Combining Statistics and Expert Systems with Neural Networks for Empirical Process Modeling

Neural networks, statistical analysis, and expert systems have been successfully applied to a number of process industries for the purpose of process modeling and control. In most of the applications, it is often beneficial to integrate statistical methods with neural networks due to the similarity of neural network training and statistical modeling. The main focus of the paper is to combine statistical methods into neural network training in various steps including data pre-processing, variable selection, network training and testing, and post validation. The statistical methods addressed here are outlier detection, missing data replacement, partial least squares regression, principal component analysis, and cross-validation. A framework of using an expert system to guide the procedures to build a neural network is proposed.