters, a neural network approximation was trained to imitate the relation between the parameters (K_p, K_i, K_d) of local optimal controllers and the flight conditions (H, Ma), so that a neutral network based optimal PID controller was obtained. Finally, this new controller was applied to a certain turbofan engine control system simulation for the steady control, acceleration control and whole flight envelope control. The simulation results show that the aero-engine control system has excellent characteristics in design points and whole flight envelope. In this approach, the controller has a simple structure, and is easy and practical for real-time realization.