Ellipsoids clustering algorithm based on the hierarchical division for WSNs

With the wide application of wireless sensor network, there has been a growing need for the use of clustering algorithms, in order to extract automatically meaningful conclusions or scientific and reasonable inference from large amounts of measured data. The author Masud Moshtaghi etc., proposed Hyperellipsoidal Clustering for Resource-Constrained Environments (HyCARCE), which is based on the Mahalanobis Distance, but the initial cluster number of HyCARCE is larger, which causes the program to run for a longer time. This paper focuses on an improved determination method of initial cluster, and puts forward resource-restricted ellipsoid clustering algorithm, which is based on hierarchical classification. The input space is divided into the cells and the number of points in every cell is calculated. The hierarchical division method is used to divide not empty cells of the data space into dense and sparse mesh regions. The dense mesh regions are the initial cluster of the algorithm. The algorithm effective separates dense and sparse mesh regions, decreasing the initial clustering high-density grid numbers and significantly reducing the running time of the algorithm, without lowering the clustering quality or increasing computational complexity, The paper realizes the algorithm and the simulation results on the actual data set and proves the feasibility of the algorithm. The proposed algorithm can be used for anomaly detection in WSNs.