Towards Accurately Explaining Battery Behavior: Learning the Parameters of Analytical Battery Models

Accurately predicting the state of charge (SoC) of batteries is essential for optimizing their performance and integration into larger energy systems.While black-box models, such as neural networks, can achieve high accuracy, they lack interpretability and validation by domain experts.In contrast, white-box models provide transparency, allowing for expert evaluation and better understanding of environmental influences on battery behavior.In this work, we propose a concept for a learning pipeline to obtain the parameters of analytical battery models, which are considered as white-box models and predict SoC evolution.Hereby, we include multiple models into our pipeline, such as a linear model and the Kinetic Battery Model (KiBaM).Within the pipeline, we fit the analytical models to a specific battery by learning from its operational data.The resulting battery model then can be used for the analysis of more general aspects like safety properties in battery-powered systems.