AI-Driven Modeling for Next-Gen Smart Transportation Systems

Abstract

This research presents a discrete-event simulation formalism for intelligent transport systems, where forecasting models serve as inputs to downstream decision support. The formalism models the infrastructure, demand, vehicles, charging stations, and discrete events for autonomous mobility-on-demand and charging of electric vehicles. The evaluation of the empirics is on trip-demand forecasting under same-city and zero-shot cross-city settings, not including dynamic control, routing, traffic coordination or energy-use optimization. The findings indicate that the single-modal time-series foundation model yields a lower symmetric mean absolute percentage error than classical machine learning, deep learning and multi-modal foundation-model baselines in the metrics reported for the forecasting tasks. The research does not consider alternative designs for transport, controllers for dynamic use, dynamic control, and routing.