Multi-objective opimization of MIMO control system using surrogate modeling

Abstract

A multi-objective optimization approach using surrogate modeling is applied to a nonlinear Multi Input Multi Outputs (MIMO) control system model to predict Pareto-front of objective functions which is defined using Integral Square Error (ISE). Typically, practical multi-objective optimization was highly expensive even in computer simulation. To address such a challenge, approximation or surrogate based techniques are adopted to reduce the computational cost. The surrogate modeling developed as surrogates of the expensive simulation process in order to improve the overall computation efficiency in multi-objective optimization problem. By using surrogate modeling, the location of the actual Pareto-front is predicted by Radial Basis Function Neural Network (RBFNN) using only a small fraction of the design space. Some case studies show that the surrogate modeling manages to predict most of the Pareto-front of the design space. The best compromise of ISE obtained from predicted Pareto-front produces optimum response for MIMO control system. The result indicates that the procedure to construct the ‘model of the model’ totally compensates the computational expense. This thesis also demonstrates that there are a number of techniques which can be used to tackle difficult multi-objective problems.