Real-Time Forecasting of Metro Origin-Destination Matrices with High-Order Weighted Dynamic Mode Decomposition

• DOI: 10.1287/trsc.2022.1128
• Publication Date: 2022-02-16T19:46:02Z
• Authors: Zhanhong Cheng; Martin Trépanier; Lijun Sun
• Keywords: Baseline (sea); Dynamic Mode Decomposition; Rank (graph theory); Matrix (chemical analysis)

Forecasting the short-term ridership among origin-destination pairs (OD matrix) of a metro system is crucial in real-time operation. However, this problem notoriously difficult due to high-dimensional, sparse, noisy, and skewed nature OD matrices. This paper proposes High-order Weighted Dynamic Mode Decomposition (HW-DMD) model for matrices forecasting. DMD uses Singular Value (SVD) extract low-rank approximation from data, high-order vector autoregression established To address practical issue that cannot be observed real-time, we use boarding demand replace unavailable Particularly, consider time-evolving feature systems improve forecast by exponentially reducing weights old data. Moreover, develop tailored online update algorithm HW-DMD coefficients daily without storing historical data or retraining. Experiments on large-scale show proposed robust noisy sparse significantly outperforms baseline models forecasting both flow. The also shows consistent accuracy over long time when maintaining an at low costs.