基于稳定阈值的吸引子传播算法

针对传统吸引子传播算法(AP)聚类性能受偏向参数影响较大的问题, 提出一种改进的吸引子传播算法, 即基于稳定阈值的吸引子传播聚类算法（STAP）. 该算法通过稳定阈值, 衡量获得真实类数时的收敛状态, 然后捕捉该状态下的偏向参数； 为加快算法的收敛速度, 采用S型函数作为收敛因子调节阻尼系数. 仿真模拟实验结果表明, 与传统吸引子传播聚类算法相比, 基于稳定阈值的吸引子传播聚类算法聚类精度更高, 收敛速度更快.

Stability Threshold-Based Affinity Propagation Algorithm

In view of the performance of traditional affinity propagation algorithm greatly influenced by parameter P, a novel affinity propagation algorithm based on stability threshold was proposed. The improved algorithm can obtain the convergence of the real class number by stabilizing threshold, and then gain the corresponding parameter P. In order to improve the convergence speed, S function as convergence factor was applied to adjust damp parameter. In addition, it was successfully applied to the field of financial evaluationof listed companies. Simulation experimental results show that the improved clustering algorithm could obtain better precision and quicker convergence, and is obviously better than traditional affinity propagation clustering algorithm.