Edge Load Balancing and Offloading Architecture Using Dynamic Graph Neural Networks in Internet of Things

In recent years, the rapid expansion of EdgeInternet of Things (IoT) networks has enabled real-times services in domains such as healthcare, smart cities, and industrial automation. However, the existing Binary LinearWeightJAYA (BLWJAYA) task scheduling algorithm is restricted to static decisions and lacks predictive load balancing solutions which degraded Quality of Service (QoS) respectively. To overcome these limitations, this research presents Edge Load Balancing and Offloading Architecture using Dynamic Graph Neural Networks (ELOAD-GNN); an energy-efficient framework for intelligent task scheduling in dynamic Edge-IoT environments. Initially, the proposed model constructs a dynamic construct graph representation of the network where nodes and edges reflect the real-time availability of resources and costs of communication. Then, it utilizes a Multimodal GNN to capture spatiotemporal variation in load, while a transformer-based task classifier prioritizes tasks based on urgency and resource demand. Offloading decisions are made using Multi-Agent Reinforcement Learning (MARL), and an energy-aware controller enforces Dynamic Voltage and Frequency Scaling (DVFS)-based load balancing. Experimental results demonstrate that ELOAD-GNN achieves in terms of energy consumption (65.23%), resource utilization (96.92%) when compared to BLWJAYA model.