Learning Dynamics on Vehicular Networks

Abstract

Traffic congestion is one of the biggest challenges facing modern services, it costs us time, money and even human lives in some extreme cases. In order to solve or rather prevent this problem, researchers must predict its state, and among all the techniques used we have chosen deep learning because studies show its superiority over all traditional methods in most fields. In our thesis, we understand the basics of car traffic and delve into its concepts, characteristics, challenges and methods used to predict its state, to this end, we offer a deep learning architecture. The main problem with the vehicular traffic is its unpredictable variables which are in most cases caused by human behavior and weather changes, so we need to include these factors in our research. In this context, the traffic state of vehicles is defined as the situation of vehicles on the road network, this state could be: free movement, congestion or a state betweenthem. Based on a review of the literature on traffic forecasts, a sequential deep learning architecture was proposed, this model was formed by data sets extracted from a simulated scenario. The results indicate a great capacity for learning the proposed model with low loss values. In our work, we tried to explore different aspects of the problem of traffic congestion, and explored the solutions proposed by the researcher, we found that this field is very large, its problems evolve over time and become more difficult simultaneously with the development of the proposed solutions and theireffectiveness