增强神经网络辨识模型泛化能力的研究

Research on enhanced generalization ability for the ANN’s identification model

摘要:

神经网络(Artificial Neural Network,ANN)辨识模型的泛化能力是其最主要的性能之一,增强ANN模型的泛化能力也是近年来国内外有关专家学者研究的重点问题。大量研究表明,ANN模型泛化能力的改善与很多因素相关联,其中恰当的性能指标函数设计是一个重要影响因素。文中在分析常见的基于均方误差最小原则的性能指标函数基础上,通过加入ANN辨识模型权值间的延迟信息,进而获得一种改进型性能指标函数。通过仿真,验证了所设计的改进型性能指标函数对增强ANN辨识模型的泛化能力是有效的。

Abstract:

Generalization ability of Artificial Neural Network (ANN) identification model is one of most main performance, also is one of key questions researched by domestic and foreign concerned experts in the recent years. Generalization ability of ANN’s identification model concerns with many factors, and appropriate designed performance index function is an important influence factor. After common performance index function is analyzed based on the mean error function smallest principle, a kind of improved performance index function is obtained through joined the power values to the time delay information in the paper. The massive simulation computation shows that improved performance index function is effective to enhance generalization ability of ANN models.