非线性动态系统中的反馈神经网络逼近*

人工神经网络常用于非线性动态系统建模,动态子系统中可以运用多种形式的静态非线性。为完成非线性动态系统建模,利用神经网络结构来近似静态非线性关系。自然地,建模过程中会出现相应的误差估计。分析了基于离散化和数据抽样系统结构因素建模所引起的误差,发掘出了等价输入输出结构可以作为误差分析的一种度量工具,刻画出了模型设计误差和逼近误差之间的相依关系并找到了如何选择合适的系统结构来减小误差。

Approximation by Neural Network in Nonlinear Dynamic Systems

ArtificialSneural network is usedSfor the modeling ofSnonlinear dynamic system.SVarious forms ofSnonlinearSstatic are used in dynamicSsubsystem. In this paper,Sin order to complete theSmodeling of nonlinearSdynamic system, we use neural network architecture toSapproximateSnonlinear static relationship.SNaturally,SthereSwill beScorresponding error estimatesSin the process of modeling. This paper analyses theSerror caused by the discretizationSmodelingSfactorsSand dataSsamplingSsystemSbased on the structure,Sdeveloped the equivalentSinput and output structure as aSmeasurement toolSfor error analysis,SportrayedSthe dependence relationship between modelSdesignSerror and approximation errorSSand findShow to select the appropriate structure toSreduce the error.