Estimation of time-varying origin-destination distributions with dynamic screen line flows

A variety of methods for dynamic origin-destination (O-D) estimation is available in the transportation literature. Among those, the research direction pursued by Bell and others in the same category offers the best promise for use in practice as it employes only link flow information from existing surveillance systems. Following the same direction, this paper presents an innovative method for estimating the dynamic network O-D matrices with time series of link and screenline flows. The proposed method takes full advantage of available link flow information, and considerably increases the observability of the dynamic interrelations between network O-D patterns and the resulting link flow distributions. Hence, one can estimate the dynamic O-D matrices with much fewer model parameters and less execution burden. With properly selected screenlines and efficient computing algorithms, the proposed model also offers the potential for real-time applications.