Faster estimation of discrete choice models via weighted dataset reduction

Abstract

We develop a novel cell-based two-stage stochastic program to address spatial, dynamic and stochastic features of traffic flow for adaptive signal control. Cell transmission model (CTM) is employed to capture the dynamic feature of traffic flow, with certain CTM cells designated as detector cells to capture real-time spatial queuing effects. We formulate a two-stage stochastic program to address uncertain demand for signal control. In stage 1, a base timing plan (BTP) is determined as the long-term default plan. In stage 2, cycle-based adaptive policies, i.e., green extension/cutoff based on the BTP, are implemented according to the detector cell states. We develop a specialised GA algorithm to search for the optimal BTP and adaptive policies. A case study of Tai Tam reservoir is conducted to elaborate the property of the proposed approach. The adaptive control plan can have 17% delay reduction compared to the optimal fixed-time plan.