CAJ | 학술논문

为弥补传统的基于隐M arkov模型在前提假设上的不足，提出了二阶隐马尔可夫模型。在研究二阶隐马尔可夫模型和凝聚算法在时空序列分析的基础上，提出了一种新的基于 HMM2的时间序列凝聚算法。该算法应用 HMM2对时间序列进行建模，合理考虑了概率和模型历史状态的关联性，按照相异度原则将序列聚成几个类，每个类用模型代表，进而对这些模型训练、合并及迭代得到聚类结果。实验比较了该算法与基于HMM算法的聚类质量，研究了聚类正确率与聚类数、距离正确率与模型距离的关系。结果表明，该算法比传统的基于HMM的聚类算法具有更好的性能。
위미보전통적기우은M arkov모형재전제가설상적불족，제출료이계은마이가부모형。재연구이계은마이가부모형화응취산법재시공서렬분석적기출상，제출료일충신적기우 HMM2적시간서렬응취산법。해산법응용 HMM2대시간서렬진행건모，합리고필료개솔화모형역사상태적관련성，안조상이도원칙장서렬취성궤개류，매개류용모형대표，진이대저사모형훈련、합병급질대득도취류결과。실험비교료해산법여기우HMM산법적취류질량，연구료취류정학솔여취류수、거리정학솔여모형거리적관계。결과표명，해산법비전통적기우HMM적취류산법구유경호적성능。
In this paper ,a second-order hidden markov model(HMM2) is proposed to overcome disadvantage of tradi-tional HMM for premise .A second-order hidden markov model-based agglomerative hierarchical time-series clustering algo-rithm is put forward on the foundation of stydying HMM 2 and agglomerative hierarchical algorithm in the use of time-spatial analysis .In this algorithm ,HMM are built from time-series ,the relationship between the probality and model's historical state is considered reasonably ,and the series are clustered according to the most similarity ,then represent by models ,and then the process of traning and merging and updating initial models is iterated until the final result is obtained .In the experi-ment ,the clustering quality of this algorithm and the algorithm based on HMM ,the relation between correctness rate and the clustering number ,the relation between correctness rate and the model distance are researched .The results show that the algorithm in this paper can achieve better performance than the traditional HMM-based clustering algorithm .