A Sliding Window Mann-Kendall Method for Piecewise Linear Representation

Piecewise Linear Representation has been widely used to compress online data which are collected by sensors. However, the current existing methods, including Piecewise Aggregate Approximation and Perceptually Important Points cannot own both advantages of calculation speed and recovery accuracy at the same time. In order to improve the Piecewise Linear Representation performance, this work proposes a Sliding Window Mann-Kendall method. This method is applied to compress four typical data series, and the compression results are compared with the existing methods. Comparing to Piecewise Aggregate Approximation method, this method has better recovery accuracy under the same compressing ratio. Comparing to Perceptually Important Points method, this method has a faster calculating speed and occupies less resources. The parameter setting of the proposed method affects the Piecewise Linear Representation result. The experiments’ results show that the optimal window width fit for different data series are different. And for change points picking, as the optimization threshold increases, the compressing ratio and the recovery accuracy decrease. Copyright © 2014 IFSA Publishing, S. L.