Advanced Remaining Useful Life of Lithium-Ion Batteries Prediction with Positive and Negative Perceptron and Convolution Neural Network

Lithium-ion batteries (LIBs) play a significant role in various applications such as electric vehicles, portable electronic devices, and renewable energy storage systems. Accurately predicting the remaining useful life (RUL) of LIBs is essential for enhancing the reliability and efficiency of battery management systems and establishing optimal maintenance plans. In this study, we propose the deep learning network with positive and negative perceptron and convolution neural network (DNPC) that combines the positive and negative perceptron (PNC) model and the convolution neural network (CNN) model, which have high prediction accuracy for the RUL of LIBs. Experimental results using the LIBs dataset show that the root mean square error (RMSE) and mean absolute percentage error (MAPE) for the RUL prediction using test dataset of DNPC are 98.86 [cycle] and 7.25%, respectively, which are lower compared to other deep learning networks (DNNs). Both RMSE and MAPE of the DNPC are also lower compared to other studies using the same LIBs dataset. When analyzing the combination results of the PNC model with other models, the DNN combining the PNC model with the LSTM model has higher RMSE and MAPE, whereas the DNN using the PNC model alone has lower RMSE and MAPE. This means that while the PNC model is suitable for prediction RUL of LIBs, its effectiveness may decrease when combined with recurrent neural network (RNN)-based models. Future research aims to develop a new deep learning model that can enhance synergy with the PNC model and validate their usefulness across various battery types and industrial fields.