论文部分内容阅读
文章回顾苏联伊凡年科和他的同事们提出的数据处理组合法的基本概念。这种方法的优点是;(1)它可以确定非线性系统的结构;(2)这种方法把拟合数据与校核数据分开,从而避免了过拟合的毛病;(3)进行模型验证时,这种方法可采用多判据目标函数。 本文评论分析了上述优点以及GMDH的另外一些特点(诸如多项式网络和分层法概念),并与非线性系统辨识所用的其它方法进行了比较,还列出了这种方法在工业和环境建模问题方面应用的实例。
The article reviews the basic concepts of the data processing combination law put forward by Ivan Nervi and his colleagues in the Soviet Union. The advantages of this method are: (1) it can determine the structure of the nonlinear system; (2) this method separates the fitting data from the checking data to avoid over-fitting; (3) model verification When this method can use multiple criteria objective function. This paper reviews the above advantages and other features of GMDH (such as the notion of polynomial networks and delamination) and compares them to other approaches used in nonlinear system identification. It also lists the advantages of this approach in industrial and environmental modeling An example of a problem application.