Detection of Distributed Denial of Charge (DDoC) Attacks Using Deep Neural Networks With Vector Embedding
Ahmed Shafee, Mohamed Mahmoud, Gautam Srivastava, Mostafa M. Fouda, Maazen Alsabaan, Mohamed I. Ibrahem
Abstract
Charging coordination mechanisms have been adopted to avoid overloading the power grid and long waiting times for electric vehicles’ (EVs) drivers at charging stations. However, adversaries could launch distributed denial of charge (DDoC) attacks against charging stations by submitting fake charging requests to reserve charging time slots. The current mechanisms in the literature postulate that the requests reported by the EVs are valid and the detection of the DDoC attacks have not been well investigated yet. In this paper, we first evaluate the ability of DDoC attacks to disrupt the charging coordination mechanisms, and propose detectors to identify the attacks using deep neural networks with vector embedding. The detection methodology is based on capturing any anomalous behavior that deviates from the normal patterns of the station’s charging demand. For training and evaluating our detectors, we build a benign charging demand dataset using real vehicles’ routes and EVs’ technical parameters. After that, we introduce several attacks and use them to generate the malicious dataset. We analyze the dataset and find temporal and spatial correlations that can be exploited to detect DDoC attacks. To accurately detect the attacks, a vector embedding layer is combined with a deep neural network to capture/learn the spatial-temporal correlations within the charging requests. Performance evaluations demonstrate that the proposed detectors have high performance regarding the detection and false alarm rates.