Adaptive Differential Evolution Based On Population Feature Information

As a population based global optimization algorithm,Differential Evolutionary(DE)stands out from the evolutionary algorithm family by virtue of its simple structure and efficient performance.The performance of DE is impacted by control parameters and mutation strategies,and adaptive method is best in improving its search performance.However,adaptive methods of both control parameters and mutation strategies have the following problems:(1)The beneficial feature information is not fully mined and utilized,which is contained in the evolution history data of population;(2)The population lack the selection pertinence and flexibility of control parameters and mutation strategies,which adopt the same parameters and mutation strategies in different states.It is well known that population feature information that refers to some mathematical statistic feature information(standard deviation of population individuals and fitness,number of consecutive stagnant iterations)of all individuals in the dimension of decision space,and it can reflect the features of the problem to be solved.The historical information of populations with similar characteristics can effectively refer to the search direction of population,control the search step,adjust the number of population(NP)and mine the accumulated information for improving the search ability of the population.To fully mine and utilize the feature information of population,a population feature information based adaptive DE framework is proposed,named PFI-DE framework for short.Additionally,PFI-DE framework based three adaptive DE variants is designed for further promoting the search performance of population.The main contributions of this paper are as follows.(1)To fully mine and utilize the beneficial feature information that is implied in the evolution history data of population,a population feature information based adaptive DE framework is proposed by this thesis.To fully mine and utilize the beneficial feature information hidden in evolutionary history population data,a population feature information based adaptive DE framework is proposed in this thesis,named PFI-DE framework.The framework mainly consists of two parts: population feature information calculation and population feature information utilization.In PFI-DE,the population feature information computing mechanism is used to mine and storage the feature information of population,and the population feature information utilization mechanism is utilized to allocate proper historically successful information for population.The historically successful information is used to select both mutation strategies and control parameters for evolution search.The feature information will be used to update the population feature information archive,generated by current population.(2)To rationally use the feature information of population for assigning control parameters more pertinently,an adaptive differential evolution based on population feature information(ADE-PFI)is proposed in this thesis,in which European distance method is used to allocate historically successful parameter information for population.The Population selects and CR to search for adaptive adjusting search step and improving the performance of population.ADE-PFI is utilized to optimize 10 benchmark functions of CEC2020 and the parameter estimation for frequency-modulated sound waves,compared with some state-to-the-art algorithms.The experimental results show that ADE-PFI is superior to other advanced algorithms in terms of convergence speed and convergence accuracy.(3)To balance the global exploration and local exploitation of the population more reasonably and effectively,an enhanced adaptive differential evolution based on population feature information(EADE-PFI)is proposed,which uses a similarity calculator to calculate the similarity between the current population and the population historical feature information archive.Then,the historical parameter information with high similarity is allocated to the current population.The current population uses the allocated parameter information to calculate F,CR and NP,and population is updated according to the current NP value.Finally,the population conducts evolutionary search according to the assigned F and CR,and feature information generated by the current population,is used to update the population feature information archive.In optimizing the 50 and 100 dimensional problems of CEC2017,EADE-PFI has achieved better performance than ADE-PFI,and also defeated other advanced algorithms.In order to prove the practicability of EADE-PFI,it is used to optimize the path planning of UAV.The experimental results show that EADE-PFI obtains better planning path than other advanced optimization algorithms.(4)To make rational use of population feature information for allocating control parameters and mutation strategies more effectively,a population feature information based differential evolution with adaptive selection of mutation strategies and control parameters(ASDE-PFI)is proposed in this thesis.In ASDE-PFI,the current population learns its empirical knowledge from the historical population with the highest similarity of population feature,and uses the empirical knowledge to generate the selection rules of mutation strategy and control parameter.Then,individuals choose appropriate mutation strategy and control parameters for evolutionary search according to selection rules so that population can obtain the better convergence ability.In order to verify the performance of ASDE-PFI,it is used to optimize the benchmark function set CEC2017 and the rule-based network intrusion problem,compared with the advanced adaptive algorithm.The results show that ASDE-PFI outperforms other advanced algorithms,and obtains promising convergence accuracy and convergence speed.Meanwhile,ASDE-PFI can effectively and efficiently generate appropriate intrusion detection rules.On the whole,three algorithms fully excavate and use the population feature information to allocate appropriate control parameters and mutation strategies for population at different stages,so that population can gain the promising convergence ability.In this thesis,a large number of experiments are used to verify the effectiveness of the proposed algorithm,which provides a reference for scientific research.Finally,the text is summarized and the future research is showed.