航空发动机递归神经网络分路式解耦控制  

Aeroengine Separate Decoupling Control Based on Recursive Neural Networks

作　　者：刘建勋[1] 李应红[1] 陈永刚[1] 尉询楷[1] 

机构地区：[1]空军工程大学工程学院,陕西西安710038

出　　处：《航空动力学报》2005年第2期287-292,共6页Journal of Aerospace Power

摘　　要：针对航空发动机多变量控制中变量之间的耦合问题,提出了一种基于递归神经网络的分路式动态解耦控制方法,给出了发动机双路式解耦控制系统的结构及其解耦原理和算法。利用递归小波网络较强的动态非线性映射能力,在线完成发动机各控制通道的模型辨识,并回馈对应的灵敏度信息;神经网络PID控制器根据回馈的信息在线自适应调整参数,实现发动机各通道的准确跟踪和分路独立控制。仿真表明,该方法在保证控制系统良好的动态和稳态性能的同时,有效地减小了各回路之间的耦合影响,能够成功应用于发动机控制系统的解耦。To solve the problem of coupling in aeroengine multivariable control system, a dynamic decoupling control method based on recursive wavelet neural networks is presented. The structure of engine decoupling control system, as well as its decoupling principle and algorithm, is given. The structure includes two separate control loops. The two recursive wavelet networks, which employed normalized wavelet basic functions, were used as the decoupling identifier, and two PID neural networks were used as the controller. The wavelet networks identify the dynamic model of the engine and feed back the sensitivity information through on-line learning. The neural network PID controller updates the weights adaptively according to the on-line sensed information, and accomplishes self-governed control of each engine control loop. The simulation results of a turbofan engine show that the proposed method can effectively reduce the coupling influence of each control loop and assure satisfactory transient performance. It can be successfully applied to decoupling of aeroengine control system.

关 键 词：航空、航天推进系统 航空发动机 多变量控制 解耦 PID 递归神经网络 小波 

分 类 号：V233.7[航空宇航科学与技术—航空宇航推进理论与工程]