Font Size: a A A

The Research On Many-objective Optimization Evolutionary Algorithm Based On Improved Diversity Maintenance Mechanism

Posted on:2022-05-13Degree:MasterType:Thesis
Country:ChinaCandidate:J L ZhouFull Text:PDF
GTID:2518306737456554Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
In the real world,many optimization problems involve multiple optimization objectives,and optimization problems with two or three conflicting objectives are called multiobjective optimization problems(MOPs).MOPs with at least four conflicting objectives are called manyobjective optimization problems(Ma OPs).Due to its wide application in the real world,multiobjective optimization research has become a hot topic.Because evolutionary algorithms have simple,intuitive,and population-based characteristics,they are suitable for solving multiobjective optimization problems.A large number of existing multiobjective evolutionary algorithms(MOEAs)all rely on Pareto dominance for convergence.However,most MOEAs based on Pareto dominance suffered severe dimensionality disasters when solving MOPs with more than three objectives.In many-objective optimization problems,this situation is called dominance resistance,and most of the solutions in population become non-dominated by each other.For constrained multiobjective optimization problems,the existing constrained multiobjective optimization evolutionary algorithms(CMOEAs)mainly distinguish candidate solutions based on the degree of constraint violations,and have not fully considered the diversity of infeasible solutions,resulting in limited algorithm performance.In recent years,many studies have been trying to use evolutionary computing technology to solve Ma OPs,and have achieved a series of research results,but there are still a series of problems to be solved.In order to better solve the above problems,this paper mainly has two key ideas: design a diversity maintenance mechanism to remove individuals with poor convergence,so as to solve the problem of dominance relationship failure in many-objective optimization;in constrained many-objective optimization,design a diversity maintenance mechanism for infeasible solutions,guide the population to explore more feasible regions,and hope to solve the constrained many-objective optimization problem with large-scale infeasible regions.The specific method of the first idea is to use the angle-based niche to estimate the density based on the basis of Pareto domination,and enhance convergence with angle selection strategy.This paper refers to the proposed algorithm as NAEA.The specific method of the second idea is based on the decomposed framework,for the infeasible solutions whose constraint violation is greater than the threshold epsilon,the adaptive parameter balances the relationship between the objectives and the constraint violation value,thereby effectively increasing the diversity of infeasible solutions.We named this proposed algorithm as MOEA/D-DMepsilon.NAEA was compared with six state-of-the-art Ma OEAs(Va EA,MOEA/DD,NSGA-III,MOEA/D,MOEA/D-M2 M and Ma OEA/IGD)on the unconstrained test suite DTLZ and WFG.The experimental results verify that the proposed NAEA is very competitive compared with peer many-objective algorithms.MOEA/D-DMepsilon was compared with four state-of-the-art decomposition based constraint algorithms(MOEA/D-CDP,MOEA/D-ACDP,MOEA/DEpsilon,MOEA/D-IEpsilon)on constraint test suites such as C-DTLZ and DC-DTLZ.By comparing the performance evaluation metrics of IGD and HV,the performance of MOEA/DDMepsilon is very competitive with other state-of-the-art algorithms on most test instances.
Keywords/Search Tags:Multiobjective optimization, Evolutionary algorithm, Dominance relationship, Decomposition, Constrained multiobjective
PDF Full Text Request
Related items