A New Possibilistic Clustering Algorithm with Its Application to Fault Diagnosis

In 2010, we proposed the improved unsupervised possibilistic clustering algorithm (IUPC) that can be run as an unsupervised clustering and overcome the weakness of the unsupervised possibilistic clustering algorithm (UPC) that it tends to generate coincident clusters. IUPC inherits the merits of UPC. In the meanwhile, IUPC solves the coincident clusters problem of UPC by limiting the feasible regions of different clusters disjoint, and it also give a more accurate solution since it uses a global optimization technique-differential evolution algorithm (DE) to optimize the proposed model. However, IUPC also has a disadvantage of sensitivity to the initializations because its feasible region of solutions is determined by the fuzzy c-means clustering algorithm, whose clustering result heavily depends on de the initial centers. In this paper, a new clustering algorithm called modified improved unsupervised possibilistic clustering algorithm (MIUPC) is proposed to overcome such problem of IUPC. The proposed algorithm adopts the subtractive clustering algorithm to initialize the cluster centers of IUPC. MIUPC not only inherits the merits of IUPC but also avoids the problem of sensitive to the initializations. The contrast experiments with UPC and IUPC show the effectiveness of MIUPC. The proposed algorithm is also applied to the fault diagnosis, and the results show its better performance.