(59e) Probabilistic Lifespan Prediction of Lithium-Ion Batteries Under Varying Operating Protocols Using Gaussian Process Regression

Predicting the lifespan of lithium-ion batteries (LIBs) under varying charge/discharge conditions is crucial for optimizing their performance and preventing safety issues. While solely data-driven approaches have shown considerable performance under limited operating conditions, they have failed to provide accurate lifespan predictions in practical circumstances for which sufficient data do not exist. For example, although [1] achieved high accuracy in predicting the lifespan through data-driven models, the data range was limited to the depth-of-discharge (DOD) of 100% whereas in actual use it varies a lot and the variations are known to make significant impact on the lifespan of the LIB.

Therefore, hybrid models that integrate data-driven approaches and model-based methods have been developed. For example, previous studies such as [2] and [3] proposed Gaussian process regression (GPR) models for an efficient prediction with sparse data, but these models did not incorporate information regarding operating patterns, and only utilized early-cycle measured data. Additionally, while [4] developed a cycle life model that considers the influence of the charging rate, operating temperature, and DOD individually, it lacks generalizability as it assumes an empirical form of each variable’s impact and also does not explain their combined effects.

In this study, a hybrid model is presented that can predict the end-of-life (EOL) of LIB under various operating conditions. The model utilizes three operating variables, including C-rate, DOD, and operating temperature. The required data for the model is generated using an electrochemical model (EM) called LIONSIMBA [5]. Since available data in the literature are limited to some specific operating conditions, new degradation data must be generated to comprehensively consider a wide variety of operating patterns. [6] is selected for validation of the EM. Then, virtual degradation experiments are conducted under varying combinations of the C-rates, environmental temperatures, and DODs until LIBs reach EOL.

The generated data is then learned by a GPR model, which can alleviate the problems due to the limited data. A 3-dimensional GPR model is trained to predict EOL. Since GPR provides a confidence interval of the predicted value, it would provide a more informative prediction of EOL in untested conditions. To train the GPR, data is randomly split into train and test datasets with a certain proportion. After training the model, model evaluation is conducted based on the test data which reside within the 95% confidence interval. Results indicate that a reasonable lifespan of LIBs under different operating conditions can be gained with small input datasets.

References

[1] Severson, K.A., et al., Data-driven prediction of battery cycle life before capacity degradation. Nature Energy, 2019. 4(5): p. 383-391.

[2] Liu, J. and Z. Chen, Remaining Useful Life Prediction of Lithium-Ion Batteries Based on Health Indicator and Gaussian Process Regression Model. IEEE Access, 2019. 7: p. 39474-39484.

[3] Jia, J., et al., SOH and RUL Prediction of Lithium-Ion Batteries Based on Gaussian Process Regression with Indirect Health Indicators. Energies, 2020. 13(2): p. 375.

[4] Omar, N., et al., Lithium iron phosphate based battery – Assessment of the aging parameters and development of cycle life model. Applied Energy, 2014. 113: p. 1575-1585.

[5] Torchio, M., et al., Lionsimba: a matlab framework based on a finite volume model suitable for li-ion battery design, simulation, and control. Journal of The Electrochemical Society, 2016. 163(7): p. A1192.

[6] B. Saha, K.G., Battery Data Set. 2007: NASA Prognostics Data Repository, NASA Ames Research Center, Moffett Field, CA.