Adaptive Resource Optimization for IoT-Enabled Disaster-Resilient Non-Terrestrial Networks using Deep Reinforcement Learning

Abstract

The increasing deployment of IoT devices across sectors such as agriculture, transportation, and infrastructure has intensified the need for connectivity in remote and non-terrestrial regions. Non-terrestrial networks (NTNs), which include maritime and space platforms, face unique challenges for IoT connectivity, including mobility and weather conditions, which are critical for maintaining quality of service (QoS), especially in disaster management scenarios. The dynamic nature of NTNs makes static resource allocation insufficient, necessitating adaptive strategies to address varying demands and environmental conditions during disaster management. In this paper, we propose an adaptive resource optimization approach for disaster-resilient IoT connectivity in non-terrestrial environments using deep reinforcement learning. Initially, we design the chaotic plum tree (CPT) algorithm for clustering IoT nodes to maximize the number of satisfactory connections, ensuring all nodes meet sustainability requirements in terms of delay and QoS. Additionally, unmanned aerial vehicles (UAVs) are used to provide optimal coverage for IoT nodes in disaster areas, with coverage optimization achieved through the non-linear smooth optimization (NLSO) algorithm. Furthermore, we develop the multi-variable double deep reinforcement learning (MVD-DRL) framework for resource management, which addresses congestion and transmission power of IoT nodes to enhance network performance by maximize successful connections. Simulation results demonstrate that our MVD-DRL approach reduces the average end-to-end delay by 50.24% compared to existing approaches. It also achieves a throughput improvement of 13.01%, an energy consumption efficiency of 68.71%, and an efficiency in the number of successful connections of 17.51% compared to current approaches.