State-of-Health Prediction For Lithium-Ion Batteries With Multiple Gaussian Process Regression Model

Abstract

State-of-health (SOH) prediction for lithium-ion batteries is a challenging and important topic in the modern industry. With the advent of cloud-connected devices, there are huge amounts of the battery degradation trend data available. How to make full use of these existing degradation data for the SOH prediction is a valuable problem deserving deep research. Aiming at this problem, a multiple Gaussian process regression (MGPR) method is proposed for the SOH prediction of lithium-ion batteries. In this work, the health indicators (HIs) are firstly extracted from the charging process curves of the batteries, and the mutual information analysis is used to select the important HIs which are strongly correlated to the SOH. These selected HIs are applied as the regression model input for describing the aging procedure of the battery effectively. Then, Gaussian process regression modeling is performed on the different batteries to bring multiple GPR models. Lastly, a weighting strategy based on the prediction uncertainty is designed to integrate the predictions from the multiple GPR models. The method validations are executed on the battery datasets from NASA, and the results show that the proposed MGPR method has higher prediction accuracy than the basic GPR method. INDEX TERMS Lithium-ion batteries, Gaussian process regression, mutual information, state-of-health.

State-of-Health Prognosis for Lithium-Ion Batteries Considering the Limitations in Measurements via Maximal Information Entropy and Collective Sparse Variational Gaussian Process

“…That is, the initial yardstick E is updated as the minimum value of E, N decreases by 1, and the while loop continues. Otherwise, the while loop ends (Steps [11][12][13][14][15][16][17][18][19]. Ultimately, the indexes of the selected features through the implementation of Algorithm 2 are outputted in I 2 (Steps 20-21).…”

“…Fig. 11(f)-(g) describes the impacts of the choice of kernel function on the prognosis results, where 1-4 denote linear (17), squared exponential (18), Matern 32 (19), and Matern 52 kernel function (20), respectively. The best prognosis performance is attained with choice 3, i.e., Matern 32 covariance function, which realizes the narrowest average RMSE range and therefore is the choice of the study.…”

“…However, it suffers from the incomplete reflection of the batteries' original status and difficulty in measurements. On the other hand, SOH is another method often to be utilized to quantify the extent of degradation and indicate the intensity of the internal chemical reactions [15], [16], [18], [24], [31], [44], [45], [47], [49]. It does not discriminate and treat any battery cell equally [5].…”

“…Nevertheless, the inner deficiencies in the measurements of such parameters are not comprehensively investigated. Moreover, the health indicators in some of the works like [18] were chosen manually, losing the ability of generalization.…”

Real-Time State-of-Health Estimation of Lithium-Ion Batteries Based on the Equivalent Internal Resistance

“…Feature extraction [10]- [12], verification, and analysis are carried out on the curve under different aging conditions to establish the relationship between these health indicators and the SoH. According to the literature [10], [11], [13]- [19], feature extraction can be conducted on the IC/differential voltage (DV) or differential thermal voltammetry (DTV) curves. For instance, Li et al established a quantitative relationship between the SoH and three peak-valley value points along with their positions on the IC curve fitted by a Gaussian process regression algorithm [13].…”