Research On Static And Dynamic Multi-objective Optimization Algorithms Based On Decomposition

Many optimization problems in the real world include multiple conflicting objectives that must be optimized at the same time.Because there are many conflicting goals,there is no solution to optimize all goals.Multi-objective evolutionary algorithms(MOEAs)can approach multiple optimal solutions in a single run.This advantage makes MOEAs a good candidate for solving MOPs.In the past decades,there has been an increasing interest in developing EA or improving its performance,which has contributed a lot to the applicability of MOEAs to MOPs.However,the performance of MOEAs depends largely on the properties of the MOPs discussed,such as static/dynamic optimization environments,simple/complex Pareto frontier features,and low/high dimensions.Different problem attributes may cause different optimization difficulties for MOEAs.For example,dynamic time-varying static MOPs are more challenging for MOEAs.Therefore,it is not easy to further study MOEAs so that they can be widely used to solve MOEAs with various optimization schemes or problem attributes.This dissertation mainly studies static and dynamic multi-objective optimization algorithms.MOEA/D is one of the most efficient and widely studied multi-objective evolutionary algorithms in recent years.Various improved algorithms have obvious advantages in solving various complex multi-objective optimization problems.Based on the research of MOEA/D,this paper improves them and proposes two improved algorithms,which are used to solve static and dynamic multi-objective optimization problems respectively.The main work and innovation of this paper are as follows:1.In order to improve the fast convergence ability of MOEA/D in solving super-multi-objective problems,a locally improved MOEA/D(MOEA/L-MSF)is proposed.Firstly,we synthesize the weight vectors by using the method of Gauss distribution and uniform distribution,which makes the new weight vectors more widely distributed.Then,the multiplicative scaling decomposition method is used to decompose the multi-objective optimization problem effectively.In addition,local optimization is added to the evolutionary operator.The improvement in this paper is very helpful to improve the performance of the algorithm.Experiments show that MOEA/L-MSF is effective in solving hyper-multi-objective problems.2.To solve the dynamic multi-objective optimization problem,a memory-enhanced dynamic multi-objective evolutionary algorithm(dMOEA/D-lp)based on lp decomposition is proposed.Specifically,dMOEA/D-lp decomposes the dynamic multi-objective optimization problem into multiple dynamic scalar optimization sub-problems,and carries out collaborative optimization at the same time,in which the lp decomposition method is used for decomposition.At the same time,the beam memory method based on sub-problems is used to store the optimal solutions of historical environment and reuse them when necessary to respond to environmental changes.The experimental results verify the validity of lp decomposition method in dynamic multi-objective optimization.The proposed dMOEA/D-lp algorithm has better performance than other memory-based dynamic multi-objective algorithms.