A Novel Ensembling of CNN-A-LSTM for IoT Electric Vehicle Charging Stations Based on Intrusion Detection System

• Published in: 2023 International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) pp. 1312 - 1317
• Main Authors: Balakrishna, G., Kumar, Aniruddh, V, Tamilselvan, Younas, Ammar, Kumar, N M G, Rastogi, Ravi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.10.2023
• Subjects: Biological system modeling; Data analysis; Electric Vehicle (EV); Electric vehicle charging; Feature extraction; Intrusion detection; Normalization; Performance evaluation; Principal Component Analysis (PCA); Smart cities
• DOI: 10.1109/ICSSAS57918.2023.10331735

The electric vehicle industry is expanding rapidly. This calls for a reliable setting capable of satisfying clients' needs. The Internet of Things (IoT) ecosystem is responsible for the explosion in EV charging station data. An electric vehicle charging station management system (EVCSMS) is the necessary technology for this. However, there is a growing concern that IoT infrastructure will be the target of a cyberattack. Intrusion detection systems (IDSs) play a crucial role in helping traditional IT networks spot harmful activities. Proposed research indicates that a classifier strategy based on machine learning is the most effective way to detect fraud in an IoT network. The suggested methodology makes use of an authentic IoT dataset collected from actual IoT traffic. There is a comparison of several different types of classification systems. Both binary and multiclass traffic models were successful. When the proposed method is implemented in the IoT-based IDS engine that supports EV charging stations, a significant number of cyberattacks that may otherwise disrupt normal life will be prevented. Preprocessing, feature selection, and model performance evaluation are all parts of the proposed method. The suggested model employs preprocessing via normalization, PCA for feature selection, and a CNN-A-LSTM for assessing model performance. When compared to the CNN and LSTM models, the proposed method performs admirably.