基于动态编程匹配的聚类序列数据挖掘算法

CLUSTERING SEQUENCE DATA MINING ALGORITHM BASED ON DYNAMIC PROGRAMMING MATCHING

针对共现聚类挖掘算法忽略共现事件的发生顺序和时间间隔,且推理精度受限等问题,提出一种基于动态编程匹配的聚类序列数据挖掘算法。利用贝叶斯推理来推断时间间隔的概率密度函数从而提高对不确定性的鲁棒性,并且同时考虑了空间接近性和时间间隔接近性;利用动态编程匹配的思想来获取事件之间发生的内在关系,从而提高时间间隔概率密度函数的推断准确性。通过使用合成数据进行的实验,验证了该算法在不确定情况下良好的推理精度,并将该算法应用于燃料电池损伤分析中,其能够准确地确定损伤模式,进一步验证了算法的有效性。

Aimed at the problem that the co-occurrence clustering algorithm ignores the occurrence order and time interval of co-occurrence events,and the inference accuracy is limited,a clustering sequence data mining algorithm based on dynamic programming matching is proposed.Bayesian inference was used to infer the probability density function of time interval to improve the robustness to uncertainty,and spatial proximity and time interval proximity were considered at the same time.The idea of dynamic programming matching was used to obtain the internal relationship between events,so as to improve the inference accuracy of time interval probability density function.The experiments using synthetic data show that the proposed method has good reasoning accuracy under uncertainty,and the proposed algorithm is applied to the fuel cell damage analysis,which can accurately determine the damage mode,further verifying the effectiveness of this method.