Font Size: a A A

Research On Micro Differential Evolution Algorithm

Posted on:2021-04-09Degree:MasterType:Thesis
Country:ChinaCandidate:X L MiaoFull Text:PDF
GTID:2428330629487227Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
Differential evolution(DE)algorithm is an intelligent optimization algorithm with simple structure and excellent performance.When solving nonlinear complex optimization problems,it has advantages such as good robustness and high solution accuracy.DE has attracted wide interest among scholars.However,most exsting research focuses on developing DE algorithms with conventional population size.The conventional DE algorithm has some limitations,such as it cannot be used in the application scenarios with low memory and high real-time requirements,such as embedded systems.In recent years,Micro-population DE(Micro-DE)algorithm has attracted much attention due to its low memory and timely response.In this thesis,three improved Micro-DE algorithms are developed,from the aspects of control parameter selection and multi-strategies combination.The main research contents are as follows:(1)Micro-DE algorithm based on Gaussian distribution is developed.To handle the difficulty of selecting control parameters,a Micro-DE algorithm based on Gaussian distribution(Micro-GBDE)is designed.Micro-GBDE uses Gaussian distribution to update individual information,so it is not necessary to set the scaling factor F,which avoids the problem of performance degradation caused by improper parameter setting.In addition,the Gaussian distribution in Micro-GBDE is improved,and a tournament based Micro-GBDE algorithm(Micro-TGBDE)is proposed.Finally,Micro-TGBDE is compared with the existing Micro-DE and adaptive DE algorithms on the CEC2014 functions,and the results show that Micro-TGBDE has better optimization performance.(2)Micro-DE algorithm based on the bimodal distribution of control parameters is developed.To solve the problems of premature convergence and stagnation caused by the small population,a bimodal distribution Micro-DE(BiMDE)algorithm is proposed.BiMDE adopts a bimodal distribution strategy based on Cauchy distribution to update the control parameters including F and CR,and F is also vectorized to enhance the population diversity.Finally,BiMDE is compared with the existing Micro-DE and adaptive DE on the CEC2014 functions,the results show that BiMDE has better optimization performance.(3)Micro-DE algorithm based on learning and abandonment strategies is developed.In order to improve the performance of BiMDE,by combining a variety of mutation operators,Micro-DE algorithm based on learning and abandonment strategies(L-BiMDE)is proposed.L-BiMDE adopts three mutation operators and determines the selection probability of the mutation operators by periodic learning,which enhances the search efficiency.Finally,L-BiMDE is compared with the existing Micro-DE and adaptive DE algorithms on the CEC2014 functions,and the results show that the L-BiMDE has better optimization performance.
Keywords/Search Tags:Differential evolution, Small population, Gaussian distribution, Bimodal Cauchy distribution, Learning and abandonment strategies
PDF Full Text Request
Related items