Research On Many-Objective Evolutionary Algorithm Based On Decomposition

Many-objective optimization problems exist in many applications of real life.Multiobjective evolutionary algorithms have been quite successful in solving multi-objective optimization problems with 2 or 3 objective optimization problems,they unfortunately encounter severe difficulties on solving the many objective optimization problem(MaOP)with objective optimization problems exceed 3.In recent years,many-objective optimization has become a really hot issue and difficult problem.It is of great significance and meaningful to design an evolutionary optimization algorithm which can efficiently solve the many-objective optimization problems.Aiming at the difficulties in the current MaOP,two decomposition based evolutionary algorithms(DCDG-EA and MOEA-APS)are proposed in this paper.The specific contents are as follows:(1)The decomposition based convergence-diversity guided evolutionary algorithm for many-objective optimization problems(DCDG-EA).DCDG-EA algorithm aims to obtain a good convergence and diversity of the set of solutions.In DCDG-EA,we propose a new operator selection(CDOS)strategy.In CDOS,an appropriate operator is selected to generate offspring with good convergence and diversity.Moreover,a new individual selection(CDIS)strategy is proposed in environmental selection,it greatly overcomes the inefficiency of Pareto dominance and use two distance metrics represent the convergence and diversity of solutions,to obtain the balance between convergence and diversity of population.The experimental results show that DCDG-EA has great advantages compared with four popular many-objective optimization algorithms.(2)The decomposition based evolutionary algorithm with angle penalty selection strategy for many-objective optimization problems(MOEA-APS).The algorithm can further improve the diversity performance of the DCDG-EA algorithm.In MOEA-APS,we considered the distribution of solutions located around the subregion boundary and proposed an angle penalty selection strategy APS.APS allows us to avoid the similar solutions to be selected simultaneously into the next generation population.Indeed,the devoted APS strategy allows improving further the diversity of the population under the premise of ensuring the convergence.In the specific implementation process of APS,the angle similarity is defined for any two solutions and is used to measure boundary solutions density.Once a good solution x in a subregion is selected,the solutions which are Pareto dominated by x or have large angle similarity with x will be punished.The statistical results clearly highlights that the proposed algorithm MOEA-APS is better or at par with five recent popular many-objective evolutionary algorithms for many objective optimization.