Research Of Fuzzy Clustering Based On Particle Swarm Optimization Algorithm

In the information age, large amounts of information bring convenience to people, but it also bring about a series of problems. For example, the amout of information is too excessive to grip and digest; the true and false of some of the information is hard to distinguish, which brings difficult to correct use of information; and the inconsistency of information organization forms have increased difficulty to process information effectively and uniformly.At the same time, people realize that the concealed information of these datas which is more deep-seated and more important can describe whole characteristics of datas and forecast development trend, so the information has significant reference value in the decision-making process. In the face of vast databases and mass of miscellaneous information, people cry out to extract knowledge from them, and further improve the utilization of information, so the research about theory and technology of data mining, which is a new study direction has formed. At present, data mining has become a multi-subject crossed research field, it deal with theory and technology of database, artificial intelligence, machine learning, statistics, knowledge-acquisition, biological-computing, and many other cross-sectoral subjects. Clustering analysis is a basic assignment of data mining,it is a course to partition physic or abstract objects into such clusters that objects within a cluster have high similarity in comparison to one another, but are very dissimilar to objects in other clusters. Clustering is a kind of unsupervised learning, and it is a significant technology that has already apply to all study fields of data mining.Conventional clustering analysis is a kind of forcible partition that every undiscriminating object can only be divided into one group or another. Since some things has no specific limits in the real world, the forcible partition is unreasonable and fuzzy clustering is proposed. Fuzzy C-Means cluster algorithm (FCM) is a widespread and sensitive algorithm, it has many advantages such as profound mathematics base, astringency proved, work simply, and operate fast. However, it also has many shortcomings as the sensibility to noisy data, it usually leads to local minimum, and it depends on initial value, especially in the instance of large amounts of clustering objects. To improve the deficiency of clustering algorithm, some scholars have import intelligent algorithms such as genetic algorithm and particle swarm optimization algorithm (PSO) into clustering, that could make up defects of traditional clustering algorithm and have better effect. This paper has used above thinking for reference, and combined PSO which is global optimum and converge fast with FCM, the PSO which is modified could have better effect. Due to the basic PSO could get into local extremum and converge slowly at the end of evolving, it is effective to increase diversity of particles to avoid precocity and not to get into local converging. Accordingly, this paper has modified PSO from two aspects to intensify diversity of particles. One method is to increase mutation based on standard PSO,and inside of the mutation include two different mutation operations with different probability. The other method is to partition the whole particle swarm to two subgroups which have different numbers of particles, each subgroup adopt independent particle swarm to evolve. After improved PSO, the algorithms used PSO instead of iterative course of FCM and used clustering rule function to form fitness function of PSO. In this way, the algorithm has strong ability of global searching, it has amended the flaw of local optimizing of FCM to a great extent, and also, it has reduced the sensitivity to initial value of FCM. According to the idea upwards, this paper proposed two improved algorithms which are fuzzy clustering algorithm based on modified PSO(FCMP) and fuzzy clustering algorithm based on multi-swarms PSO(FCMSP).This paper has applied the modified algorithms to two groups of datas. FCMP work easily and has better effect to the first group of datas, and that FCMSP has better effect to the second group of datas but work complicatedly. The two experiments show that the effect of fuzzy clustering algorithm based on modified PSO is much better compared with fuzzy clustering algorithm based on genetic algorithm and fuzzy clustering algorithm based on basic PSO. They have accelerated convergence rate, advanced work efficiency, and achieved global optimum at the beginning, however, it still has to reduce rate of making mistakes.