Uma solução para offloading computacional em VANETs baseada na predição do tempo de vida do enlace

• Ano de defesa: 2023
• Autor(a) principal: Rocha, Paulo Henrique Gonçalves
• Orientador(a): Rêgo, Paulo Antonio Leal
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.ufc.br/handle/riufc/74992

Vehicular networks (VANETs) facilitate the deployment of intelligent applications in urban mobility scenarios, such as real-time traffic data dissemination or resource sharing among VANET components. However, the communication time (referred to as link lifetime or TVE) between nodes is typically short due to the dynamic nature of vehicular mobile scenarios. This limited communication duration has an impact on applications and processes within VANETs, including the decision-making process for task offloading. Furthermore, there is limited integration of entities such as pedestrians in existing research on VANETs, particularly in the context of offloading applications. An architectural model is proposed for computational offloading in VANETs, which involves intelligent prediction of the link lifetime to support the decision-making process. Firstly, a machine learning (ML) model is proposed to predict TVE between nodes in VANETs. The aim is to enhance the efficiency of predictions and improve the decision-making process for computational offloading. Several ML models were trained to evaluate the feasibility of predicting TVE in both Highway and Urban scenarios. Subsequently, an offloading decision algorithm is developed to distribute tasks among resource servers from pedestrian devices. The algorithm’s efficiency is evaluated in comparison to the random decision algorithm (FIFO, First in First out), as well as from different perspectives based on the available servers. The results demonstrate that TVE prediction modeling using SVR (Support Vector Regression) is effective, leading to a 50% reduction in task loss rate compared to the traditional approach. The decision algorithm exhibits lower recovery and false negative rates during the offloading process compared to the random approach, with the number of false negatives or local runs being approximately 40% lower.