A method for automatic generation of plant-wide inference engines

Abstract

An automatic modeling method, which creates an inference engine out of raw data, is suggested. The inference engine is used by the automation system to assist operators in decision making. We aim at plant-wide modeling of industrial processes and we therefore prioritize fast and approximate solutions. The suggested method is capable of creating models with hundreds of variables. As a basic structure we utilize multi-dimensional histograms, which at a lower level model the relations of two or three variables. These sub-models are connected in a tree structure. Both the variable selection of sub-models and the tree structure connections are based on Shannon entropy.