Research On Interactive Preference-based Multiobjective Evolutionary Algorithm Based On Weight Vector Adjustment Strategy

When multi-objective evolutionary algorithm(MOEA)solves multi-objective ooptimization problem(MOP),the decision maker(DM)emphasizes on each optimization objective differently.The DM is only interested in part of the solution set on the pareto front(PF)sometimes.Too many pareto optimal solutions will increase the decision-making cost of the DM.Meanwhile,when we solve many-objective optimization problem,due to the existence of a large number of non-dominated solutions in the population,the lack of selection pressure of the MOEA leads to a significant decrease in convergence performance.Preference-based MOEA introduces preference information into MOEA to guide the evolution of the population and obtain the optimal solution set that satisfies the preference information of the DM which not only saves computing resources,reduces decision-making burden of the DM,but also improves the optimization performance of the MOEA.In recent years,more and more researchers have devoted themselves to the research of MOEA and have achieved a large number of research achievements.However,there are still some drawbacks which can be described as follow: 1.The overall performance of many preference-based MOEAs are greatly affected by the position of the reference point.2.The convergence and distribution performance of some preference-based MOEAs decline sharply dealing with many-objective optimization problem.3.For different MOPs,the DM has specific setting for the size of the Region of Interest(ROI).However,it is difficult for the DM to control the size of the ROI in many preference-based MOEAs.4.In the process of optimization,as the DM's knowledge of the MOP continues to deepen,the DM may adjust his/her preference information.However,current preference-based MOEAs are generally unable to meet such requirement.To solve above problems,this thesis proposes an interactive Preference-based MOEA based on weight vector adjustment strategy(MOEA/D-WVA).The main contributions of this thesis are as follows: 1.We transform the reference point into a preference vector so that the performance of the preference-based MOEA is no longer affected by the position of the reference point.2.We use the framework of decomposition and discompose the MOP into a set of single-objective optimization problems(SOPs)which will be optimized simultaneously to ensure the performance of the MOEA solving many-objective optimization problem.3.We propose a weight vector adjustment strategy which will move weight vectors that uniformly distributed in the objective space towards somewhere near the preference vector in a specific way so as to guide the population to evolve towards the ROI and finally help us obtain the optimal solution set satisfying the preference information of the DM.The DM can adjust the size of ROI by adaptive parameter ?.4.We propose an interactive decision-making strategy to help the DM adjust his/her preference information in the optimization process to guide population evolve in a better manner.This thesis conducts an experiment by comparing the proposed algorithms with other four advanced Preference-based MOEAs on three kinds of test benchmark sets.The results show that the proposed algorithm can solve the Preference-based MOP well and help the DM obtain the optimal solution set that satisfies his/her preference information.The proposed algorithm is superior to contrastive algorithms in most test cases and shows excellent convergence and distribution.