在时间序列数据挖掘中,在线检测在时间上存在任意缩放的相似模式是一个具有挑战性的问题.本文对基于模型匹配的分段半马尔可夫模型进行改进,通过引入偏移量分布、振幅差值分布和前项状态,克服该模型参数难以确定、鲁棒性差的缺点.实验表明,改进分段半马尔可夫模型能够快速准确检测出在时间上存在任意缩放的相似模式. In time series data mining, it is a challenging problem to detect similar patterns with arbitrary scaling in time.In this paper, the improved semi-Markov model based on model matching is introduced. By introducing offset distribution, amplitude The distribution of the difference and the state of the previous item are overcome, the parameters of the model are difficult to be determined and the robustness is poor.The experiments show that the improved sub-Markov model can quickly and accurately detect the similar pattern with arbitrary scaling in time.