Sliding Multi-Windows Based Trend Change Detection on Time Series Stream
Xin Zhang · Dianzi xuebao · 2010
Trend change detection has been applied widely to applications of time series stream.For the issue of detecting the change of variable length,a detection approach based on sliding multi-windows is proposed to scales up the sliding window to detect variable change.For the issue of long-term change detection with a memory constraint,a synopsis of incremental PLA is proposed to approximate to the original data,and then the error under the L2 distance is analysis theoretically,by which an amendatory L2 is given to reduce the ratio of true negative.The approach in this paper achieves quite high detection accuracy and efficiency within the extensive experiments.