基于递阶遗传算法的RBF神经元网络板形板厚综合控制

Strip Flatness and Gauge Complex Control of RBF Neural Network Based on Hierarchical Genetic Algorithm

该文首先用递阶遗传算法 (HGA)设计RBF神经网络 ,不仅可以同时确定网络参数 (连接权、隐节点中心和宽度 ) ,而且解决了网络拓扑结构的优化训练问题 ;而后针对板带材轧制是一个复杂的非线性过程 ,板形控制 (AFC)和板厚控制 (AGC)又是相互耦合的一个综合系统等特点 ,建立了基于过程最优的权值在线自学习算法的RBF神经元网络的板形板厚多变量综合控制系统。仿真结果证明了此AFC -AGC控制系统具有良好的自适应跟随和抗扰性能 ,其控制效果优于传统的解耦PID控制。

In the paper, the authors firstly use hierarchical genetic algorithm t o design RBF neural network. This method trains not only network parameters such as centers, widths and connection weights, but also solves configuration proble m of RBF network. Secondly, because strip rolling is a very complicated nonlinea r process, and characterized by couple between automatic flatness control (AFC) and automatic gauge control(AGC), authors present strip flatness and gauge com plex control system of RBF neural network based on weights online self-learning algorithm of process optimization. The simulation results show that this kind of new controller has good performances of adaptively tracking target and resistin g disturbances and is superior to the conventional decoupled PID control in term s of improving the strip flatness and gauge accuracy.