Estimation of traversal speed on multi-lane urban arterial under non-recurring congestion

This paper presents a data-driven model to estimate the traversal speed of public transport on urban arterials under non-recurring congestion. We group unidirectional links of an arterial and use the concept of Macroscopic Fundamental Diagram to achieve a smooth speed-density relationship along the arterial. The methodology comprises two main steps: first, developing the model and estimating fit parameters using k-means clustering and locally weighted regression to achieve a more flexible model. To do so, a training dataset is obtained from microscopic traffic simulation, which is used to estimate the fit parameters. In the second step, to validate the model, a lane closure scenario is modeled in the simulation and k-nearest neighbor classification is employed to estimate traversal speeds based on flow-density relationship. Comparing the results of the proposed methodology with the speed values derived from a mesoscopic simulation shows that the model outperforms the mesoscopic model in almost all conditions.