Estimating Battery Temperature in Dynamic Driving Conditions Using Physics Informed Neural Networks

Battery temperature plays a critical role in determining electric vehicles' performance, safety and lifespan. Accurate estimation of battery temperature is essential for improving the monitoring and control of battery thermal management systems. This paper presents a novel approach for estimating battery temperature using physics-informed neural networks (PINNs) in dynamic driving conditions. The proposed approach combines physics-based models with data-driven neural networks to produce a hybrid PINN model capable of accurately and efficiently estimating battery temperature. The PINNs are trained using a dataset of battery temperature measurements against the varying current, voltage, state-of-charge, and time generated from real-world driving condition simulations on MATLAB Simulink using the NEDC Drive Cycle. The proposed approach's effectiveness is shown by differentiating it from the conventional data-driven methods without physics implementation. The results reveal that the proposed approach can achieve significantly higher accuracy and faster computation times than these existing methods with much lesser data. This work concludes that the proposed approach can potentially improve the reliability and safety of electric vehicles by providing more accurate and timely information about battery temperature.