非线性动态系统中的反馈神经网络逼近

Approximation by Neural Network in Nonlinear Dynamic Systems

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。

Artificial neural network is used for the modeling of nonlinear dynamic system.Various forms of nonlinear static are used in dynamic subsystem.In order to complete the modeling of nonlinear dynamic system,neural network architecture was used to approximate nonlinear static relationship.Naturally,there will be corresponding error estimates in the process of modeling.The error caused by the discretization modeling factors and data sampling system based on the structure were amalyred,the equivalent input and output structure as a measurement tool for error analysis were developed,the dependence relationship between model design error and approximation error were prorated and how to select the appropriate structure to reduce the error was found.