In many real application, the volume of time series data increases seriously. How to store and process data becomes more interesting and challenge things. Effective representations can make storage less, processing more easily. In this paper, we contribute to construct a new piecewise linear approximation algorithm for segmenting online time series with error bound guarantee. To beat our targets, we combine a disconnected segment strategy into Feasible Space Window method, and to test our algorithm, we compare with algorithms that adopts the above strategies on both real and synthetic data sets. The time complexity of our algorithm is O(n) and the number of segments is smaller than FSW algorithm on all tested data sets.