Application of deep learning for prediction of remaining useful life of li-ion batteries

The Remaining useful life (RUL) estimation of batteries like lithium batteries have been studied using the Convolutional Neural-networks (CNNs) and Necurrent NN (RNN). Concurrently the transformers-based architectures have been built to predict the Natural Language Processing (NLPs). Utilization of transformers in predicting the text especially the RUL estimation has not been attempted. The research proposed is a novel approach that uses deep learning based custom model of six layers that utilizes feed-forward NN with sequential layers. Data is obtained from Hawaii Natural Energy Institute (HNEI), where 14 lithium batteries of type NMC-LCO 18650 are adopted. The Adam optimizer is used for increasing the model performance. For evaluating the model, the RMSE, MAE and R2 scores are used. The model obtained 99.21% accuracy with R2 score and the obtained MAE is with RMSE as. The model is found significant and efficient in estimating the lithium batteries' RUL.