2026-10-06T21:38:54Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33516952 2026-09-30T04:49:01Z category_43791 portal_791 item_type_3 month_year_09_2026

Lithium-ion Battery State of Health Estimation on the Basis of Impedance Data Using the Collaborative Architecture of Polar-lights-optimizer-based Random Forest and Transfer Learning (Supporting Information) Zhipeng GUO (24757052) Chaoyu WANG (24757054) Qiaozhen JI (24757057) Hongliang HAO (24757059) Jiahao XU (24757068) Xingyu ZHANG (24757069) Qiangqiang LIAO (24757073) Fei WANG (24757076) Chemistry Lithium-ion Battery SOH Estimation Electrochemical Impedance Spectroscopy Transfer Learning <p dir="ltr">Accurately and rapidly estimating the state of health (SOH) of lithium-ion battery is crucial for the safety management of battery systems. A high-precision estimation method for lithium-ion battery SOH is proposed based on electrochemical impedance spectroscopy (EIS). After the characteristic parameters of impedance spectra are analyzed using a suitable equivalent circuit model, two new health factors, namely the resistance capacitance ratio (RCR) and the mid-low-frequency regional line segment (<i>L</i>_AB), are proposed. A mathematical relationship called Bradley model is established to associate RCR with SOH while a cubic polynomial model is employed to describe the relationship between <i>L</i>_AB and SOH. A collaborative architecture of polar-lights-optimizer-based random forest and transfer learning (PLO-RF-TL) is designed to address the issues of the trap of local optimum and feature distribution shifts caused by differences in battery types and environmental temperature, thereby achieving efficient transfer of cross domain knowledge. The results indicate that the two new health factors derived from EIS are negatively correlated with battery SOH for both LiNi_0.83Co_0.11Mn_0.07O₂ (NCM), LiNi_0.86Co_0.11Al_0.03O₂ (NCA) and LiCoO₂ batteries at different temperatures. The PLO-RF-TL model reduces the root mean square error (RMSE) by 67 %–90 % compared to stand-alone polar-lights-optimizer-based random forest (PLO-RF) models. Meanwhile, compared to the DNN-TL model, RMSE has also significantly decreased. In addition, the proposed method was validated using only 10 % of the entire lifecycle data, with an estimated <i>R</i>² of approximately 0.84 or higher for SOH.</p> 2026-09-30T00:00:00Z Dataset Dataset 10.50892/data.electrochemistry.33516952.v1 https://figshare.com/articles/dataset/Lithium-ion_Battery_State_of_Health_Estimation_on_the_Basis_of_Impedance_Data_Using_the_Collaborative_Architecture_of_Polar-lights-optimizer-based_Random_Forest_and_Transfer_Learning_Supporting_Information_/33516952 CC BY 4.0