Early prediction of battery degradation in grid-scale battery energy storage system using extreme gradient boosting algorithm

Abstract

The growth of battery energy storage systems (BESS) is caused by the variability and intermittent nature of high demand and renewable power generation at the network scale. In the context of BESS, Lithium-ion (Li-ion) battery occupies a crucial position, although it is faced with challenges related to performance battery degradation over time due to electrochemical processes. This battery degradation is a crucial factor to account for, based on its potential to diminish the efficiency and safety of electrical system equipment, thereby contributing to increased system planning costs. This implies that the health of battery needs to be diagnosed, particularly by determining remaining useful life (RUL), to avoid unexpected operational costs and ensure system safety. Therefore, this study aimed to use machine learning models, specifically extreme gradient boosting (XGBoost) algorithm, to estimate RUL, with a focus on the temperature variable, an aspect that had been previously underemphasized. Utilizing XGBoost model, along with fine-tuning its hyperparameters, proved to be a more accurate and efficient method for predicting RUL. The evaluation of the model yielded promising outcomes, with a root mean square error (RMSE) of 90.1 and a mean absolute percentage error (MAPE) of 7.5 %. Additionally, the results showed that the model could improve RUL predictions for batteries within BESS. This study significantly contributed to optimizing planning and operations for BESS, as well as developing more efficient and effective maintenance strategies.

Keywords

• Battery energy storage system
• Battery degradation
• Remaining useful life
• Extreme gradient boosting algorithm
• Hyperparameter tuning