Parallel Algorithm to Detect Structural Changes in Time Series

Abstract

Analysis of long time series is relevant in many current applications. These investigations are usually carried out on multiprocessor computers (supercomputers). However, supercomputer calculations are efficient only if the time-series processing algorithms are sufficiently parallelized. In this article, we propose a parallel algorithm that detects shift points of the time-series mean — a highly important task in many applications. The algorithm breaks the time series into segments and looks for shift points on each segment using a statistical test. The critical values have been calculated for this test. An additional test reduces the number of false detections.