Multi-objective optimization using differential evolution for dynamic model structure selection

Abstract

Model structure selection is one of the procedures of system identification. The main objective of system identification is to select a parsimony model that best represents a dynamic system. Therefore, this problem needs two objective functions to be optimized at the same time i.e. minimum predictive error and model complexity. This research presents a new developed algorithm called multi-objective optimization using differential evolution. One of the main problems in identification of dynamic systems is to select a minimal model from huge possible models to be considered. The important concepts in selecting good and adequate model are used in the proposed algorithm are elaborated, including the implementation of the algorithm for modelling dynamic systems. The related issue such as parameter tuning on the proposed algorithm is discussed.