文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Radial Basis Functions for Process Control
侵权投诉

Radial Basis Functions for Process Control

Radial basis function (RBFs) neural networks provide an attractive method for high dimensional nonparametric estimation for use in nonlinear control. They are faster to train than conventional feedforward networks with sigmoidal activation networks ("

Radial Basis Functions for Process Control Lyle H. Ungar Tom Johnson Richard D. De Veaux University of Pennsylvania Voice Processing Corp. Princeton University Radial basis function (RBFs) neural networks provide an attractive method for high dimensional nonparametric estimation for use in nonlinear control. They are faster to train than conventional feedforward networks with sigmoidal activation networks (\backpropagation nets"), and provide a model structure better suited for adaptive control. This article gives a brief survey of the use of RBFs and then introduces a new statistical interpretation of radial basis functions and a new method of estimating the parameters, using the EM algorithm. This new statistical interpretation allows us to provide con dence limits on predictions made using the networks.

Abstract

KEYWORDS: radial basis functions, nonparametric regression; function approximation; nonlinear control.

1 Introduction Radial basis functions (RBFs), are a form of neural network1 for approximating nonlinear relationships. They typically take the form

y= X wj j (x) j=1

k

(1)

where the functions j (x) are Gaussian basis functions with centers j and widths j or, as we will interpret them in this paper, Gaussian density functions with mean j and standard deviation j . In control applications, the mapping the RBFs represent could be either a model of a plant (where e.g. x is the current state of the plant and the current control action and y is the next state of the plant) or a controller (where e.g. x is the current state of the plant and the desired next state of the plant and y is the control action to be taken). In real applications, the true state is generally not measurable, so an ARMA-style input consisting of a window of current and past measurements is used. Once one has a plant model, it can be incorporated into any model-based nonlinear control scheme such as MPC Psichogios and Ungar, 1991; Hunt et al., 1992]. 1 The term\neural network" is sometimes controversial, but radial basis functions are loosely inspired by parts of the visual cortex, where signals from neighboring receptors in the eye are combined with a roughly Gaussian grouping.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email