Efficient Methodology for Benchmarking Dynamic Origin–Destination Demand Estimation Methods

Abstract

One of the key traffic variables required for the ex post and ex ante evaluation of traffic management and policy measures is origin–destination (O-D) demand matrices. Without ground truth O-D information, however, it is difficult, if not impossible, to assess the quality of an O-D estimation method because so many unknowns are involved. One indicator of the quality of an O-D estimation method is the sensitivity of the method to, and its robustness against, random and structural perturbations of the input from a few typical test networks (e.g., data from sensors, prior O-D matrices). In this paper, an assessment methodology is proposed on the basis of the Latin hypercube method, which is an efficient alternative to Monte Carlo sampling and particularly suited for high-dimensional estimation problems. The methodology is demonstrated on a real urban corridor network for a well-known O-D estimation method (the minimum information estimation method) to illustrate the results that can be obtained and how these results can be used to benchmark different O-D estimation methods.