Average density estimation for urban traffic networks: application to the Grenoble network

This paper deals with the problem of the average density estimation in largescale traffic networks, without requiring to know the density of each individual road. This is achieved by the design of a reduced-order open-loop observer. In general, this type of observers require some specific graph properties, but we show that it is possible to find a virtual representation of the network that satisfies these conditions, by diving each road into a number of cells of specific length. The virtual network is shown to provide close approximations to the average density of the real system for large enough networks. Algorithms to efficiently calculate the observer parameters are proposed. This approach is based on the assumption that traffic dynamics are linear, and is applied at first in free-flow regime. Using microscopic simulations and real data we evaluate the observer performance even when congestion is present in the network.