Abstract:

In this paper,a new memetic algorithm named SaS-MA ( Self-Adaptive Subpopulations-based MemeticAlgorithm) is proposed by combining the differential evolution ( DE) with the covariance matrix adaptation evolutionstrategy ( CMAES) .In order to improve the synergy of the global exploration and local exploitation abilities ofSaS-MA,a self-adaptive subpopulation strategy consisting of the deletion mechanism and the addition mechanism isdesigned.Then,a numerical experiment is implemented to compare SaS-MA with four algorithms,namely,DE,LCMAES,PSO and IP-CMAES.Finally,a ten-story two-bay concrete frame is adopted to carry out the displacementperformance optimization,and the results obtained by using SaS-MA are compared with those obtained from theoriginal paper and through four algorithms,namely,DE,CMA,PSO and GA,which are reliable and widely usedat present.Numerical results indicate that the proposed strategy improves the performance of SaS-MA by means ofdynamic deletion and by constructing subpopulations,and that,SaS-MA is efficient and reliable,and it statisticallyoutperforms the former four existing algorithms.The optimization results show that SaS-MA is more efficient in termsof the concrete frame displacement optimization,and that,as compared with the other four algorithms under thesame conditions,SaS-MA is capable of achieving the best solution,and it possesses excellent optimization abilityand preferable convergence.

Key words: differential evolution, covariance matrix adaptation evolution strategy, self-adaptive subpopulations based memetic algorithm, self-adaptive subpopulation strategy, performance optimization