State of health estimation for lithium-ion batteries using Gaussian process regression-based data reconstruction method during random charging process

Abstract

State of health (SOH) estimation is a critical technology to guarantee the safe and reliable operation of battery energy systems. Data-driven methods have been widely studied in the field of lithium-ion battery SOH estimation. However, random charging in real operating scenarios will result in difficult extraction of health features, which in turn limits the online application of data-driven methods. In this paper, a SOH estimation method using a piece of random charging data is proposed. In random charging scenarios, the health features screened within the predefined charging range are easily lost. For this reason, the partial charging data is reconstructed by the Gaussian Process Regression (GPR) model to intelligently supplement the health features that are not sampled. In terms of health feature selection, this paper selects the charging time sequence within a voltage interval as the input of the Long-Short-Term Memory (LSTM) estimation model, so that the complicated feature extraction work can be avoided, such as the peak of incremental capacity (IC) curve, the slope of the charging voltage and so on. The experimental results show that the data reconstruction method based on the GPR model can reconstruct the missing charging data accurately under a large number of simulated random charging data, and the reconstruction error of the charging data under two types of aging paths is less than 4.30%. Finally, using most known health features and a few reconstructed health features within certain charging intervals, the proposed GPR-LSTM method achieves a root-mean-squared percentage error (RMSPE) of 2.51% in SOH estimation for the two well-known laboratory datasets, and compared to 3.89% for the IC peak feature-based method, a reduction by 1.47%. Meanwhile, for a real-world battery data, the RMSPE of the proposed method is below 3%, illustrating the usefulness of the proposed method under the on-road conditions.