Graph-Vector Autorregressive Model for Lithium-Ion Cell Capacity Estimation

Battery degradation is a critical consideration in ensuring the longevity and reliability of energy storage systems, particularly in devices deployed in remote locations. Characterizing battery degradation allows for the allocation of maintenance resources in networks deployed in remote locations. Some chemistry types such as in lithium-ion batteries, that is very sensitive to cycling conditions, which cause degradation in the battery capacity. In this paper, a data-driven model was designed to characterize the importance of the relevant variables in this process and the estimation of battery capacity fading. This approach involves learning the similarities between different operating conditions represented by time series and their influence in the capacity degradation process. This is performed by learning a Graph Vector Autoregressive model that captures the evolution of the degradation with past values of the time series and the dependencies across degradation. The proposed model was evaluated in a real dataset composed of twenty-four times series of capacity degradations over time cycled under constant conditions of temperature, charging current cut-off and discharging current. The results show that the G-VAR model can estimate new cell information with a determination coefficient, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>, higher than 95%. It was compared with similar state-of-the-art models such as ARIMA or other parametric models showing better performance.