A Self-Organizing Map-Based Adaptive Traffic Light Control System with Reinforcement Learning

Abstract

Increasing urban congestion is a common problem in big cities all over the world. Some Adaptive traffic control (ATC) systems have been proposed to reduce the total travel delay time, which is the main factor of congestion cost. However, previous solutions need a great number of sensors and hardware devices, which are hard to deploy. Fortunately, the advent of advanced Internet of Things (IoT) has made possible more effective and efficient solutions for the congestion issue. Besides, with the availability of the Machine Learning (ML) models, there is hope that the ATC system can be improved with the IoT approach, adopting the ML models. In this paper, we propose a machine learning model for adaptive traffic light control system. First, we assume traffic data is collected from internet-connected IOT sensing devices in vehicles on the roads. Next, the proposed machine learning model receives the data analyzed in cloud, and generates an optimal traffic light period as output. Finally, the optimal traffic light period is transformed to traffic light setting signals to be delivered to the IoT actuating devices in the crossroads. For verifying the proposed model, we build a traffic simulator. For a 24-hour simulated period, the proposed model reduces 55.7% waiting time and 12.76% maximum road occupancy on average, as compared with the Fixed Traffic Light Control System (FTLCS). We also simulate different traffic levels, and our model performs consistently better than the FTLCS in the overall waiting time and maximum road occupancy. The experimental results show that the proposed model is able to alleviate the traffic congestion problem.