Interpretable Data-Driven Digital Twins For Forecasting Battery Conditions In Electric Vehicles

Abstract

The swift automotive paradigm shift brought on by the rapidly growing electric vehicles (EVs) has made the precise prediction of the battery state paramount in optimizing performance, ensuring safety, and thus, ultimately, prolonging battery life. The paper describes a new technique for predicting battery states in EVs using Explainable Data-Driven Digital Twins. Using deep learning, the model includes the latest and most commonly used techniques such as DNN, LSTM, CNN, SVR, SVM, Feedforward Neural Networks, RBF networks, Random Forest and XGBoost. The aim is to improve predictions of battery-important parameters such as SOC and SOH, for different working scenarios.On the other hand, explainable AI techniques help to identify key factors that affect battery performance. Based on the synergistic effects of these algorithms, the digital twin model surpasses existing ones with respect to predictive accuracy and robustness. This work aims to convince the scientific community about the need for designing intelligent and adaptive battery management systems laying the foundation of tomorrow's sustainable electric mobility.