Research On Differential Evolution Algorithm With Fitness Landscape Information

The computational intelligence method is a cluster of algorithms for solving problems that are inspired by the laws of nature.Compared with traditional optimization methods,computational intelligence method is very suitable for solving complex optimization problems with multi-objective,large-scale,constrained,dynamic and uncertain characteristics that are difficult to solve by traditional optimization methods because it do es not need accurate mathematical or logical modeling of the problem.Differential evolution algorithm is an important member of the computational intelligence.It generates an offspring mainly through the interaction between the parents in current population,and then vertically compares the two generations to preserve one of them more suitable for the environment,so as to promote the evolution of the whole population in a promising direction and search for the optimal solution step by step.Because of its easy implementation,simple and efficient,strong robustness and other characteristics,differential evolution algorithm has attracted the attention of many scholars at home and abroad,and has made great progress in theoretical research and application.The main operations of traditional differential evolution algorithms can be summarized as population initialization,population evaluation,mutation,crossover and selection.Among them,the design of mutation,crossover and selection operators will have a great impact on the optimization performance of the algorithm,especially mutation operators,which often show different effects on different optimization problems.However,observation shows that in the original differential evolution algorithm,the pro cess of selecting interactive individuals is accompanied by greater randomness,and the selection of mutation operators lacks scientific guidance,which makes it difficult to guide the population search.In order to solve this problem,experts and scholars in related research fields have proposed many schemes to improve the differential evolution algorithm.Among them,using fitness landscape information to guide population search is a relatively new angle,which has great research space and research value.Based on the above considerations,this paper introduces the concept of fitness landscape,aiming at fully mining population information from the perspective of fitness landscape,and applying this information to the selection of mutation strategies and the selection of individuals in mutation strategies,so as to better guide the population to evolve in a promising direction.Based on the fitness landscape information,four kinds of differential evolution algorithm frameworks are proposed to improve the optimization performance of the differential evolution algorithm.The main work of this paper can be summarized as follows:(1)Aiming at the problem that different mutation strategies have different optimization performances for different fitness landscape problems and the lack of effective guidance information for the selection of mutation strategies,a differential evolution algorithm based on function modal utilization mechanism is proposed.After detecting the approximate fitness landscape of the function to be optimized,different mutation strategies are applied to different fitness landscape modal questions.(2)In order to effectively utilize the information of superior and inferior individuals to guide population evolution and reduce the waste of the number of evaluations,a differential evolution algorithm based on historical information utilization mechanism is proposed.It includes two key operators: Proximity-based Replacement Operator and Negative Direction Operator.The combination of these two operators can speed up the convergence of the algorithm while maintaining the diversity of the population.(3)Since the individuals with different fitness locate in different regions in the search space and exhibit different search behaviors,so the search mechanism of individuals needs to match their search behavior.On this basis,a differential evolution algorithm based on individual search behavior is proposed,which consists of three stages: construction,separation and guidance,and a heuristic rule is designed in each stage.Through these three stages,we can make full use of the difference information between individuals in search behavior to guide the evolution of population.(4)In the mutation strategy guided by the best individual of the current population,the region of the guiding individual is often ignored,which may lead to the algorithm searching for local optimal solution in the hopeless region.In this paper,a differential evolution algorithm based on adaptive multi-crowds learning strategy is proposed.By using the crowding-based guiding mechanism and the crowding-based replacement mechanism,the whole population is divided into several crowds and is used to guide selection of individuals.The exploration and exploitation ability of the algorithm are balanced by diversifying the guiding individuals to guide the mutation process and utilizing the information of promising trial vectors.In summary,in order to solve the problem of lacking effective information to guide the selection of mutation operators and the selection of individuals in mutation operators,this paper studies the promotion mechanism of differential evolution algorithm for solving global optimization problems from four aspects: historical information,individual search behavior difference and and multi-crowds learning by explores and utilizes the fitness landscape information.Through a large number of experimental evaluations,it is verified that the proposed frameworks have good optimization performance for the algorithm,and provide an effective reference for scientific and engineering fields.