A Convergence Acceleration Technique for Multiobjective Optimisation


Abstract

A memetic algorithm which addresses the requirement for solutions convergence towards the Pareto front of a multiobjective optimisation problem is discussed. The memetic algorithm is designed by incorporating a Convergence Accelerator Operator (CAO) in existing algorithms for evolutionary multiobjective optimisation. The discussed convergence accelerator works by suggesting improved solutions in objective space and using neural network mapping schemes to predict the corresponding solution points in decision variable space. Two leading multiobjective evolutionary algorithms have been hybridised through introduction of the CAO and tested on a variety of recognised test problems. These test problems consisted of convex, concave and discontinuous test functions, with numbers of objectives ranging from two to eight. In all cases introduction of the CAO led to improved convergence for comparable numbers of function evaluations.