Decoding degradation: The synergy of partial differential equations and advanced predictive models for lithium-ion battery

Abstract

Recent advancements in machine learning (ML) algorithms have transformed Li-Ion battery analysis, focusing on crucial parameters like State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL). However, due to increasing computational complexity and lack of fundamental process understanding within the developed models, conventional predictive tools suffer its integration in real world applications. To address this gap, our study introduces a hybrid modeling approach consisting of two stages. In the first stage, we apply various machine learning algorithms to predict battery degradation using empirical data, focusing on capturing the initial patterns of battery behavior under different operating conditions. In the second stage, we enhance these predictions by integrating Partial Differential Equations (PDEs) that incorporate fundamental physical principles governing battery performance. This combination creates a physics-informed ML model that bridges the gap between empirical data and theoretical understanding. The integration of PDEs significantly improves the model's accuracy in predicting both degradation and discharge capacity. Our results demonstrate a marked enhancement in key performance metrics, with the hybrid model achieving a Mean Square Error (MSE) of 0.2091, Root Mean Square Error (RMSE) of 0.4572 and Mean Absolute Error (MAE) of 0.2555. In comparison, the errors from the initial stage-one ML predictions were substantially higher, with MSE, RMSE, and MAE values of 10.3648, 3.2194, and 0.7392, respectively. These findings highlight the hybrid model's effectiveness and its potential to significantly improve battery management practices, ultimately contributing to the extension of battery lifespan.