Graph convolutional networks and deep reinforcement learning for intelligent edge routing in IoT environment

Abstract

The rapid growth of the Internet of Things (IoT) has increased the demand for Quality of Service (QoS) in various applications. Intelligent routing algorithms have emerged to meet these high QoS requirements. However, existing algorithms face challenges such as long training time, limited generalization capabilities, and difficulties in handling high-dimensional continuous action spaces, which hinder their ability to achieve optimal routing solutions. To address these challenges, this paper proposes a novel intelligent edge routing optimization (RO) algorithm that integrates node classification (NC) using a graph convolutional network (GCN) with path selection (PS) based on deep reinforcement learning (DRL). This approach aims to intelligently select optimal paths while meeting high QoS requirements in complex, dynamically changing IoT Edge Network Environments (IENEs). The NC module reduces the computational complexity and enhances the generalization capability of the RO algorithm by transforming network topology and link state information into node features, effectively filtering out low-performing nodes. To cope with high-dimensional continuous action spaces and meet QoS requirements, the PS module utilizes the refined network topology and state information from NC to determine optimal routing paths. Simulation results show that the proposed algorithm outperforms state-of-the-art methods in key performance metrics such as average network delay, packet loss rate, and throughput. In addition, it shows significant improvements in convergence speed and generalization ability.