Sub Clustering Algorithm In Large Data for Non-Uniformly Sampled Data

In recent years, affinity propagation clustering algorithm with its unique advantage has become focus of research. The difference between this algorithm and some algorithms (for example, k-means algorithms) is that the number of clustering doesn’t have to be determined before clustering, instead of taking all data points as underlying representative points, through information transmission between different data points, to iteratively compute optimal representative point, and finally realize clustering. During simulation study of clustering algorithm, it is found in the Thesis that: the value of initial deflection degree is closely associated with accuracy rate of algorithm and complexity of computation time. After analysis, this issue concerns multi-objective optimization, combined with common solutions addressing such issues, the Thesis establishes optimization model for multiple single objective with limited algorithm time, determines optimal value of initial deviation, finally maximizes accuracy rate of clustering, and at the same time minimizes complexity of time, so as to improve clustering performance of algorithm effectively.