This project develops efficient optimization algorithms for the calibration of stochastic microscopic multi-agent urban traffic models. Efficiency is achieved by exploiting problem structure and by bringing together ideas from two already highly developed fields, the mathematical discipline of simulation-based optimization (SO) and the applied discipline of calibrating traffic simulation models from real sensor data. As part of the research, metamodel SO techniques are designed, efficient metamodels are formulated based on analytical probabilistic traffic models and point selection techniques for small-sample size problems are proposed. Such techniques are suitable to address complex calibration problems within a tight simulation budget. They respond to the needs of transportation simulation users by allowing them to address complex problems in a practical manner.
If successful, the techniques of this project will allow traffic simulators to be more accurately and efficiently calibrated, leading to more reliable results. This is important given that federal, state, regional and local transportation agencies, as well as transit agencies and a variety of transportation consultants develop and rely on microscopic simulation tools to identify network design or traffic management strategies that mitigate congestion as well as its negative economic, environmental and health impacts. This award will be carried out in collaboration with a regional planning agency. This allows us to design techniques informed by the current and future needs of practitioners. Additionally, a large-scale case study of the city of Berlin will be carried out. By testing the performance of these tools on complex large-scale case studies, we will demonstrate the benefits of our proposed approach to transportation practice. This project will engage undergraduate minority students. It will integrate its findings within advanced graduate transportation and operations research subjects.
