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

Area of Science:

• Cybersecurity
• Electrical Engineering
• Computer Science

Background:

• Electric Vehicle Charging Station (EVCS) security faces increasing cyber threats.
• Traditional Intrusion Detection Systems (IDS) lack the capability to detect novel or complex attacks.
• Existing systems often fail to identify known and undiscovered threats due to reliance on conventional machine learning and feature selection.

Purpose of the Study:

• To propose a Transfer Learning (TL) framework for enhanced cyber-physical attack detection in EVCS.
• To improve the accuracy and scalability of threat detection in EV charging infrastructure.
• To address the limitations of traditional IDS in identifying sophisticated cyber threats.

Main Methods:

• Developed a Transfer Learning (TL) framework utilizing pre-trained Deep Neural Network (DNN) model weights.
• Applied data normalization and min-max scaling techniques for model training.
• Compared the TL model's performance against Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), LSTM-RNN, and Gated Recurrent Unit (GRU) models.
• Evaluated the framework using the CICEVSE2024 (EVSE-A and EVSE-B) datasets.

Main Results:

• The proposed TL model achieved an accuracy of 93% in detecting cyber-physical attacks.
• The TL framework demonstrated superior performance compared to other deep learning models evaluated.
• The study confirmed the effectiveness of pre-trained models in enhancing EVCS security.

Conclusions:

• Transfer Learning (TL) offers a robust solution for improving EVCS cybersecurity.
• The TL model effectively detects malicious attacks, enhancing the overall security of electric vehicle charging infrastructure.
• This approach provides a high degree of symmetry between EVCS security and malicious attack detection.