Peak-Hour Subway Passenger Flow Forecasting: A Tensor Based Approach

Passenger flow forecasting plays a critical role in intelligent transportation systems. Existing studies mainly focused on the accuracy of forecasting. In this paper, we propose a tensor-based framework combined with “priori modeling” and “posterior analysis”, which can forecast peak-hour passenger flow with both accuracy and interpretability. This framework consists of three parts: 1) construct 5-way tensor model based priori features of passenger flow (e.g. multiple periodicity, temporal variabilities and spatial information); 2) design a tensor completion algorithm to forecast peak-hour passenger flow; 3) qualitatively and quantitatively assess the reliability and importance of priori features. Our framework is evaluated on a real world large-scale passenger flow dataset in the largest city in China. Experimental results showed that our approach achieves 25%-40% improvement over baselines. We also provided in-depth analysis for traffic meanings of modes of tensor and the rank of importance of priori features.