State of Charge Estimation of Li-Ion Batteries using Artificial Neural Networks

The race towards net zero carbon emissions raises interest in fields such as electric vehicles and renewable energy. The advancement in these areas is closely related to the progress of battery technology. Although ongoing research into new battery chemistries shows promising results, Li-Ion batteries are still considered to be the state of the art, mainly because of their superior specific energy. Due to the high reactivity of lithium, the deployment of battery management systems (BMS) is crucial to ensure the safe and optimal use of Li-Ion cells. A precise state of charge (SOC) estimation is key for such applications. This paper proposes an adaptive solution for state of charge estimation, using a feedforward artificial neural network and machine learning. The training data consists of series of charge and discharge cycles for a Panasonic 18650PF Li-Ion battery, recorded at temperatures between -20 degrees C and 25 degrees C. The model's accuracy is validated using a variety of test datasets over a wide range of ambient temperature. The mean absolute error (MAE) obtained is between 1 % and 2 %, depending on the ambient temperature.