Caraterização de estradas em tempo real através de métodos preditivos

Abstract

The increasing number of people visiting cities has led to a greater number of vehicles traveling through main access roads. Queues and more frequent incidents require wider network monitoring, faster incident detection and quicker response times. This situation happens on Via de Cintura Interna (VCI), a highway that surrounds Porto and Vila Nova de Gaia cities.In order to improve performance and capacity of this road, it is proposed the development of algorithms, methods or frameworks which allow to understand network state, dynamics and places where incidents could occur. After completing this process, it should be possible to forecast network state in real-time within a maximum interval of 30 minutes. The resulting computation should be based on data collected by loop sensors installed on VCI. The final solution will be integrated on a traffic management system currently in use.This dissertation presents some concepts and related work following model-driven and data-driven approaches. Relatively to the first approach, study is focused on mesoscopic simulation which allows representation and simulation of road networks and entities with adequate levels of detail and performance. Relatively to data-driven approach, it is referred the statistical learning concept and some frequently used algorithms. All these concepts supported the development of a new framework and different forecasting algorithms which will be compared and assessed.