Aiming to construct a new regression model for high-order and high-dimensional time series using tensor methods. In recent years, tensor decomposition has become an important tool for data processing and analysis, gaining significant attention. Based on tensor singular value decomposition (T-SVD), a novel low-rank autoregressive model for tensor time series prediction under tensor T-product is proposed. The model parameters are estimated through an alternating minimization algorithm. Numerical experiments demonstrate the model’s superior prediction accuracy. The established model have its effectiveness and practical applicability.