A statistical perspective of neural networks for process modeling and control
S. Joe Qin · 2002
Multilayer neural networks have been successfully applied to industrial process modeling and control. The prediction variance of neural networks from gradient based learning is analyzed in the presence of correlated process inputs. Several biased regression approaches, including ridge regression, principal component analysis, and partial least squares, are integrated with neural net training to reduce the prediction variance. Examples are given to illustrate the improvement of the integrated approaches.>