Abstract: Multivariate time series collected by industrial monitoring systems often have problems such as numerous raw data, repeated segmentation results, redundant intersections and blurry boundaries in the process of using data mining technologies to acquire internal existing unknown patterns, leading to unsatisfied mining results when the dataset involves mutation variables or inferior relevance among the data. To resolve these problems, this article introduces a new multiple time sequence clustering algorithm. Experimental results show that this algorithm can overcome the shortage that the accuracy of clustering is often affected by initial values in the Gath-Geva algorithm. It can exhibit the potential variation of raw data and thus efficiently deal with segmentation in multivariate time series to get ideal mining results.