Transfer learning for securing electric vehicle charging infrastructure from cyber-physical attacks

Electric Vehicle Charging Station (EVCS) security is a growing concern in today's connected world due to the growing complexity and frequency of cyber threats. Traditional Intrusion Detection Systems (IDS) for EV chargers struggle to detect novel or unexpected attacks due to their usage of predetermined signatures and limited detection capabilities. Existing EV charging station security systems are unable to identify many known and undiscovered threats since they primarily rely on feature selection and categorization accuracy. It is common for these systems to be constructed using conventional machine learning algorithms. So many common signs of attacks are ignored. This paper proposes a Transfer learning (TL) framework for cyber-physical attack detection in EVCS in order to overcome these difficulties and improve both accuracy and scalability. The weights preserved from the Deep Neural Network (DNN) model after implementing data normalization and min-max scaling techniques utilized for training are used to initialize a new model termed Transfer Learning. The study also provides a comparison with different DL models such as Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), Long Short-Term Memory-Recurrent Neural Networks (LSTM-RNN), and Gated Recurrent Unit (GRU). The CICEVSE2024 (EVSE-A and EVSE-B) datasets are used to assess the framework, where one dataset is used to train and store weights, and the second is used to evaluate the learned patterns using transfer learning. Several evaluation matrices are used to evaluate the suggested model. The experimental results demonstrate that the TL model attained 93% accuracy. Consequently, the pre-train TL model provides a high degree of symmetry between EVCS security and the detection of malicious attacks.