EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

820. An active learning-based optimization for road traffic networks with uncertain service

Invited abstract in session MA-55: Transportation Network Modelling and Optimization I, stream Transportation.

Monday, 8:30-10:00
Room: S02 (building: 101)

Authors (first author is the speaker)

1. Suh-Wen Chiou
Information Management, National Dong Hwa University

Abstract

For a road traffic network with uncertain service, a learning-based optimization is presented. In order to appropriately address spatial evolution of traffic congestion inside road links, a time-varying traffic model with uncertain capacity at links downstream is considered. Accounting for road users’ behavior, a learning-based bilevel program can be proposed. A learning-based optimal signal settings can be determined at the upper level via reinforcement learning and users’ equilibrium traffic flow can be decided at the lower level. In order to effectively solve the proposed bilevel program, a learning-based optimization can be decomposed into two sub-problems. A Quasi-Newton update is introduced to find solution with globally asymptotical convergence. In order to ensure feasibility of solution found against high-consequence realization of stochastic capacity, a learning-based robust model is proposed. Numerical experiments are performed at a moderate traffic grid. As compared to conventional approach, obtained results obviously showed that the proposed model can exhibit sufficient gain of achieving effectiveness while attenuating time-varying congestion in the presence of stochastic capacity at links downstream.

Keywords

Status: accepted

Back to the list of papers