Enhanced Dynamic Origin-Destination Matrix Updating with Long-Term Flow Information

Abstract

The problem of updating dynamic origin-destination (O-D) matrices by exploiting a long-term time series of link traffic counts in large-scale transportation networks without the need for surveys or census data is investigated. Different time-recursive mechanisms for analysis of these data to enhance the performance of the models currently used to synthesize within-day dynamic O-D matrices are suggested. The efficiency of the proposed procedure is investigated with respect to different formulations and related solution algorithms (based on entropy maximization and generalized least-squares). The impacts of different assumptions on model performance are also examined. These include the length of the time scale in which the flow information is updated and the selection of the weekday for which the flow information is collected. The results of the statistical analysis and the model performance measures demonstrate that the proposed time-recursive procedure for an information updating period of 2 years can produce an improved prior O-D matrix that may significantly enhance the subsequent updating of dynamic O-D matrices corresponding to a series of days of the week.