Improvement Of Biogeography-based Optimization Algorithm And Its Application To Clustering Optimization Problem

Optimization problems are everywhere and are closely related to people's lives.In order to deal with the optimization problem efficiently,the swarm intelligence optimization algorithm comes into being.Biogeography-Based Optimization(BBO)is one of swarm intelligence optimization algorithm.It mimics the species' migration behavior among different habitats and the living environment mutation.BBO is simple and easy to implement,so it has drawn much attention,and applied in many fields such as data mining,image processing and mechanical design.However,with the development of society and science technology,the optimization problems in science and engineering fields are more and more complex.At present,the performance of BBO still has a lot of space for improvement.Clustering optimization is an important branch of data mining.K-means algorithm is a classical clustering algorithm.K-means algorithm is simple,it has good scalability and high efficiency,but it also has some problems.For example,the number of K can not be determined and the algorithm is sensitive to initial values.So some scholars have tried to use the swarm intelligence algorithm to solve the problems of the K-means algorithm.BBO has good performance and wide application.It has the potential to deal with clustering optimization problems better.However,there are few related researches.Therefore,BBO has great research value in the application of K-means clustering optimization.In this paper,the research background and significance of BBO and clustering optimization are introduced.The steps of BBO are described.The main defects of BBO are analyzed.The related studies of BBO are briefly reviewed.In order to further improve BBO and expand its application,aiming at the problems of poor performance,low efficiency and weak universality BBO in dealing with high-dimensional and practical problems,three BBO's variants are proposed and applied to K-means clustering optimization problem.The main work of this paper is summarized as follows:(1)In order to enhance BBO's optimization performance,an improved Biogeography-Based Optimization algorithm with differential migration and global-best mutation(DGBBO)is proposed.The emigration habitats selection method is improved to overcome the defect which may select a worse habitat to share the information to a better habitat.The migration operator is improved to overcome the defect that the searchable locations are limited in the solution space.The mutation is improved to overcome the defect which may destroy the superior solution.Moreover,the computational complexity is reduced from several aspects.At finally,DGBBO is obtained.The computational complexity of DGBBO is analyzed.In the experiments,DGBBO is compared with other state-of-the-art algorithms on 16 benchmark functions.The experimental results show that,DGBBO has better optimization performance.(2)In order to improve BBO's optimization efficiency,an efficient and merged Biogeography-Based Optimization algorithm(EMBBO)is proposed.Firstly,BBO's mutation operator is got rid of to greatly reduce the computational complexity.BBO's migration operator is improved to make up the removed mutation operator and enhance the local search ability.Secondly,the single-dimensional and all-dimensional alternating strategy is merged into the improved migration operator to balance exploration and exploitation and reduce the computation complexity further.Thirdly,the opposition-based learning approach is merged into the algorithm to avoid the algorithm falling into the local optima to some degree.At finally,EMBBO is obtained.The stability of EMBBO is analyzed.In the experiments,EMBBO is compared with other state-of-the-art algorithms on 21 benchmark functions and CEC2017 test set.The experimental results show that,EMBBO has higher optimization efficiency.(3)In order to solve the K-means clustering optimization problem better,a hybrid algorithm based on Biogeography-Based Optimization and Grey Wolf Optimizer(HBBOG)is proposed.Firstly,BBO and GWO are improved respectively to enhance their performance.Then,two improved algorithms are hybridized by the single-dimensional and all-dimensional alternating strategy to complement each other's advantages and to overall balance exploration and exploitation.At finally,HBBOG is obtained.The global convergence of HBBOG is analyzed.In the experiments,HBBOG is compared with other competitive algorithms on 30 benchmark functions and 9 clustering datasets.The experimental results verify the universality of HBBOG and show that HBBOG overall performs the best in solving K-means clustering optimization problems.For these three researches,in terms of algorithm design: The first study is an innovative improvement based on BBO.It emphasizes the enhancement of the performance of the algorithm.Based on the first study,the second study borrows some innovations and proposes some new improvements.It not only emphasizes the enhancement of the performance of the algorithm,but also emphasizes the significant reduction in computational complexity,then obtains the high optimization efficiency and operability.Based on the second study,the third study borrows some innovations and proposes some new improvements.In addition to the enhancement of the performance of the algorithm,it is also required to deal with more types of optimization problems,and ultimately achieve strong universality.For these three researches,in terms of experiment design: The first study makes the experiments on a set of commonly used benchmark functions.The second study makes the experiments on more benchmark functions.The third study not only makes the experiments on benchmark functions,but also tests the clustering data set.On the whole,these 3 studies follow the logic relationship from simple improvement to complex improvement of the algorithm,from single improvement study to improved and applied comprehensive study.The latter is more perfect than the former,the performance is more significant,and the experiments are more abundant.It also corresponds to the process of deepening the research in this paper.