Clustering dans l’IoT basé machine learning

Abstract

The Internet of Things (IoT) connects billions of heterogeneous electronic devices, enabling and massive data collection. However, the limited me-mory and energy of these devices pose significant challenges, particularly in data management and energy efficiency. Clustering, as a key method for organizing and structuring IoT networks, plays a central role in improving energy efficiency and re-source management. This technique involves grouping devices into clusters and electing a cluster head (CH) for each cluster to reduce costs, extend device life- times, and optimize load balancing within the networks. To date, numerous clustering approaches have been proposed to enhance data collection performance in IoT. Howe- ver, most of these methods focus on partitioning networks with static topologies, limiting their effectiveness in addressing the uncertain and dynamic aspects of IoT This observation motivates the exploration of new intelligent solutions to overcome these li-mitations. This thesis aims to develop innovative clustering solutions leveraging machine lear-ning (ML) for IoT networks, with a particular focus on wireless sensor networks (WSNs),which are at the core of IoT, as well as edge computing, whose computational capabi-lities are often utilized to enhance IoT . The first contribution of this work proposes integrating the U-k-means algorithm, a clustering method capable of automa-tically determining the optimal number of clusters in WSNs, where this parameter is often unknown in most applications. This method is combined with a genetic algorithm (GA) for efficient CH selection. The second contribution introduces a routing mecha-nism optimized with Q-learning, which dynamically selects CHs in WSNs to balance the load while maximizing network performance. Finally, the third contribution presents an advanced optimization method for load balancing in IoT networks managed by edge servers. This method relies on deep learning to predict traffic variations, coupled with a GA for the intelligent distribution of the load across servers.