Research

Enhancing Traffic Control Modeling: A Comparative Study of Model Predictive Control and Store-and-Forward Techniques with Reduced Offline Parameter Computation

Project Details

Abstract

This research project explores advancements in traffic signal control (TSC) through Store-and-Forward (SF) modeling and Model Predictive Control (MPC). SF modeling, known for its simplicity and efficiency, is increasingly used in the domain of TSC despite some trade-offs in accuracy and realistic behavior. This project presents a dual contribution: it introduces an open-source script named Network Model Extractor Script (NMES) to streamline the implementation of various SF models by reducing offline computation needs, and it compares the performance of MPC using different SF extensions. The study highlights that while the SF extension reported a 36.17% improvement, the MPC implementation showed a 16.54% improvement. The NMES significantly decreased the offline computation steps required, by 72.73% for classical SF and over 90% for SF extensions, thus facilitating easier and more efficient SF model implementations.

The methodology of this project involved quantifying offline computations, developing NMES, and implementing both deterministic and probabilistic SF models alongside MPC. The results revealed substantial improvements in efficiency through NMES but little improvement from the MPC implementation of SF. The improvements in MPC were inconsistent, with better results in some network scenarios compared to others. The continued work on this project will focus on addressing the MPC calibration issues, exploring non-convex optimization for probabilistic models, and expanding NMES to support a broader range of SF formulations. This ongoing research aims to enhance the practical applicability of these models in real-world traffic scenarios, ensuring more reliable and optimized traffic control systems.

firas-abou-karnib-01.jpg