Improved Clustering Methods Based On Swarm Intelligence Theories And Their Applications

In the era of big data,the information explosion has produced huge amounts of data.Cloud computing,internet and mobile internet are in various fields of people's daily life.Internet banking,e-commerce,information technology and other advanced technology generate data and produce billions to trillion bytes of data every day.Data structure becomes more and more complex and data dimension becomes more and more diverse.Deep excavating implicit information in huge amounts of data is being the urgent demand of the era of big data for managers.Clustering is an important method of data mining.The distribution of the data set can be found as a separate tool,also can be used for analyzing the data and other data pretreatment process of the algorithms.But almost all well-known clustering algorithms need to manually set the key parameters and the choices of parameters depend entirely on the researchers' prior experience and subjective judgment.The effects of these parameters on the clustering results are very significant.In the face of large amount of data and complex data sets,people could not be completely accurate to set the global parameters.In statistics,two step clustering,K-means clustering and Hierarchical clustering are commonly used in SPSS statistical analysis software.The researchers use these three clustering method to solve all types of data clustering analysis.In the face of high-dimensional,vast complex high-dimensional data,the calculation can only get a rough calculation results.For managers,accurate data analysis is the fundamental guarantee for right decisions.As to different learning strategies,it is necessary to use different clustering method.There is no clustering method can be used for all types of data analysis.In this paper,I proposed some improved clustering methods and enriched the clustering theory.The density peak clustering method was proposed in Science in 2014.The DPC algorithm is the latest methods with rapid clustering speed,simple calculation,strong scalability and suitable for big data analysis.Therefore according to the density peak clustering method,the parameter dc need to be set artificially is insufficient.So the different strategies of swarm intelligence optimization algorithm are used to calculate the parameters to in density peak clustering method.The main idea of this paper is basing on the characteristics of the latest proposed fruit fly swarm intelligence optimization algorithm and the cuckoo search optimization algorithmto improve the original algorithms.The improved algorithms would be more scientific and robust and then to optimize the parameters in density peak clustering algorithm.In this paper,the main contents and methods are as follows:(1)An improved fruit fly optimization algorithm(FOA)based on knowledge learning was developed and applied it to optimize the cutoff distance of density peak clustering algorithm.Put forward density peak clustering method based on the improved fruit fly optimization algorithm.FOA is simple with high efficiency,application ability and other characteristics with less parameter.It can be done according to the fruit flies individual search process.There is no limit to the function constraints,information sharing,and other fruit flies pass information to each other.But as the population location update using this algorithm is a kind of completely random strategy.This strategy is simple and has great blindness search process,easily trapped in local minima.Aiming at this problem,this paper introduced knowledge learning strategies and improved the global search ability and convergence speed algorithm.When fruit flies optimization algorithm into local extremum,fruit fly population variation scale have different knowledge by learning.By simulation experiments,the improved flies were validated by test function optimization algorithm and optimization ability is stronger.Optimization peak density clustering algorithm,the algorithm for some data sets have shown the stronger ability of clustering.(2)An improved cuckoo optimization algorithm(CS)based on dynamic discovery probability was proposed.The algorithm was applied to optimize the peak density clustering algorithm in the cutoff distance,further cuckoo optimization peak density clustering method is put forward.Cuckoo optimization algorithm has less parameter,good robustness,the advantages of the global search ability is strong,but there are defects such as slow speed optimization algorithm.The detection probability of cuckoo algorithm in this paper,which is associated with the current function value changes,dynamic update the bird's nest,through such improved completely random characteristics of the original algorithm.Cuckoo optimization algorithm can be improved according to the current optimal,worst nest position distance and direction to control the size of the probability.The improved algorithm has faster convergence of optimal speed and more accurate precision.Through simulation experiment,the improved cuckoo was validated by test function optimization algorithm optimization ability is stronger.Optimization peak density clustering algorithm,the algorithm for some data sets show the stronger ability of clustering.(3)The intelligent selection optimization algorithm was proposed.Fruit flies optimization algorithm has stronger local search ability.At the same time the cuckoo search algorithm has stronger global searching ability.Combined with the advantages of both algorithms,introduce collaborative reorganization operator and mutation factor based on chaos theory and initialize the population divided into two populations.Then put forward theimproved fruit flies optimization algorithm.Respectively,using optimization algorithm and improved cuckoo optimization algorithm to implement the double-population synergy evolution strategy and completed in each iteration Make use of collaborative reorganization operator to two populations to introduce the global optimal solution of each other and realize that the real-time information exchange between the populations and the two kinds of algorithm to get the optimal solution.Through roulette way to select the optimal position,and as a cuckoo species of bird's nest during the next iteration and fruit flies population global optimal food source.Thus the improved algorithm achieved cooperative coevolution and made the algorithm had the ability of local and global parallel serial mining exploration.This paper comprehensively improved the precision,convergence speed and global optimization performance of algorithms.Intelligent selection optimization algorithm has better optimization precision and strong ability of clustering.