Sistema de semáforo inteligente con aprendizaje automático para una intersección de alto flujo vehicular. Caso de estudio: Cuenca, Ecuador.

Abstract

Vehicle congestion in Cuenca, Ecuador, is a growing issue that demands innovative solutions. This thesis focuses on simulating the implementation of an intelligent traffic light system that uses the Deep Q-network machine learning algorithm to adjust traffic light phases according to real-time traffic flow conditions. To address the limitations of traditional traffic lights, this research aimed to reduce vehicle waiting times at intersections. To achieve this goal, SUMO (Simulation of Urban MObility), a simulation tool, was used to model and analyze the behavior of the vehicular network. Two approaches were evaluated: (i) the incorporation of traffic light phases with prioritized turns, referred to as FN (new phases), and (ii) the implementation of traffic light phases with fixed duration intervals, determined based on real measurements at the intersection, called DFV (fixed variable duration). The results showed a significant reduction in waiting times, with the FN approach achieving an 8.73 % decrease and the DFV approach achieving a 14.10 % decrease compared to the traditional system. The second approach, DFV, proved to be the most efficient, reducing the metric by 5.90 % compared to the FN approach, suggesting its potential for real-world implementation. These findings demonstrated that the intelligent traffic light system not only reduces waiting times but also optimizes traffic flow, promising to improve urban mobility in Cuenca.