Explainable Data-Driven Digital Twins for Predicting Battery States in Electric Vehicles

Abstract

Advancements in battery management systems (BMS) involve using digital twins to optimize battery performance in electric vehicles. The state of charge and health estimations are essential for battery efficiency and longevity. Digital twins allow for precise predictions of the state of charge and state of health by simulating battery behavior under different conditions. Using artificial intelligence (AI) in digital twins improves predictive capabilities, as demonstrated through studies employing deep neural networks (DNN) and long short-term memory networks (LSTM). However, incorporating AI presents challenges due to the opaque nature of the models, necessitating the need for explainable artificial intelligence (XAI) and trustworthy digital twin models. This study pioneered XAI methods such as SHapley Additive exPlanations, Local Interpretable Model-agnostic Explanations, and linear regression-based surrogate models to explain the predictions of DNNs and LSTMs in digital twin-supported BMSs. The results reveal that the DNN and LSTM digital twin models are more reliable for state-of-health and state-of-charge estimation due to higher $R^{2}$ scores, lower mean residuals, and better XAI results.

Keywords

• Battery management systems
• digital twins
• artificial intelligence
• XAI
• explainable artificial intelligence
• machine learning