Dynamic congestion pricing for Dublin

Road pricing usually targets delay or revenue. This project treated pricing as an environmental controller: charges rise with modelled NO₂ exposure and influence route choice before core concentrations exceed 12 µg/m³. Dublin supplied a real road network, detector data and a clear test of the trade-off between local exposure, mobility and affordability.

I led development of the custom discrete-time Python engine and geographic-data workflow, turning the team's system logic into a 1,440-step daily simulation. OSMnx supplied road topology, Rustworkx and NetworkX handled repeated routing, and NumPy and SciPy supported demand and spatial exposure. Full-scale runs took 12-14 hours.

I assigned each agent a departure time, vehicle type, origin, destination, willingness to pay and time sensitivity. I built travel demand from Dublin SCATS detector counts and estimated direct NO₂ emissions with COPERT v5 factors and roadside evidence.

Three nested zones charged entry according to local exposure. Agents balanced time against toll cost, while the model assigned route emissions to a spatial grid and decayed them minute by minute.

Dynamic pricing preserved completed trips while cutting core through-traffic by 80.8%. It detoured 619,520 trips, collected 55,525 tolls and generated €385,159 over the modelled day.

Dynamic pricing did not reduce exposure across the entire network. External-boundary trips rose 18.9%, creating new exposure along parts of the cordon. Protecting the core can transfer the burden unless the controller optimizes the surrounding network as well.

The model simplifies atmospheric dispersion, scales demand from relative detector counts, assumes perfect-information routing and uses reduced-scale sensitivity tests. These choices limit direct policy interpretation. A next iteration should add queueing, sensitive locations, price inertia and a network-wide exposure objective.