Origin-Destination Trip Table Estimation for Dynamic Traffic Assignment: A Synthesis of the State-of-the-Art
Corresponding Author: Ramachandran Balakrishna, Caliper Corporation
Presented By: Ramachandran Balakrishna, Caliper Corporation
Dynamic origin-destination (OD) trip tables are an essential input to advanced network models and dynamic traffic assignment (DTA). OD demand variables can run into the millions, and typically dominate the model calibration process as flows must be estimated for each feasible OD pair and for each (short) departure time interval. Dynamic OD matrix estimation (ODME) involves the calibration of these numerous OD flows to match field-observed, time-varying traffic data (counts, speeds, travel times, etc.) recorded over short intervals such as 15 or 30 minutes. We present a roadmap of the ODME problem for practitioners by discussing its technical challenges, a typology of existing solution approaches, and an analysis of their strengths and limitations. We base our discussion on a thorough literature review and nearly two decades of our own ODME research and experiments.
Dynamic ODME has rightly received substantial research focus, given its well-documented challenges. First, the evaluation of a given OD solution requires a network model whose outputs are compared to field measurements to compute a performance metric. The metric can then be optimized by modifying the model’s inputs. The model often involves traffic simulation, with its associated challenges of running time, scalability and randomness. Further, the high fidelity of traffic simulation comes at the cost of a non-linear performance measure, a difficult optimization property. Unconventional solution algorithms have thus been evolved to counter local optima and a stochastic objective function. The ODME solution is also typically non-unique: more than one solution may yield the same traffic output. If the network model predicts a lower flow than the measured count on a particular link, it could either be from low demand or from an artificially high demand causing an upstream bottleneck. Successful ODME thus requires additional measurements such as speeds, which can help determine if the low counts correspond to congested or uncongested conditions. We discuss the evolution of various linear and non-linear ODME techniques including least squares, maximum entropy, Kalman filtering, genetic algorithms and stochastic optimization. We draw from several real-world case studies to illustrate the primary findings, concluding with an overview of current research and future directions.