Hybrid-neural modeling of a complex industrial process

Abstract

This paper deals with a complex industrial modeling problem the modeling of a Linz-Donawitz steel converter. The main purpose of the paper is to show that in such cases where classical modeling methods cannot be applied successfully and where the nature of knowledge available is heterogeneous hybrid intelligent approach can give new possibilities. The proposed hybrid advisory system is composed of different neural networks and rule-based systems exploiting the advantages of both approaches. The paper describes the main features of the modeling task, lists the most serious difficulties of this industrial problem and presents the motivations behind the construction of hybrid solution. At the end it gives details about the architecture of the proposed system and an overview about the results achieved.