点击次数：

论文类型：期刊论文

发表时间：2010-01-01

发表刊物：Journal of Computational Information Systems

收录刊物：EI、Scopus

卷号：6

期号：1

页面范围：197-203

ISSN号：15539105

摘要：HMM(Hidden Markov Model)is widely used as a tool to analyze various types of time sequences. Conventional training approach for HMM, such as BW(Baum-Welch)algorithm can only lead to local optimal solutions. Meanwhile, PSO (Particles Swarm Optimization)is a kind of swarm intelligence algorithm with outstanding characteristic of global optimization. To overcome the drawback of BW, PSO is introduced into DHMM(Discrete HMM)training in this paper. By combining BW and PSO, DHMM parameters are trained and applied in time sequences classification. The hybrid algorithm not only balances well between global exploration and local exploitation but also improves the convergence speed of PSO. Classification application of the hybrid algorithm was summarized. Experimental results on an artificial PHONE dataset show that the proposed method is superior to the BW algorithm and beneficial for improving both average probability and classification accuracy. Copyright ? 2010 Binary Information Press.