SmartTRO: Optimizing topology robustness for Internet of Things via deep reinforcement learning with graph convolutional networks

The reliability problems caused by random failure or malicious attacks in the Internet of Things (IoT) are becoming increasingly severe, while a highly robust network topology is the basis for highly reliable Quality of Service (QoS). Therefore, improving the robustness of the IoT against cyber-attacks by optimizing the network topology becomes a vital issue. Heuristic algorithms as the mainstream idea to solve the network robustness optimization, but their high computational cost cannot meet the timeliness requirements of real IoT scenarios. This paper proposes a Smart Topology Robustness Optimization (SmartTRO) algorithm based on Deep Reinforcement Learning (DRL). First, we design a rewiring operation as an evolutionary behavior in IoT network topology robustness optimization, which achieves topology optimization at a low cost without changing the degree of all nodes. Then, SmartTRO learns the evolutionary behavior characteristics of IoT network topology by combining Graph Convolutional Network (GCN) and policy network, where the training of neural network parameters is completed by DRL. Experimental results demonstrate that SmartTRO improves the ability of IoT topology to resist cyber-attacks effectively and outperforms the state-of-the-art heuristic algorithm in terms of both topology robustness optimization performance and computational cost.