Integrated Multi-Model AI Architecture for Battery Health Monitoring and Maintenance of Electric Vehicle Based on Federated Learning
Jelia Anita, Dedid Cahya Happyanto, Akhmad Hendriawan · 2025
Electric vehicle (EV) adoption faces significant barriers due to limited monitoring capabilities and user concerns about battery life, charging efficiency, and maintenance timing. Current EV monitoring systems provide basic data without predictive insights or integrated analysis. This paper introduces a combined AI system that uses Long Short-Term Memory (LSTM) networks to predict battery health, Convolutional Neural Networks (CNN) to analyse driving habits, and Random Forest algorithms for maintenance predictions, all while keeping user privacy safe through federated learning. The system was validated through comprehensive field testing with 50 EV owners across four geographic regions over six months, covering eight different EV models. Our approach achieved substantial improvements over existing methods: battery life prediction accuracy improved by$\mathbf{8 5 \%}$(mean absolute error of$\mathbf{2. 4 \%}$vs. manufacturer estimate of 5.7 %), charging optimisation reduced charging time by 23.2 % while decreasing battery stress by 30.7 %, and predictive maintenance achieved 92 % accuracy with a 31 -day advance warning compared to 8 days for traditional methods. The federated learning implementation achieved 94.2% of centralised model performance while preserving user privacy and reducing data transmission by$\mathbf{9 8. 7 \%}$. Economic impact analysis demonstrated annual savings of $990.85 per vehicle, representing an$11.1 \times$return on investment. These results address key EV adoption barriers while providing a scalable, privacy-preserving solution for intelligent vehicle monitoring.