A novel method for the modeling of the state of health of lithium-ion cells using machine learning for practical applications

TLDR: The conducted analyses demonstrated that the degradation process in the case of LIBs was characterised by high variability depending on the cyclic operation parameters, which was taken into account in the new model, which is an improvement on the currently existing methods.

Abstract: In this article, the authors propose an original method for the modeling of the state of health of cyclically operating lithium-Ion batteries (LIBs), based on Gaussian process regression. This method allows for the estimation of the degradation of the LIBs during an equivalent duty cycle at various load patterns. The results of many years of research on the degradation of LIBs have been analyzed in two aspects. The first one concerned degradation under constant loads, and the second was related to degradation taking into account randomly variable loads. The conducted analyses demonstrated that the degradation process in the case of LIBs was characterised by high variability depending on the cyclic operation parameters (the charging and discharging half-cycle). Furthermore the degradation of LIBs depends, to a significant extent on the current state of health. For this reason, this parameter was taken into account in the new model, which is an improvement on the currently existing methods. The developed model has been verified by simulating the variable load of the cells during its entire lifespan — the obtained percentage prediction error margin during the whole simulation did not exceed 5%, which confirmed its practical usefulness.