Research On Many-objective Evolutionary Algorithm Based On Decomposition

Multi-objective optimization problems(MOPs)with more than three objectives are called Many-objective optimization problems(MaOPs).Since MOEA/D using decomposition idea was pro-posed,the idea of decomposition has been increasingly used to deal with MaOPs.MOEA/D decomposes an multi-objective problem into a group of single-objective subproblems through a set of reference vectors,and then searches the optimal solution for each subproblem in an efficient and coevolutionary way,and finally obtains a set of solutions with good convergence and diversity.Although MOEA/D has achieved good results when dealing with MOPs with regular Pareto Front(PF),when it encounters MOPs with complex PFs,MOEA/D cannot fit the shape of these PFs well and may lose some diversity.Therefore,based on the framework of MOEA/D,this thesis takes the goal of balancing the convergence and diversity in MaOPs,and combines the weight vector adaptive method to carry out in-depth research.The main research work of this thesis is as follows:1.In order to improve the performance of MOEA/D in MaOPs with complex PFs,a vector adaptive strategy with multi-criteria constraints is integrated into MOEA/D,and the MOEA/D-PAW algorithm is proposed.Firstly,we analyze and summarize the problems existing in the proposed vector adaptive methods,and then introduce a period information to help judge the validity of the weight vector,which can effectively reduce the probability of deleting the ef-fective weight vector.When adding the new weight vectors,a two-stage method is proposed to properly explore the potential regions in the external archive to enhance the diversity of the obtained solution set.Finally,MOEA/D-PAW and its comparison algorithms are tested on the DTLZ1-7 test problem set.The experimental results show that MOEA/D-PAW has better abil-ity to balance the convergence and diversity in tackling MaOPs by comparing with the classical algorithm.2.In order to improve the performance of MOEA/D algorithm when handling MaOPs with regular and irregular PFs,a two-stage algorithm with weight vector adaptive strategy named MaOEA/D-2AV is proposed.In the first stage,a competitive evolutionary operator is used to accelerate the convergence of the population,and the evolutionary degree of the population is determined by the number of convergent subproblems.In the second stage,MaOEA/D-2AV periodically activates the weight vector adaptive strategy in order to adjust the evolutionary di-rection of the algorithm in time to better simulate the complex PFs.To verify the performance of MaOEA/D-2AV in dealing with regular PFs,three MaOEAs are selected as comparison al-gorithm.The results show that MaOEA/D-2AV is better than the comparison algorithm in the overall performance,which indicates that the vector adaptive strategy has no significant effect on the optimization of regular PFs.To verify the performance of MaOEA/D-2AV in dealing with irregular PFs,three MaOEAs with vector adaptive strategies are selected as comparison algorithm.The results show that MaOEA/D-2AV can be well adapted to irregular PFs with up to 20 objectives.The above results show that the proposed algorithm has good performance when dealing with regular and irregular PFs.