Many algorithms that monitor and optimize the operation of battery energy storage systems (BESS) require accurate cell models, which, in return, must be parametrized accordingly. With the use and degradation of the battery cells, these parameters have to be continuously adapted during the operation to correct for model drifting. Many methods have been previously developed to fit the battery models with the help of field data. A common requirement is a readily available open-circuit voltage (OCV). This assumption, however, might not always hold, especially since, like with other model parameters, the shape of the OCV curve changes with cell degradation.

In this work, we introduce a novel approach to build a battery cell model exclusively from field data, with no previous knowledge required. We leverage the flexibility of Gaussian process regression (GPR) to enhance the full parametrization of an equivalent circuit model (ECM) in a data-driven fashion. As part of this study, we evaluate the model’s accuracy and compare it to a state-of-the-art parametrization method using experimental data from cells under increasing aging conditions.

The data-driven enhanced model effectively adapts to the aging OCV and internal resistance, unlike the conventional method, which loses accuracy with increasing cell degradation. However, the model can only make reliable predictions within the SOC range it was trained on, so it’s important to ensure the dataset covers a wide range of operating conditions to maximize its effectiveness.