Multi-Channel Detection for Abrupt Change Based on the Ternary Search Tree and Kolmogorov Statistic Method

To fast detect abrupt change from large-scale time series, we propose an improved method based on the Ternary Search Tree and modified Kolmogorov statistic method (TSTKS, for short). First, two ternary search trees are built by adding a virtual middle branch into existing binary trees; and then the multi-channel detection is implemented from the root to leaf nodes in terms of two search criteria. Simulations show that TSTKS has an encouraging improvement on our previous HWKS method, because of better sensitivity and efficiency than HWKS, especially higher hit rate and accuracy near the middle boundary. Meanwhile, the results of abrupt change analyses on the real Electromyography (EMG) signals in the CAP sleep datasets suggest that the proposed TSTKS is very helpful for distinguishing the different states of sleep disorders, and it is a quite encouraging method for useful information detection from all kinds of large-scale time series.