基于结构的神经网络在系统建模中的应用  

Application of Neural Networks Based on System Architecture in Nonlinear Modeling

作　　者：乔俊伟[1] 詹永麒[1] 施光林[1] 

机构地区：[1]上海交通大学机械工程学院,上海200030

出　　处：《系统仿真学报》2002年第5期557-561,共5页Journal of System Simulation

摘　　要：针对复杂非线性系统的建模问题,提出一种基于结构的神经网络建模方法。将一个复杂系统转化为若干个较简单的子系统,分别用函数链神经元建立各子系统的模型,然后根据子系统间的相互作用关系连接成一个完整的网络,即基于结构的神经网络。分析了网络的性能,探讨了在复杂系统建模中的应用。对Y2-Hc10型先导式溢流阀的仿真表明了该方法是可行和有效的。A novel method of nonlinear system modeling is presented, which employs neural networks based on system architecture. This method translates a complicated system into some simple sub-systems. Each sub-system is modeled with a functional link neuron. Then the neurons are connected into an integrated network according to the relationship between these sub-systems. The performance of the networks is analyzed, and its application in complicated system modeling is discussed, too. The feasibility and validity of the method is illustrated by a simulation on Y2-Hc10 pilot operated relief valve.

关 键 词：神经网络 系统建模 函数链神经元 液压元件 计算机仿真 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] TH137.5[自动化与计算机技术—计算机科学与技术]