TTS-Norm: Forecasting Tensor Time Series via Multi-way Normalization

Tensor time series (TTS) data, a generalization of one-dimensional on high-dimensional space, is ubiquitous in real-world applications. Compared to modeling or multivariate series, which has received much attention and achieved tremendous progress recent years, tensor been paid less effort. However, properly coping with the more challenging task, due its complex inner structure. In this paper, we start by revealing structure TTS data from afn statistical view point. Then, line analysis, perform T ensor ime S eries forecasting via proposed Multi-way Norm alization ( TTS-Norm ), effectively disentangles multiple heterogeneous low-dimensional substructures original Finally, design novel objective function for forecasting, accounting numerical heterogeneity among different subspaces TTS. Extensive experiments two datasets verify superior performance our model.