Neural Network based State of Charge Prediction of Lithium-ion Battery

Abstract

Accurate State of Charge (SoC) prediction is the solution to problems entailing Li-ion batteries, especially in the backdrop of increasing Electric Vehicle (EV) usage globally. The challenges including over/undercharging issues, protection, safety, battery-health and reliable operation of an EV, have paved way for devising accurate estimation models. In this paper, a thorough investigation has been made in selecting the Feed forward Neural Network (FNN) for the prediction of SoC. The network is trained with a particular driving cycle condition under different temperatures and is tested in another driving cycle conditions to prove the efficacy of the proposed FNN. To improve the estimation accuracy, a new current integral feature along with the measured current, voltage and temperature is utilized for the training of the model. The trained FNN is capable enough to predict SoC with high accuracy throughout all temperature range. Also, the model is robust as it is found to be working effectively, even under noise conditions.