State of Health Estimation of Lithium-Ion Batteries Based on Stacked-LSTM Transfer Learning With Bayesian Optimization and Multiple Features

Abstract

It is critical to accurately estimate the state of health (SOH) to ensure the safe and efficient operation of lithium-ion batteries. To reduce the training amounts of existing data-driven methods, the transfer learning (TL) method has attracted more attention. However, most previous studies lack validation with different battery types and working conditions. Furthermore, the shared knowledge just relies on raw current and voltage data, resulting in insufficient accuracy. This article proposes a stacked-long short-term memory (LSTM) TL method based on Bayesian optimization (BO-Stacked-LSTM), which integrates multiple features to estimate SOH. By improving the structure of the BO-Stacked-LSTM networks and the fine-tuning strategy of TL, as well as employing a Bayesian optimization (BO) algorithm to optimize hyperparameters, the proposed method can achieve accurate SOH estimation. Experimental results demonstrate that it just requires a small quantity of target dataset to accurately estimate SOH on the target dataset. Furthermore, experiments were performed on three different lithium-ion battery datasets, to validate the effectiveness.