一种基于工程力学与有限元分析的聚类算法

lustering is an important technique as a branch of data mining which is often comprehensively used in many fields such as market-analysis, forecast of financial investment, satellite or medical image disposal etc. Meanwhile at present most clustering algorithms take the distance or density among data as measure to partition clusters, but distance can’t embody the relationship among more than two objects and it can’t satisfy the requirement in practice. Principals of mechanics are introduced to clustering in this paper which converts the relationship among data to gravitation among points and construct the plane truss by connecting points with elastic pole. The stress and strain of all poles are computed which can be use to get the value and direction of all points. According to principals of minimum potential energy and points-relocation, proper clusters are acquired. Experiments have been done and indicated that the result of this algorithm has great improvements.