Nonlinear Simulator model Identification of a Walking Beam Furnace Using Recurrent Local Linear Neuro-Fuzzy Network

The walking beam furnace (WBF) plays a critical role in a steel production factory, and has complex nonlinear dynamic behaviour. Thus, using the conventional physical principles to build a model of the process leads to a time-consuming modeling procedure and when the number of operating set-points in walking beam furnace increases the conventional modeling problem difficultly will be solved, indeed analytical models cannot be applied or can not give satisfactory results. This paper proposes intelligent nonlinear simulator model for a real walking beam furnace in a steel production factory using nonlinear sub-system identification technique based on the recurrent local linear neuro-fuzzy (RLLNF) network. This model is trained using the local linear model tree (LOLIMOT) algorithm, which is a tree-structure divide-and-conquer algorithm. It is the first time that such nonlinear simulator (recurrent) model for a real WBF based on locally linear neuro-fuzzy modelling is developed. The recorded data of Iran Alloy Steel Company are used to identify and evaluate the RLLNF simulator model of walking beam furnace.