基于HGA优化RBF网络的污水总氮软测量  

Soft Sensor for Total Nitrogen in Waster Water Based on RBF-NN Optimized by Hierarchical GA

作　　者：梁倩[1] 吕勇哉[1] 

机构地区：[1]上海交通大学自动化系,上海200240

出　　处：《计算机仿真》2008年第7期137-140,183,共5页Computer Simulation

摘　　要：污水处理系统是一个包含海量信息的非线性复杂系统。目的是对某污水处理厂生物脱氮系统的出水总氮(TN)进行软测量建模。先用主元分析方法实现输入变量的降维和去相关,简化径向基函数(RBF)网络的输入。再应用递阶遗传算法(HGA)确定合理的RBF网络隐层节点数、基函数宽度和中心。能够同时优化网络参数和拓扑结构,在全局范围内寻找RBF参数的最优解,实现了RBF网络的自适应优化。应用该模型对出水总氮软测量进行仿真,结果表明了该网络模型的可靠和有效,说明该软测量模型具有工业应用价值和意义。Waster water treatment system is a complex non-linear system containing a huge amount of information.The purpose is to build a predictive model for the development of soft sensor in Total Nitrogen（TN） estimation in the outlet of a biological Nitrogen removal system of a waster water treatment plant.PCA is used to perform both dimension-reduction and de-correlations in input space to simplify the inputs of RBF Neural Network.The hierarchical GA is further used to optimize the number of hidden layer nodes,width and center of the basis function.The proposed approach can optimize the network parameters and structure at the same time,search for the optimal solution of RBF parameters in overall range and realize the self-adapting optimization.The simulation is conducted on TN soft–sensing and shows the reliability and effectiveness of the network model.The results indicate the value of the soft-sensing model in industrial application.

关 键 词：主元分析 径向基函数网络 递阶遗传算法 污水处理 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] X830.3[自动化与计算机技术—控制科学与工程]