Structure optimization for multiple model control

Abstract

This paper deals with structure optimization of the network of linear local models. The proposed technique assumes initial equidistant partition of the operating space. The structure of the model is then optimised through iterative algorithm that includes merging similar local models. The merging condition is based on the computation of Mean Squared Error and similarity of the local models in closed loop given by the gap metric. The approach is illustrated by a simulation study of a pH neutralization process.