本研究提出一套適用於張量因子模型的有效估計方法，並發展資料驅動架構，透過穩定估計所得的載荷空間，選取張量時間序列之外積展開與內積展開程序中的滯後資訊聚合。針對所提方法，進一步建立其近似誤差的收斂速率。 為評估方法表現，透過廣泛的模擬研究，探討其在不同訊號強度、樣本數與動態結構設定下的估計效果。此外，所提方法亦應用於 2025 年紐約市曼哈頓計程車 小時資料與 2025 年台北市 YouBike 小時資料。數值結果顯示，該方法具有穩定且可靠的表現。;This study proposes an effective estimation scheme for tensor factor models. A data-driven framework is developed to select the lag aggregations in the outer- and inner-product unfolding procedures for tensor time series by stabilizing the corresponding estimated loading spaces. The convergence rate of the approximation error associated with the proposed method is established. Extensive simulation studies are conducted to evaluate its performance under varying signal strengths， sample sizes， and dynamic settings. The proposed method is further applied to two datasets: hourly Yellow Taxi trips within Manhattan， New York City， in 2025 and hourly YouBike trips in Taipei City in 2025. The numerical results demonstrate that the proposed method achieves satisfactory and reliable performance.
