Abnormal Series Detection Based on Trend Analysis with Point Compression
Li Bi · Chuangan jishu xuebao · 2014
As a special mode of time series,abnormal sequence plays a very important role. But most time series use the distance-based method in similarity measure,ignore the morphological feature of time series itself. Therefore,this paper proposes an abnormal series detection algorithm based on trend contrast,makes use of the bottom-up linear approximation method with combination of important point and the piecewise linearization. And to extract trend features with higher accuracy,the paper fuses the adjacent subsection on the premise of minimizing the two new optimal object function. Moreover,for the purpose of reducing the complexity of the algorithm,time series were collected after deleting redundant point by Douglas-Peucker algorithm,which keeps the whole trend feature of the sequence and to some extent reduces the amount of calculations. The results of simulation demonstrate that the detection algorithm is feasible,and it not only improves the detection accuracy,but also enhances the sequence visualization of the change in trend.