A Novel Dual FedAux-based Harvested Energy Prediction for Road Safety Systems

In the evolving landscape of road safety, the role of solar-powered Internet-of-Things (IoT) devices, crucial for supporting connected and autonomous vehicles (CAVs) with enhanced computing and storage capabilities, is increasingly significant These devices, often deployed in remote areas such as transcontinental highways, rely heavily on their energy harvesting capabilities for optimal operation. Given their expected widespread deployment, this paper explores the challenge of accurately predicting harvested energy in such IoT devices. To do so, we introduce "Dual FedAux", a novel approach that blends Federated Learning with a dual-model architecture, integrating long short-term memory (LSTM) and Transformer models. This combination, which is enhanced with meta-learning, i.e., stacking ensemble learning, is particularly adapted for time series analysis, such as for energy prediction. Through experiments, we demonstrate that the proposed Dual FedAux significantly outperforms benchmark models in terms of root-mean-square error (RMSE) and convergence, thus underscoring its potential to enhance the efficiency and reliability of large-scale road safety IoT-based systems.