Towards intelligent and sustainable IOT system

The rapid integration of Artificial Intelligence into the Internet of Everything (AIoE) has led to the ubiquitous presence of embedded devices, playing crucial roles in various aspects of our lives. However, the limited battery life of these devices poses a significant challenge as they are expected to deliver an increasing number of services and applications. Consequently, the sustainability of embedded devices has become a paramount concern for both academia and industry. In response, Energy Harvesting (EH) has emerged as a promising solution, enabling devices to harvest energy from the surrounding environment, such as radio frequency and thermal energy, to power them- selves perpetually. While EH has extended device lifetimes, effectively utilizing EH-powered de- vices in large-scale deployments remains a challenge. The transient nature of energy harvesting ne- cessitates that EH devices alternate between active (discharging) and dormant (recharging) states, resulting in frequent interruptions during sensing and communication activities. Traditional strate- gies for sensing, communication, and energy allocation are ill-suited for EH devices, as those tradi- tional strategies assume devices can be activated at any time, contrary to the characteristics of EH devices. Although prior research has focused on EH-aware sensing and communication strategies, most existing studies have predominantly approached optimization from an individual perspective. To address these challenges, this dissertation proposes a sustainable and intelligent IoT system leveraging emerging technologies, such as deep reinforcement learning and compressed sensing. The proposed framework encompasses three key components: 1) A comprehensive sparsity-aware spatiotemporal data sensing framework for EH IoT systems, aiming to optimize data collection by selectively sensing and processing relevant data based on sparsity characteristics with the consid- eration of the intermittency of EH IoT devices. 2) Environment adaptive multi-hop routing and energy allocation for EH IoT networks, considering the intermittent nature of EH devices. This component incorporates dynamic changes in energy harvesting and jointly optimizes routing poli- cies and energy allocation for efficient data routing. 3) A unmanned aerial vehicles (UAVs)-assisted EH IoT framework that assists ground EH devices in completing data collection tasks such as environment monitoring. By leveraging the capabilities of UAVs, this integrated approach enhances data collection efficiency and extends the reach of EH-powered IoT systems. Through these contributions, this dissertation addresses the challenges associated with EH IoT systems, emphasizing energy efficiency, sustainable operation, and intelligent decision-making. The proposed framework integrates deep reinforcement learning and compressed sensing tech- niques, fostering the development of a resilient and efficient IoT ecosystem.