一种具有较强泛化能力的神经网络模型研究与应用  

Study and Application of a Class of Neural Networks Model with Better Generalization Ability

作　　者：王迎春[1] 耿长福[1] 

机构地区：[1]北京控制工程研究所,北京100080

出　　处：《航天控制》2002年第2期6-11,17,共7页Aerospace Control

基　　金：国家自然科学基金 ( 6 9984 0 0 7);贵州省科技攻关计划项目

摘　　要：分析了影响神经网络模型泛化能力的因素。以电解铝过程中氧化铝浓度的神经网络软测量为例 ,提出了利用先验知识确定网络结构 ,采用特定实验保证样本数量和质量 ,离线训练加在线学习修正模型等措施 ,改善神经网络模型的泛化能力。现场应用表明这些措施是有效的。这样建立的神经网络模型准确 ,泛化能力强 。This paper studies on the factors affecting the generalization ability of neural networks model Taking alumina concentration′s soft sensing in the process of aluminum electrolysis as an example, some measures are presented to improve model′s generalization ability They include constructing neural networks with prior knowledge, ensuring the quantity and quality of samples through the special experiments, and training neural networks both off-line and on-line The practical application shows their effectiveness The neural network model based on these design methods proved to be precise It has better generalization ability and provides reliable guarantee for the advanced process control

关 键 词：神经网络 建模 泛化能力 氧化铝浓度 铝电解 工业控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP183[自动化与计算机技术—控制科学与工程]