Improvement Of Krill Herd Algorithm And Its Application In Data Clustering

Krill herd algorithm(KH)is a biological heuristic algorithm for solving optimization problems,which was proposed by American scholar Gandomi and Iranian scholar Alavi in 2012.As a random search algorithm,KH has been successfully applied to solve many optimization problems in different fields because it contains exploration and exploitation strategies based on foraging and induced motion by other individuals.However,similar to most well-known optimization algorithms,it is always difficult to achieve the best balance between the global exploration ability and the local exploitation ability of the KH algorithm in the optimization process,which leads to the low accuracy of KH.It is easy to fall into the shortcomings of local optimum and low search efficiency.In view of this,on the basis of previous work,this paper improves the standard krill herd algorithm and applies it to solve the problem of data clustering.The main contents are summarized as follows:(1)Krill herd algorithm based on generalized opposition-based learning was proposed and its application in data clustering.In order to solve the problem of premature convergence due to the decrease of population diversity in the optimization process of krill herd algorithm,.a new algorithm based on generalized opposition-based learning was proposed.Firstly,determining step size factor by cosine decreasing strategy to balance the exploration and exploitation ability,then,a generalized opposition-based learning strategy was added to search each krill,which enhanced the ability of the krill to explore the surrounding space.The improved algorithm was tested on 20 benchmark functions and was compared with standard KH,particle swarm optimization and its four improved algorithms.The experimental results show that the improved algorithm can effectively avoid premature convergence and has higher accuracy.In order to demonstrate the effectiveness of the improved algorithm,the improved krill swarm algorithm and the K-means algorithm were combined to solve the data clustering problem that was,the worst individual was replaced by the optimal individual or the new individual after the K-means iteration after each iteration.Five data sets of UCI were used to test and improved algorithm was compared with the K-means and six optimization algorithms for clustering problems,the experimental results show that the fusion algorithm is suitable to solve the data clustering problem with strong global convergence and high stability.(2)A data clustering algorithm based on improved KH and KHM clustering was proposed.In order to solve the problem that K-means clustering is too dependent on random initial clustering center and poor global convergence,a data clustering algorithm based on improved krill herd algorithm and K-Harmonic Means is proposed.Firstly,an improved krill herd algorithm with Levy flight and crossover operator was proposed to improve stagnating local optimum and low search efficiency of krill herd algorithm.That is,after each standard krill herd location updating,a new location updating method is added to further search to improve the search ability of the population,at the same time,Levy flight and crossover operators are used alternately to carry out greedy search for the current population position to enhance the global search ability of the algorithm.The improved algorithm is compared with the related algorithm.The improved algorithm can quickly and effectively search for the optimal solution with higher accuracy under the condition of fewer iterations.Then,the improved krill herd algorithm and the K-harmonic mean clustering algorithm are fused to solve the data clustering problem,that is,the worst individuals are replaced by the optimal individuals or the new individuals after the K-harmonic mean iteration after each iteration.The test results of five real data sets on UCI show that the fused clustering algorithm overcomes the defect that K-means is sensitive to the initial clustering center and has stronger global convergence.