Adaptive Neighborhood Size Adjustment Of MOEA/D

As a classical multiobjective evolutionary algorithm,multiobjective evolutionary algorithm based on decomposition(MOEA/D)translates the multi-objective problem into many single objective subproblems.Dynamically allocates resource strategy for decomposition-based multiobjective algorithm(MOEA/D-DRA)is a widely used variant of MOEA/D by dynamically allocating computation resource.In this thesis,neighborhood adjustment and adaptive operator selection mechanism of MOEA/DRA are considered.Due to the dynamically allocating computation resource,Pareto front of MOEA/D-DRA is not well in some cases.To overcome this shortcoming,an adaptive neighborhood adjustment strategy is designed,and we call it as MOEA/D-ANA.In this strategy,the neighborhood is adjusted dynamically by the density of individuals.If the density is small,some individuals will move to other region,and vice versa.To test the validity,CEC2009 multiobjective benchmarks are employed,and simulation results show it is effective.In order to further improve the performance,an ensemble strategy is proposed including DE/rand/1,DE/rand/2 and CMX crossover operator,with this manner,the individual can choose the proper operator to improve the performance of MOEA/D-ANA.Simulation results show our modification is superior to MOEA/D-DRA,MOEA/D-ANA,MOEA/D-STM and MOEA/D-FRRAMB.Signal reconstruction problem of compressed sensing theory is viewed as two-objective(sparsity and measurement residuals)optimization problem.To test the performance,MOEA/D-EANA is applied to solve it.Simulation results with four random signals show our modification achieves best performance when compared with other three algorithms.