Fine-tuned transfer learning and deep gated recurrent unit methods for state-of-health estimation of the whole life-cycle of lithium-ion batteries

Abstract

Effectively estimating the whole-life-cycle state-of-health (SOH) of lithium-ion batteries is crucial to ensure their safety and reliability. To address this need, this paper proposes a method based on fine-tuning transfer learning and deep gate recurrent unit (DGRU) to estimate the SOH of different types of lithium-ion batteries. Firstly, the DGRU model is pre-trained by extracting highly relevant health indicators (HIs) from random known lithium-ion battery dataset to capture the dynamic characteristics of SOH over aging. Then, the fine-tuning pre-trained model is applied to the dataset of other batteries through transfer learning technology, enhancing the generalization capability of the proposed model on different battery types. Experimental results show that compared with baseline methods such as single-layer gate recurrent unit (SGRU), Gaussian process regression (GPR) and convolutional neural network (CNN), the proposed method has better comprehensive performance on different datasets. Specifically, the mean square error (MSE), mean absolute percentage error (MAPE), and the maximum error (MAXE) of the fine-tuning transfer learning-based deep gate recurrent unit (FTDGRU) model on B0006 cell are only 5.52e⁻⁴%, 0.22 %, and 0.0086, respectively, and the MSE, MAPE, and MAXE on CS2_36 cell are only 2.36e⁻⁴%,1.93 % and 0.0918, respectively. This method not only improves the accuracy of SOH estimation, but also demonstrates strong adaptability and versatility, making it widely applicable to various lithium-ion battery application scenarios.