Network-Level Adaptive Signal Control Optimization Under a Closed-Loop Online Updating Framework

EasyChair Preprint 14740

Abstract

Adaptive signal control is a flexible solution for network signal control but the tradeoff between computational complexity and solution quality is long-standing, especially for large-scale network. Besides, prediction accuracy of the future traffic state also carries considerable weight in control efficacy. Therefore, this study proposed a centralized cycle-based adaptive traffic light control (CATLC) model to optimize the green time of each intersection on a cyclic basis aiming at network delay minimization under a network signal control framework using a closed online updating strategy. Using the real-time link volume measurement, the cycle-based link turning ratio is predicted through a gated recurrent unit (GRU) deep learning network trained with incremental historical data to update the parameters of the CATLC model. An extended multi-cycle-based adaptive traffic light control (MCATLC) method is further developed based on model predictive control (MPC) strategy, which extends the prediction step to multiple cycles considering traffic progression in the neighborhood of each intersection. Simulation evaluation was conducted through SUMO with a network of 56 intersections in Singapore, and the MCATLC showed an improvement of over 44% and 19% in waiting time, compared with the fixed time scheme and SCATS scheme, respectively, which is promising for real-time control in practice.

Keyphrases: Delay minimization, Model Predictive Control, closed-loop online updating strategy, cycle-based adaptive network signal control, turning ratio prediction

@booklet{EasyChair:14740,
  author    = {Jiarong Yao and Rong Su and Chaopeng Tan},
  title     = {Network-Level Adaptive Signal Control Optimization Under a Closed-Loop Online Updating Framework},
  howpublished = {EasyChair Preprint 14740},
  year      = {EasyChair, 2024}}