发动机神经网络自适应控制仿真及分析

The Simulation and Analysis for Neural Network Adaptive Control of Aero-engine

针对航空涡喷发动机,给出了一种神经网络自适应控制结构。研究了该控制回路中辨识网络和控制网络的结构、离线训练以及初始权值的确定等问题。发动机控制系统仿真表明:该方法用于发动机稳态控制能取得较好的控制品质,当偏离设计点时,通过神经网络很强的学习、适应能力,在线修正网络参数,使控制系统仍保持良好性能,鲁棒性好。

A neural network adaptive control structure for a certain aero-engine was presented. The problems such as structure, off-line training, determination of initial weight values and control network involved in the control loop were studied. The simulation of the engine control system indicates that aero-engine steady-state control can achieve good performance with this manner. When deviating from the design point, the parameter of the neural network will be modified on-line with its own strong learning and adaptive ability, which make the control system maintain good performance and strong robustness.