Deep Transfer Learning Enabled State of Health Estimation of Lithium-Ion Battery Using Voltage Sample Entropy Under Fast Charging Profiles

Abstract

Accurately monitoring the battery state of health (SOH) is of great significance in the safe, reliable, and efficient operations of battery energy storage systems. With the increasing demands for sophisticated battery control strategies and working conditions, this article proposes deep transfer learning (TL)-enabled SOH estimation techniques using voltage sample entropy under fast charging profiles. First, considering the unstandardized fast charging protocols, sample entropy is employed to quantitatively assess the fluctuation degree of voltage profiles during fast charging control, which is demonstrated as an effective label correlating to battery SOH. Then, a deep recurrent structure using bidirectional long short-term memory with attention mechanism (BiLSTM-Att) is proposed to capture long-term dependence, whereas the attention processes particular information in prolonged sequences by dynamically focusing on different parts. Third, a TL algorithm is proposed to deal with knowledge transfer between different fast charging profiles, which shows the superiority in reducing data requirements and time cost of model training for different knowledge domains. Through a widely used open-source benchmark dataset, the experimental results indicate that the proposed algorithm can reproduce accurate SOH estimation and save almost 64.5% of time cost during knowledge transferring.