Capacity Prediction and Remaining Useful Life Diagnosis of Lithium-ion Batteries Using CNN-LSTM Hybrid Neural Network

To obtain an accurate remaining useful life (RUL) prediction is a key objective of battery prognostics and health management. Traditional prediction methods depend on sufficient prior knowledge of battery degradation mechanism and complicate artificial feature engineering. Artificial features are lack of generalization of model adaptability and may lose significant information to depict the long-term dependencies. To embrace these challenges, we proposed a novel deep-learning enabled approach for battery RUL prediction based on hybrid neural networks. The one-dimensional convolutional neural network (1D CNN) was employed to automatically extract multiple features of the charging curve. Then, the long shortterm memory (LSTM) network was implemented to capture the long-term underlying relationship among the degraded cycles and battery capacity. Experimental results demonstrated that the proposed method could predict battery RUL with a maximum mean error of 15.65 cycles on NASA dataset. Compared with traditional RUL prediction methods, the proposed method can provide a novel and promising end-to-end solution in RUL diagnosis.