微粒群并行聚类在客户细分中的应用

Customer segmentation application of PSO parallel cluster

提出了基于自适应微粒群优化的并行聚类算法,采用了任务分布方案和部分异步并行通信,降低了计算时间。这种并行自适应微粒群算法结合了并行微粒群算法的快速寻优能力和自适应参数动态优化特性,保持了群体多样性从而避免了种群退化。最后将该算法应用于电信客户细分中。实验证明,该算法在并行机群上具有了较好的准确性、加速性和可扩展性。

The paper presented the parallel cluster algorithm of adaptive particle swarm optimization, which adopted task parallelization and partial asynchronous communication to decrease the computing time. The proposed algorithm combined the fast search optimum ability of parallel particle swarm optimization with parameters dynamical optimization property of adaptive. It could maintain the individual diversity and restrain the degenerate phenomenon. Finally, the presented algorithm was used to analyze the telecom customer segmentation. The experiments indicate the presented algorithm on the cluster maintains preferable accuracy, the speed-up and scaled-up.