Enhancement of vehicular ad hoc networks using machine learning-based prediction methods
- Autores: Leticia Lemus Cárdenas
- Directores de la Tesis: Mónica Aguilar Igartua (dir. tes.), Ahmad Mezher (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2020
- Idioma: español
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Society is aware that the ecological predation of the planet is due to human activity. Therefore, it is necessary to carry out actions to reverse this damage. Besides, the trend of population growth in big cities and the uncontrolled volume of traffic cause serious problems such as traffic delays, traffic jams, increased CO2 emissions, and traffic accidents. In this sense, so-called smart cities are motivated to create a greener and safer environment where both efficient mobility and public services seek to mitigate those problems. These initiatives are supported by smart technologies such as the Internet of Things (IoT) and information and communication technology (ICT), that provide the basis for creating and implementing smart city projects. In this context, modern vehicles today are equipped with a variety of sensors that enable them to detect and share information. This detected information can not only be useful for other vehicles but also to collect relevant data related to traffic management. This data can help to generate smart mobility solutions and help to improve city services. Vehicle ad hoc networks (VANET) enable communication between vehicles (V2I) and also between vehicles and the city's fixed infrastructure (V2I). Also, VANET routing protocols minimize the use of fixed infrastructure as they employ multi-hop V2V communication to reach the road side units (RSUs) of the city.This thesis aims to contribute to the design of VANET routing protocols for urban environments. We have started our research work analyzing some important aspects of the hop-by-hop forwarding routing based on the evaluation of node metrics. We have analyzed different weighting strategies to compute a multimetric score to arrange the candidate nodes to be as the next node to forward the packet. As a result, we have proposed a weighted power mean function (W-PMF) to improve the selection of the best forwarding candidate. We have shown that the best way to combine several metrics is by implementing the geometric mean function. Then, we have improved the selection of forwarding nodes by accurately estimating their current position at the moment of forwarding messages, instead of using the position information received in the last beacon.Nowadays, many applications and services are based on big data because valuable information can be extracted when the data is properly processed. In this sense, historical data about vehicular network conditions can provide us useful information to design new routing strategies based on predictions, A large number of simulations under different representative scenarios has been conducted to the collect the routing metrics and the related binary output (successfully delivered or not at destination). A statistical model to each metric has been obtained and used to design a new routing protocol. We have proposed a probability-based multimetric routing protocol (ProMRP), which selects the candidate nodes based on the highest probability to deliver the packet at the destination. Also, the accurate estimation of the node's position has been included to obtain an enhanced version called (EProMRP).The last contribution is focused on the applications of machine learning techniques to enhance routing decisions. To this end, a new data set has been collected from five routing metrics and have used to generate two different machine learning models: (i) a decision tree and an (ii) artificial neural network. Our proposals are called (i) multimetric predictive ML-based routing protocol (MPML) and (ii) multimetric predictive ANN-based routing protocol (MPANN). Different VANET scenarios and different city maps have been considered to evaluate each proposal. We have also assessed the level of flexibility of our proposals to adapt to new network conditions. For this purpose, MPANN (which has shown to be the best proposal) has been tested to different city scenarios, different from the city map used to train the model
