PSO-Enhanced reinforcement learning for resource allocation in LoRaWAN IoT Network slicing

Mardi, F. Z., Hadjadj-Aoul, Y., Bagaa, M. et Benamar, N. (2026). PSO-Enhanced reinforcement learning for resource allocation in LoRaWAN IoT Network slicing. ISSN 1084-8045

Prévisualisation

PDF Disponible sous licence Creative Commons Attribution. Télécharger (3MB) | Prévisualisation

Résumé

Recent years have witnessed drastic growth in the use of wireless sensor networks, mainly due to the increasing adoption of Internet of Things (IoT) applications in various sectors. However, efficiently allocating transmission parameters in these networks, particularly within the LoRaWAN framework, poses a significant challenge due to the diverse requirements of multiple services, each demanding varying levels of bandwidth and reliability. To address this challenge, this paper introduces three innovative resource allocation approaches for LoRaWAN network slicing: Deep Q-Network-3 (DQN-3), Particle Swarm Optimization (PSO), and the hybrid Particle Swarm Optimization–Deep Q-Network (PSO–DQN). These methods aim to dynamically assign transmission parameters, including transmission power (TP), spreading factor (SF), and coding rate (CR), to optimize network performance while meeting the Service Level Agreement (SLA) requirements of the supported applications. The DQN-3 approach focuses solely on learning optimal policies through experience, while the standalone PSO efficiently explores the parameter space with low complexity but lacks adaptability in dynamic environments. However, the proposed PSO–DQN approach combines the exploratory capabilities of PSO with the decision-making strengths of DQN. The PSO facilitates the exploration of the parameters’ configurations, enabling the algorithm to identify optimal resource allocations even in complex and dynamic environments. These configurations are then passed to the DQN, which refines them to enhance efficiency and performance. Our findings reveal that all three approaches significantly improve reliability and energy efficiency while meeting the Service Level Agreement (SLA) requirements for various services in LoRaWAN networks. The combined method, PSO–DQN, surpasses both the DQN-3 and PSO strategies, highlighting the benefits of merging heuristic techniques and deep learning. Our simulation results demonstrate that the proposed framework outperforms existing methods.

Type de document:	Article
Mots-clés libres:	Deep Q-network Internet of things LoRaWAN Network slicing Particle swarm optimization Resource allocation
Date de dépôt:	03 juill. 2026 15:36
Dernière modification:	03 juill. 2026 15:36
Version du document déposé:	Version officielle de l'éditeur
URI:	https://depot-e.uqtr.ca/id/eprint/12961

Actions (administrateurs uniquement)

Éditer la notice