State‐of‐Health Estimation of Lithium‐Ion Batteries Based on Thermal Characteristics Mining and Multi‐Gaussian Process Regression Strategy

Abstract

Accurate prediction of the state of health (SOH) is essential to ensure the safety and reliability of battery operation. The thermal factor is an important indicator of SOH and many methods based on temperature are sensitive to measurement noise. To improve SOH estimation precision, a new health indicator (HI) directly extracted from the temperature curve is developed and an integrated multi-Gaussian process regression (MGPR) model is proposed. First, based on the trend analysis of the charging temperature curves with battery degradation, three features that can reflect thermal characteristics are extracted as the HI. Second, considering that the model generated by machine learning is influenced by the training dataset and the inherent inconsistencies in batteries, MGPR model is proposed to improve the model fitness. The training data is reformed and multiple GPR models are established. The multiple models are weighed by taking into account the prediction uncertainty to get the final SOH estimation result. Finally, two types of open-source data relative to different ambient temperatures and operating profiles are used to verify the performance. Experiment results show that the HI developed can characterize the battery degradation well and the MGPR model has high robustness and can obtain high-precision estimation results.

The Impact of Calendering Process Variables on the Impedance and Capacity Fade of Lithium‐Ion Cells: An Explainable Machine Learning Approach

“…In the recent years, artificial intelligence (AI) and specially machine learning (ML) techniques have progressed as powerful tools to support the state estimation and performance analysis of batteries [29,30] and control their manufacturing processes. [31,32] Considering each individual process of Li-ion production chain, ML models are developed to relate the manufacturing variables and the physical characteristics of the cathode in studies.…”

A Novel Remaining Useful Life Prediction Approach Combined eXtreme Gradient Boosting and Multi-Quantile Recurrent Neural Network

“…Therefore, the RUL prediction methods using point estimation are unsuitable for making maintenance decisions while considering risks, since risk is usually measured through uncertainty. The uncertainty quantification in RUL prediction mainly depend on statistical model-based methods, such as Gaussian process regression [32], hidden Markov model [33], etc. Only a few researchers have paid attention to uncertainty quantification of RUL prediction results based on machine learning in recent years.…”

Self‐Discharge Voltage Drop Estimation Method Based on Improved Gaussian Process Regression

“…[ 16 ] At present, there is no research on estimating the self‐discharge voltage drop by GPR. However, GPR method has been widely used in the field of SOC [ 17 ] and state of health (SOH) [ 18 ] of lithium batteries. Deng et al [ 17 ] and Guo et al [ 18 ] show that the GPR model has good performance in the estimation of SOC and SOH.…”

“…However, GPR method has been widely used in the field of SOC [ 17 ] and state of health (SOH) [ 18 ] of lithium batteries. Deng et al [ 17 ] and Guo et al [ 18 ] show that the GPR model has good performance in the estimation of SOC and SOH. As the self‐discharge process of a lithium‐ion battery is a complex, dynamic, and nonlinear electrochemical process, it is often difficult to accurately estimate the self‐discharge voltage drop.…”