We propose a new way of modelling very large road systems based on new multidimensional multilevel network structure (hypernetworks). This is an holistic method able to capture the multiscale dynamics at all levels from the microlevel of individual vehicles to the macrolevel of flows between cities, regions, countries and continents.

We have implemented and run the TRANSIMS simulation system for the city of Milton Keynes, which has a population of a quarter of million people, is an employment and retail attractor for a large region, and is a commuting centre for London by rail and many other places by road. The TRANSIMS Transportation Analysis Simulation System is able to simulate the emergent traffic dynamics from millions of travellers, based on a synthetic micropopulation modelling individual people and their family structures, where they live, and their activities. From this, origin-destination travel demands are generated using located activity data. Travel times on road links depend on flows, which depend on routes selected which depend on travel times on road links. TRANSIMS breaks this cycle by a series of simulations that assigns vehicles to shortest routes based on expected link travel times, computes their micro-dynamics, and updates the link travel times. At the lowest level, TRANSIMS computes the second-by-second interaction of the vehicles from which the local flow-time relationship emerges. TRANSIMS effectively models to the coevolution between individual route choices, the impact of those individual choices on link and route travel times, and switching between routes as travellers learn from experience and seek shorter routes.

Our motivation for using TRANSIMS was to investigate road traffic as a multilevel phenomenon. TRANSIMS is entirely bottom-up, with all dynamics emerging from the simulated interactions of individual vehicles. We were interested in emergent relationships at higher level that can provide a coherent way of aggregating local bottom-up simulations to give global dynamics. An unexpected finding in our study of Milton Keynes using TRANSIMS was that the emergent dynamics are very volatile, and that the iterated simulations rarely converge to a given steady state. This finding is reinforced by passing comments in the literature, but it seems there has been no systematic study of the phenomenon.

Another unexpected outcome of that research was that the traditional flow-density of road traffic theory does not apply to the majority of road systems. The real curve reflects the subtlety that flow is a measurement at a point while density is a measurement across a distance. Our approach to modelling the microdynamics of road systems is based on a more appropriate density-flow relationship. The model has been tested by simulation at the microlevel and we are working on integrating this into a multilevel model.