eFLEET: a framework in federated learning for enhanced electric transportation

Purpose A fully integrated electric transportation system needs to face some challenges. Therefore, the purpose of this paper is to address this problem by conducting a study of local mobility in the city of Madrid (Spain) with the aim of determining the importance of the vehicle routing problem (VRP) and the need to optimize a set of routes for a fleet of autonomous electric vehicles (EAVs). Design/methodology/approach This study has allowed to propose a framework in federated learning for enhanced transportation (eFLEET) with dynamically implement routing solutions through a federated learning (FL) system that makes decisions based on images using computer vision (CV) algorithms. In addition, an additional layer based on Dag technology provides security and privacy in images to be consumed by the federated model. Findings The use of geodata through images has proven to be more effective in this work. To do this, the authors compared different algorithms of machine learning (ML) and CV to determine the most effective method for calculating the cost of vehicle traffic in the central district of Madrid. The selection of the most suitable algorithm has been through the MCC metric that has been used to compare the types of ResNet algorithms that might provide better results. Finally, a federated model has been developed to speed up the selection of the most suitable vehicle for a given real traffic situation. In addition, it has been integrated with other predictive systems that have been proposed in other studies to validate this approach with data prediction for the first quarter (Q1) of 2024 and with an additional security layer. Research limitations/implications The work is centered in “Madrid Central” district, which has more restrictions than any other area in Madrid. The potential for application and expansion in other districts and cities is enormous. Furthermore, these experiments have been limited to data from the past full year (2023) and its predictions for Q1 2024. Improving the ML model is possible through the addition of more data. Practical implications The potential for planning safe future routes, which can improve the planning and forecasting of traffic changes, is the most significant contribution of this work. This method enables the use of images to assist in decision-making with security and privacy. Furthermore, this eFLEET framework facilitates rapid scalability with a large number of EAVs in a fleet. Social implications Improved route planning has implications for not only the energy efficiency of vehicles in urban environments for but also pollution. This is because it avoids traffic jams. Originality/value The added value of the work is that other applications only provide real-time traffic information for choosing the best route. This framework allows to predict routes based on climatology, calendar, local and pollution, among others. This opens up a range of possibilities for creating applications that use this framework and allow routes to be planned in advance.