Data granularity studies for the development of NMC Li-ion battery models

This paper presents a comparative study of various models used to characterize the behavior of Li-ion batteries. Specifically, this work focuses on analyzing data granularity and examining important features such as model accuracy and required computing resources. The article presents the aforementioned study employing different types of algorithms for voltage and state of charge (SOC) estimation. Each of these models are formulated and calibrated using different techniques based on their typology. To conduct these studies, real Nickel Manganese Cobalt Oxide (NMC811) lithium-ion batteries with a 4.8 Ah capacity are considered. These cells were tested at different temperature conditions and under real-world application profiles, including electric vehicle (EV) and stationary battery energy storage system (BESS), to analyze and compare the impact of data granularity. For low power variability scenarios, such as the stationary BESS, using a 60s time resolution barely affects model accuracy while reducing execution time by 50%–75% and memory usage by 98%. High power variability scenarios such as EVs require finer resolutions (5–10s) to maintain voltage prediction accuracy and still allow simulation time and memory reductions of 10%–50% and 80%–90%, respectively. SOC estimations reflect higher tolerance to data granularity, and a 30s resolution provides an optimal trade-off, reducing simulation time by 25%–50% and memory demands by up to 97%. The study concludes by assessing the energy consumption associated with different granularities and execution frequencies and analyzes key communication requirements, proposing in this way a benchmark for developing more efficient digital twins of battery systems.