Proceedings Article | 25 September 2023
Wei He, Haowen Huang, Can Feng, Guosheng Lu, Peigang Han, Zhigang Zhao
KEYWORDS: Data modeling, Education and training, Machine learning, Feature fusion, Artificial neural networks, Performance modeling, Feature extraction, Evolutionary algorithms, Neural networks
Accurately predicting the Remaining Useful Life (RUL) of lithium-ion batteries is of great significance to practical production life. In recent years, the data-driven RUL prediction method based on machine learning has become a popular research topic both domestically and internationally. However, a significant number of research works fail to consider the connection between fore-and-aft data, leading to inadequate accuracy of RUL prediction or excessive model complexity. To address this issue, we propose a hybrid deep learning method, named CNN-GRU-DNN, which combines the Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and deep neural network (DNN). The CNN-GRU-DNN fusion model can deeply explore the high-dimensional features and long-term dependencies in the time series data of lithium-ion batteries, and effectively leverage the connection between fore-and-after data to enhance RUL prediction accuracy. The CNN hyperparameters are obtained through simulation, and five feature parameters with high correlation, including capacity, equal voltage drop discharge time, average discharge temperature, average discharge voltage, etc., are selected as input features of the CNN-GRU-DNN fusion model. Experiments are conducted on the NASA lithium-ion battery aging dataset using three statistical indicators, namely MAE, R², and RMSE, and numerical values are used to evaluate the prediction results. Comparative analysis with other hybrid methods (CNN-GRU), single neural network algorithms (CNN, GRU), and various machine learning (ML) algorithms demonstrates the superiority of the proposed CNN-GRU-DNN fusion model in terms of RUL prediction accuracy. This provides a high reference value for the study of high-accuracy prediction of RUL of lithium-ion batteries. In conclusion, the proposed CNN-GRU-DNN fusion model effectively captures the complex relationship between the high-dimensional features and long-term dependencies of time series data of lithium-ion batteries, and provides an accurate prediction. This study offers valuable insights into the development of high-precision RUL prediction methods using machine learning techniques.
