Analyzing the Performance of AI-Based Battery SoC Estimation: A Metrological Point of View

Battery State of Charge (SoC) estimation systems play a crucial role in modern energy infrastructures challenges, including the integration of renewable energy, grid stability, and the electrification of transportation. The established technologies of Artificial Intelligence (AI) and Machine Learning (ML) have proven instrumental in achieving heightened accuracy and efficiency within battery state estimation frameworks. Highlighting a critical gap in current applications, the research underscores the need for a comprehensive treatment of measurement uncertainty in AI-driven battery state estimation. The proposed methodology introduces a novel approach that incorporates measurement uncertainty into the evaluation of the ML model, investigating how the SoC estimation system is influenced by the measurement accuracy, and contributing to a deeper understanding of uncertainties associated with AI systems. The investigation focuses on the application of data-driven ML techniques, particularly the Nonlinear AutoRegressive with eXogenous inputs (NARX) model for its proficiency in SoC estimation. The results provide valuable metrological insights into the ML model and a starting point toward reliable battery SoC estimation systems.