State of health estimation and remaining useful life assessment of lithium-ion batteries: A comparative study

Abstract

Lithium-ion batteries are widely used due to their attractive features. They have emerged as the primary storage system for electric cars, solar power, and marine vehicles. Consequently, their internal health state estimation has attracted extensive attention. An accurate health assessment results in a safer and more efficient battery management system (BMS), which estimates the battery's state and predicts premature failure. Recurrent neural network (RNN) has been extensively utilized to diagnose and prognosis lithium-ion batteries, as it has demonstrated superior performance. Improving the proficiency of Machine Learning (ML) algorithms has always been a subject of research. Among these attempts, various studies have proposed the combination of RNN and the convolutional neural network (CNN). In this paper, the CNN is combined with the long short-term memory (LSTM) network for the state of health estimation and remaining useful life assessment of lithium-ion batteries. To facilitate the features exploitation procedure for the ML algorithm, a K-means clustering algorithm is proposed for data classification. A comparative study of LSTM and the hybrid CNN-LSTM method is conducted to show the superiority of the proposed method.