부산대학교 도서관

This paper presents an innovative solution for real-time data collection in the Internet of Things (IoT). Our research focuses on optimizing data collection in IoT networks through the development of a reinforcement learning-based algorithm that efficiently structures the network. Specifically, we propose a novel method utilizing Q-learning to optimize routing in IoT by dynamically selecting an optimal set of Cluster Heads (CHs). The reward function is carefully designed to achieve multiple objectives: increasing network throughput, improving CH dispersion, minimizing the energy consumption of network nodes and CH nodes, reducing the standard deviation of energy consumption among CH nodes, mitigating message loss from CH failures, and decreasing transmission delays. These enhancements contribute to the resilience, efficiency, and overall performance of IoT networks.