Data driven approach for state-of-charge estimation of lithium-ion cell using stochastic variational Gaussian process

Abstract

Modern electric vehicles rely on lithium-ion batteries. Electric vehicles (EVs) utilize intricate battery packs that require the oversight of a battery management system (BMS) to ensure safe, reliable, and efficient operation. The state estimation of the battery pack is an important responsibility carried out by the BMS. Accurately estimating the State-of-Charge (SOC) poses a considerable engineering challenge since it cannot be directly measured at the battery terminals. This study introduces a novel data-driven methodology for accurately estimating the SOC in Lithium-ion batteries, with a particular focus on its relevance in EV contexts. The framework is built upon the Stochastic Variational Gaussian Process (SVGP)—an improved version of the conventional Gaussian Process (GP). Unlike GP, It can scale up to very large datasets. Furthermore, SVGP uses variational inference to estimate the posterior instead of calculating, making it computationally efficient. The model training process involves using laboratory test data from an 18650 Lithium-ion Nickel Manganese Cobalt (NMC) cell that has gone through eight dynamic drive cycles. The findings showcase a remarkable level of precision in estimation, as indicated by an average R² value of 0.99 and a Mean Square Error (MSE) as low as 0.02.