HGNN: A Hierarchical Graph Neural Network Architecture for Joint Resource Management in Dynamic Wireless Sensor Networks

Abstract

During the flourishing era of the Internet of Things (IoTs), wireless sensor networks (WSNs) have emerged as a critical backbone for sensing, connectivity, and automation in 6G communications. Due to limited energy sources, minimizing power consumption is the primary focus in extending the lifespan of WSNs. Unfortunately, conventional approaches often face difficulties with scalability and computation complexity, thereby making them insufficient for large-scale WSNs. To address these challenges, graph neural networks (GNNs) have gained significant research attention, thanks to their scalability and generalization capabilities. Nonetheless, existing GNN architectures may struggle to effectively capture the hierarchical topology of WSN systems, where interactions between different levels significantly influence overall network performance. To overcome this challenge, this article proposes a novel hierarchical GNN (HGNN) architecture to learn power allocation and sensor-access point (AP) selection policies that minimizes power consumption in hierarchical WSNs (HWSNs). In this architecture, node and edge update mechanisms are designed to reflect the internal structure of WSNs. Besides, the proposed HGNN is guaranteed representational power, ensuring its ability to capture the graph’s information. Numerical results demonstrate the superior performance of the solution produced by the proposed HGNN in reducing power consumption under various network settings. The HGNN can reduce total power consumption by approximately 30% compared with the model-based approaches.