Reinforcement Learning-Based Low-Delay Data Collection in UAV-Assisted IoT for Secondary Geological Hazard Monitoring

The complexity and suddenness of secondary geological disasters make it difficult to predict the risks in advance. Therefore, implementing rapid and sustainable monitoring of these hazards is crucial. However, due to the variability of geological disasters across regions and their evolution over time, ensuring timely data collection and the ability to respond flexibly to disaster conditions presents a significant challenge. This paper designs a heterogeneous unmanned aerial vehicles (UAV)-assisted Internet of Things (IoT) system architecture for secondary disaster monitoring scenarios, including the follower UAV (FUAV) and the collector UAV (CUAV). Based on this architecture, a reinforcement learning-based low-delay data collection scheme (RLDC) is introduced. The scheme prioritizes the data collection from sensor nodes (SNs) with higher importance and dynamically plans the paths of UAVs to adapt to environmental changes. First, to address the challenge of data delay caused by the inability to precisely locate SNs, a path search algorithm based on fuzzy SN location is proposed. Second, to reduce data transmission delay between the FUAV and the CUAV, a load status update algorithm for FUAV is presented. Lastly, considering the dynamic changes in the importance of SNs, a bidirectional long short-term memory network multi-agent deep deterministic policy gradient path planning algorithm (BMADDPG) is proposed. Extensive experimental results demonstrate that the scheme can adapt to dynamic secondary geological disaster monitoring scenarios and achieve low-delay data collection from SNs.