A Data-driven Approach to Estimate the Probability of Pedestrian Flow Congestion at Transportation Bottlenecks

In public gathering places, pedestrian flow congestion may happen in transportation bottlenecks. Limitations exist in the conventional judgment of congestion by considering the crowd density or the walking speed merely. In this paper, a data-driven mathematical approach based on Kernel Density Estimation (KDE) to analyze the probability of pedestrian flow congestion is proposed, which comprehensively considers the walking speed, the crowd density and the flow rate when pedestrians walk towards possible bottlenecks. With the case study of Nanjing metro station during rush hours, the advantages of non-parametric KDE compared with traditional parametric normal distribution estimation are analyzed and the optimal bandwidth for KDE is also discussed. The case study shows that the proposed method can obtain a more reliable quantitative assessment of congestion risk, as it overcomes the limitation of parametric estimation that relies on experience, and also avoids biased assessment of congestion that merely concerns single parameter of pedestrian flow. Finally, an assessment framework for dynamic congestion risk at bottlenecks is suggested. With this framework, the change of congestion situations of a monitored region can be mastered through data-driven approach, and thus the transformation of normal flowing to congestion of the crowd can be quantified through probabilistic analysis.