Abstract: Changes in the value of time series data often represent the occurrence of events. The change query on time series data, that is, to find the subsequences within a certain length that meets a certain threshold of increase or decrease, can mine events and has important practical significance. Existing methods cannot efficiently solve this problem. To this end, a method based on segmentation and constructing a segmentation relationship graph is proposed. Experiments show that this method can still return results within 100 milliseconds under a million-length time series, and the storage overhead of the segmentation relationship graph is also small. For data sets with less fluctuation, the storage size can reach less than 30% of the original data set size. Moreover, two optimization methods are further proposed, which can reduce the storage overhead by about 50% on the original basis, and at the same time do not affect the query efficiency too much.