UAV-Enabled Dynamic Data Collection and Energy Replenishment in Large-Scale IoT Networks

The increasing frequency of forest fires has become a serious threat to both ecological environments and public safety. Recent advances in unmanned aerial vehicle (UAV) technology have provided new opportunities for monitoring field environments, owing to UAVs’ high maneuverability and real-time data collection capabilities. To address the challenges of real-time performance and energy efficiency in forest fire data collection networks, our work integrates UAVs with Internet of Things (IoT) technologies to efficiently collect and process environmental parameters and human activity information. The aim of our study is to develop an instantaneous and energy-efficient forest fire data collection system that rapidly predicts and responds to fire risks by optimizing UAV data acquisition and trajectory planning. First, we propose a model based reinforcement learning framework and a dataset difficulty definition method to refine the training sample distribution. Subsequently, an adaptive learning mechanism is introduced to gradually increase dataset difficulty, thereby accelerating model convergence to better adapt to environmental changes. Furthermore, considering the dual challenges of information freshness and UAV energy constraints, a novel node partitioning strategy is designed to decompose the overall problem into multiple subproblems for cooperative solution. Extensive simulation results demonstrate that the proposed method outperforms existing approaches in terms of real-time capabilities and energy efficiency, validating its effectiveness for forest fire detection and warning systems.