Large-scale Agent Data Partitioning Based on DensityRepel-K_medoids

Abstract

Large-scale Agent data partitioning is the premise of parallel distributed computing in the process of ABMS (Agent-based Modeling and Simulation) . Based on the distance-based K_medoids clustering algorithm, this paper proposes an improved K_medoids algorithm (DensityRepel-K_medoids), which is implemented by high-performance programming language X10 and applied to large-scale Agent data partitioning simulation based on distance interaction. The DensityRepel-Kmedoids algorithm first determines the density value and the repulsion value of each Agent data in the Agent data set, and secondly pre-selects the cluster center according to the density value and the repulsion value of each Agent data, and finally uses the pre-selected cluster centers as the initial cluster centers for continuous iterative clustering until convergence. The algorithm avoids the defects of K_means clustering algorithm sensitive to outliers, and avoids the shortcomings of K_medoids clustering algorithm for large-scale data processing. By comparing and analyzing the simulation experiments of Agent data sets of different scales, the algorithm presents a better performance.