Research On Evolutionary Many-objective Optimization Algorithm Based On Nadir Point Search

Many optimization problems involve in optimizting multiple conflicting objectives at the same time,which is usually referred to as multi-objective optimization problems(MOPs)and many-pbjective optimizaiton when the number of objectives exceeds three.Different from a single objective optimization problem that has a unique optimal solution,multi-objective optimization problems exist a set of incomparable solutions,which is called Pareto set and the projection of it on the objective domain is called Pareto front(PF).The population-based evolutionary algorithm is very suitable to address MOPs.This thesis further proposes two kinds of evolutionary many-objective optimization algorithms based on quick search on the nadir point to address many-objective optimization problems.The first algorithm,NSGA-II-BS,as an improved version of NSGA-II,can be divided into two phases.In the first phase,the nadir point is approximated.In the second phase,the objective space is divided into inner and outter space by the approximated nadir point.Only the solutions located in the inner space is further selected by the nondominated sorting and crowding distance in NSGA-II to further improve the convergence for many-objective problems.The proposed algorithm is compared with six state-of-art many-objective optimization algorithms and the experimental results show that NSGA-II-BS is very competitive with the compared algorithms.The second algorithm aims to address many-objective optimization problems with incomplete PFs,by a robust search of nadir point and angle based selection.The proposed algorithm,called MOEA-R&A,is compared with eight state-of-art many-objective optimization algorithms and the experimental results show that MOEA-R&A outperforms other algorithms.