Study Of Multi-objective Evolutionary Algorithm For Adaptive Preference Radius To Divide Region

In real life,many problems are often high complexity,more dimension of the objective,and a nonlinear optimization problem.When this problem can't get user satisfaction solution by traditional way,random search evolutionary algorithm,especially multi-objective evolutionary algorithm,can get satisfactory solution of decision-makers.The principle of evolutionary algorithm is to simulate the evolution rule of natural organisms,"survival of the fittest,the elimination of unsuitable persons" and adopt elitism preserving mechanism to make excellent solutions evolve.Therefore,the multi-objective evolutionary algorithm is getting more and more attention from researchers.With the development of multi-objective evolutionary algorithm,it is found that it usually does not need to get the optimal Pareto surface,but rather requires a solution or a decomposition set on the Pareto optimal surface.Therefore,more and more attention has been paid to the research of preference multi-objective evolutionary algorithms.The preference-based multi-objective evolutionary algorithms are the sort of evolutionary algorithms to assist the decision maker(DM)in finding the interested Pareto optimal solutions.At present,the inappropriate locations of the reference points sometimes seriously impact the convergence performance of the algorithms when the locations of the reference points are used as the preference information during the optimization.Moreover,the size of the preferred region is difficult to control.And the comprehensive performance of the algorithms will degrade in dealing with the many-objective problems.To address the above issues,in this paper,the self-adjustable preference-based radius is calculated to build a new preference relation model,and by dividing region of interest(RoI),a preference-based multi-objective evolutionary algorithm based on the division of RoI is proposed.The proposed algorithm is compared with four reference points based multi-objective evolutionary algorithms(g-NSGA-?,r-NSGA-?,angle-based preference algorithm and MOEA/D-PRE).The results show that,the proposed algorithm has well convergence and diversity,and the DM can control the size of the preferred region,and it has a good convergence in addressing the many-objective problems.