The Research Of Resource Allocation Strategy And Hybrid Operator Strategy Based On MOEA/D

Multiobjective optimization problems(MOPs)wildely arise in many fileds,such as engineering design,manufacturing and science research.MOP includes several objectives that should be optimized simultaneously and each objective may conflict with others,so it is impossible to find a single solution that can optimize all objectives at the same time.Therefore,we must consider all objectives and aim to find a set of best trade-off solutions among all the objectives when solving an MOP.Multiobjective evolutionary algorithm(MOEA)is a kind of heuristic search algorithm and it has been found to be an efficient tool for solving MOP.It is designed based on the biological evolutionism.Among all MOEAs,multiobjective evolutionary algorithm based on decomposition(MOEA/D)is an important research filed.It decomposes an MOP into a number of subproblems by utilizing a set of evently distributed weight vectors and each subproblem is optimized in a collaborative way.By performing a detail research and analysis on the MOEA/D variants in this thesis,we find that even though many MOEA/D variants obtain promising results when tackling MOPs,but these algorithms may exist some drawbacks,such as loss of diversity and reduction of search ability,when the problems become more complicated.This makes them not work as well as before.Therefore,we do some analysis on these drawbacks and propose the improved MOEA/D variants in this thesis.First,this thesis proposes a diversity-enchanced resource allocation strategy based on MOEA/D,named MOEA/D-IRA.The new resource allocation strategy introduces a diversity indicator(solution density around each subproblem),to combine with the original convergence indicator(relative improvement of aggregated function value).So it can properly balance the convergence and diversity when assigning computational resources for subproblems.Besides,MOEA/D-IRA uses a new selection strategy to replace the original random selection for neighbor parent.It makes the nearer neighbors have more opportunity to be choosen for evolution,so as to generate more promising solutions.Sencondly,this thesis proposes a hybrid DE strategy in gene level based on MOEA/D,named MOEA/D-GHDE.It designs two operator pools and each one is consist of two different kinds of DE operators,i.e.one DE for convergence and the other DE for diversity.When evolving a solution,the algorithm picks an operator pool and uses its two DE operators to evolve all dimensions of the solution in gene level,which is different from the traditional way by using only one DE in all dimensions of a solution.Also the parameter p used to adjust the hybrid rate of two DE operators in one operator pool is adaptively controlled according to the evolution phase.The proposed algorithm obviously enhances the search ability based on the utilization of hybrid DE operators in gene level.