Network Fundamental Diagram based Dynamic Routing in a Clustered Network

Dynamic routing algorithms aim to find the shortest (fastest in most cases) path in a road network prone to time-dependent traffic states. Conventional approaches assume the availability of link-level travel time data. Due to the limited number of sensors in real road networks, for large parts of a road network often no travel time data are available. Link-level travel times are therefore often estimated as constants. Consequently, predicted travel times and routes are not accurate, especially under congested traffic conditions. In this paper, we develop a macroscopic routing algorithm in a clustered network based on loop detector data. Traffic speeds in each cluster are assumed to scale homogeneously and are estimated based on the cluster-specific network fundamental diagrams. A macroscopic routing approach is implemented, which reduces the complexity of finding an optimal path. As a result, missing link-level data are imputed with an expected traffic state in each cluster based on the fundamental diagram. Preprocessed routing information within the clusters and a macroscopic network lead to fast route computations. The approach is evaluated from two sides. Using one month of processed empirical trajectory data collected from a large fleet of vehicles in Munich, our predicted travel times are proved to be more accurate compared to a baseline routing algorithm and a one-cluster (network) method. Re-routing can also be observed from free-flow routes using synthesized trips, showing that our macroscopic routing algorithm is capable of avoiding congested clusters.