Research And Application Of Fuzzy Clustering Based On Artificial Bee Colony Optimization Algorithm

In the time of big data, there is tremendous irregular data accumulated by many trades. This data contains the potential information and knowledge that can effectively help human compress pattern classification, predict risk trend and make decisions. To better understand and fully use this vast data and mine the potential valuable information, it is urge to present a clustering analysis way to help human explore this data.As an important unsupervised way to analyze and comprehend data in recognition, fuzzy clustering (FC) analysis is more proper to handle incomplete, inaccurate and boundless clustering problems with the introduction of thoughts of fuzzy set and fuzzy maths, which formulates uncertainty between training data and each category. Although FC can exactly describe and represent the clustering problems in real world, it is usually too sensitive to the initial value, easy to stuck in the local optimal value with a poor ability of global optimal value searching, which influences the final result partly. Thus, it is a gradual trend of development of FC research to overcome obstacle of initial value and global optimum searching in FC algorithm with combining the basic FC algorithms and modern intelligent optimization algorithms. In this paper, FC is combined with artificial bee colony algorithm which has unique global optimum searching ability. Primary research content and innovations are shown as follows:At first, the research background and meaning of FC is presented as well as its developing process and research status. The existing primary problems and developing trend are also explicitly analyzed. At the same time, the maths model and basic algorithm of FC are introduced and studied, then, a detailed implement process of solving a general FC problem is presented.Secondly, considering the slow convergence velocity at the end of algorithm and probability of trapping by local optimum solution in basic ABC, ABC based on simulated annealing (SA-ABC) is proposed by introducing the thought of simulated annealing into the process of ABC greed optimum searching. The core thought is to adopt the SA mechanism to accept a poorer new solution by a some certain probability in the process of leader bees and follower bees searching the new neighboring solution, which increases the possibility of obtaining the global optimum solution and avoids getting into the local optimum value to some extent. It is verified that SA-ABC has a better ability of global optimum value searching and convergence in the test of high dimensionality function.Thirdly, a possibilistic fuzzy entropy clustering algorithm(PFECM) is proposed, combining with PFCM, UPC and probability entropy. Meanwhile, the global ABC is introduced to optimize the proposed model Then its effectiveness and feasibility is testified in the UCI standard database. At the end of this chapter, the proposed model is applied to the error detection analysis of voltage transformers, which fully suggests the good clustering performance and practical value of the proposed algorithm.At last, FC based on probability fuzzy entropy of Gaussian kernel method is proposed by introducing Gaussian kernel function into PFECM. The primary idea is to effectively overcome the unstability of classical FC algorithms dealing with high dimensionality, non-linear detachable and non-convex structure data in the way of using Gaussian kernel function, to map the original low dimensional sample data into high dimensional feature space, the purpose of which is to amplify the feature diversity in all dimensionality. Experimental results in artificial dataset and standard testing dataset show the good performance of the proposed algorithm.