Alibaba Group Holding Limited
Change of variance detection in time series data

Abstract:

The present disclosure provides detecting change of variance points of a target time series based on baseline assumptions of statistical stationarity. A target time series signal is modeled as including a trend component signal and a residual component signal. A signal cost function including at least a difference penalty function and at least one regularization term is optimized for the target signal to output a trend component signal thereof. A signal cost function including at least a difference penalty function and at least one regularization term is optimized for the residual component signal to output estimated variance thereof. Both of these cost functions may be optimized by applying an augmented Lagrangian operator. A centered cumulative sum is computed based on a cumulative sum of the estimated variance. The centered cumulative sum is segmented to yield change of variance points thereof. Such methods may provide improved performance over existing methods.