REM: A Multi-Model Fusion Framework for Real-Time Cyberattack Detection in Electric Vehicle Charging Systems

Authors: Son Duong, Secil Ercan, Mahdi Zargayouna
Affiliation: GRETTIA-COSYS, Université Gustave Eiffel, 77454 Marne-la-Vallée, France

The integration of Electric Vehicle Charging Systems (EVCS) into smart grids introduces new cybersecurity vulnerabilities due to insecure communication protocols. Threats such as denial-of-service attacks, malware injection, and spoofing compromise system reliability and safety. We introduce the Resilient Electric Vehicle Charger for Cyberattack Classification Using Multi-Model Fusion (REM), a scalable and real-time framework for detecting cyberattacks. REM processes heterogeneous data sources—hardware performance counters and power consumption metrics—through three stages: (i) data preprocessing, (ii) independent model training for each data type, and (iii) probabilistic fusion of outputs to enhance prediction accuracy. Using the CICEVSE2024 dataset, REM achieves 98.57% accuracy in binary classification and 89.47% in multi-class tasks, with a response time of just 0.26 seconds.

The REM framework consists of three main stages:

1. Preprocessing Stage: Independent processing of Power and HPC data through cleaning, feature selection, and dimensionality reduction
2. Learning Stage: Training dedicated models (Transformer, LSTM, RF, SVM, LR) on each dataset to capture modality-specific attack signatures
3. Prediction Stage: Late fusion of probabilistic outputs using weighted averaging to generate unified classification

REM-framework/
├── data/
│   ├── processed/      # Processed data files
│   ├── raw/           # Raw data files
│   ├── test/          # Test data
│   └── train/         # Training data
├── experiments/       # Experimental results and logs
├── src/
│   ├── fusion_test.py      # Late fusion evaluation
│   ├── hpc_processor.py    # HPC data preprocessing
│   ├── model_builder.py    # Model architectures
│   ├── power_processor.py  # Power data preprocessing
│   └── train_models.py     # Model training pipeline
├── models/            # Saved model checkpoints
├── docs/              # Documentation
├── requirements.txt   # Python dependencies
└── README.md

• Python 3.8 or higher
• CUDA-capable GPU (recommended for deep learning models)
• 16GB RAM minimum

1. Clone the repository

   git clone https://github.com/CongSon01/UGE-Research-Project.git
   cd UGE-Research-Project

2. Create virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Download the CICEVSE2024 dataset

   http://cicresearch.ca/IOTDataset/CICEVSE2024%20Dataset/

Process the Power and HPC datasets independently:

# Preprocess Power consumption data
python src/power_processor.py --input data/raw/power_data.csv --output data/processed/

# Preprocess HPC data
python src/hpc_processor.py --input data/raw/hpc_data.csv --output data/processed/

Train models on the preprocessed data:

# Train all models
python src/train_models.py --config configs/default_config.yaml

# Train specific model type
python src/train_models.py --model transformer --dataset power

Evaluate the late fusion mechanism:

# Run fusion evaluation
python src/fusion_test.py --alpha 0.5 --n_samples 1000

If you use this code in your research, please cite our paper:

@article{duong2025rem,
  title={REM: A Multi-Model Fusion Framework for Real-Time Cyberattack Detection in Electric Vehicle Charging Systems},
  author={Duong Son and Ercan Secil and Zargayouna Mahdi},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  year={2025},
  publisher={IEEE}
}

This project is licensed under the MIT License - see the LICENSE file for details.

Source Code: https://github.com/CongSon01/UGE-Research-Project.git