采用遗传算法训练对角递归神经网络预测控制器
TRA ININGOF D IAGONAL RECURRENT NEURAL NETWORK PRED ICTIVE CONTROLLER USING GENETIC ALGORITHM S
Abstract
摘 要: 本文提出了一种基于广义预测控制的神经网络预测控制方案. 预测控制器由对角递
归神经网络预测控制器和前向神经网络静态补偿器组成. 两种神经网络均采用遗传算法进行训
练. 仿真实验表明, 对于带纯时延的非线性被控对象, 采用遗传算法设计的对角递归神经网络预测
控制器具有令人满意的控制性能.
Abstract: Th is paper p ropo ses a neural netwo rk p redict ive cont ro l scheme based on generalized p redic2
t ive cont ro l (GPC). The p redict ive cont ro ller is made of diagonal recurrent neural netwo rk p redict ive con2
t ro ller (DRN PC) and feed fo rw ard neural netwo rk steady2state compensato r (FNC). Two k inds of neural net2
wo rk s are t rained using genet ic algo rithm s. The simulat ion results show sato sfacto ry perfo rmance of the neu2
ral netwo rk p redict ive cont ro ller fo r nonlinear p lants w ith dead t ime.