Non-linear model approximation using new multivariate Hammerstein block structure model for large processes

Abstract

In this paper, a new approach is discussed to achieve a full order approximation model which leads to reduced order models for the large scale industrial process models. The model class used for this purpose is “block structure” models. The final aim is to achieve the reduced order model which is derived from the full order block structure model. Previously, Hammerstein structures have been used to approximate a mathematical non-linear model of a process by black box identification mainly. Input-Output Hammerstein structure, defined as classical Hammerstein model has been extended to new Hammerstein structure making use of states and inputs, hence called as input-state (IS) Hammerstein structure. During the research, it is proven that expansion of Taylor series leads to IS Hammerstein structure. The accuracy of the approximation is improved by including higher order approximation. IS full order Hammerstein model has been implemented on test case (high purity distillation column). It gives good approximation of the original non-linear system in full order. Over an operating domain of a process, the Input-State Hammerstein structure provides opportunities to achieve the reduced order model which is final goal of the research.