Deep Learning-based Malicious Energy Attack Detection in Sustainable IoT Network

Through the use of renewable energy, sustainable Internet of Things (IoT) network can significantly enhance its sustainability and scalability. However, it faces a unique security challenge known as malicious energy attack (MEA), which compromises information security by selectively charging nodes to manipulate the routing path in the network. To efficiently counter MEA, we introduce a two-stage deep learning framework to accurately detect the presence of MEA. It is composed of a stacked residual network (SR-Net) for classification and a stacked LSTM network (SL-Net) for prediction. This model is capable of determining whether an IoT network is under MEA attacks and identifying the affected nodes. Our experimental results verify the efficacy of our proposed model, with the SR-Net demonstrating an average binary cross entropy of less than 0.0590, and the SL-Net showcasing an average mean-square error of approximately 0.0215. These results suggest a high degree of accuracy in detecting MEAs, underscoring the potential of our approach in fortifying the security of sustainable IoT networks.

I. Introduction

The exponential growth of Internet of Things (IoT) presents substantial sustainability challenges, as traditional power solutions like power grids or batteries are not scalable to accommodate the deployment of a vast number of devices over extensive areas. To fulfill these sustainability needs, energy harvesting emerges as a promising solution [1], [2]. It empowers wireless devices by scavenging energy from the environment, such as solar, RF energy, and biothermal energy, allowing nodes to operate semi-perpetually without battery replacements [3], [4]. In this work, we consider the sustainable IoT network where nodes harvest ambient RF energy. Although the energy harvesting feature can greatly enhance the sustainability and scalability of IoT network [5]–[7], it arises a unique security challenge known as malicious energy attack (MEA) [8].