Lithium-sulfur cell state of charge estimation using a classification technique

Lithium-Sulfur (Li-S) batteries are a promising next-generation technology providing high gravimetric energy density compared to existing lithium-ion (Li-ion) technologies in the market. The literature shows that in Li-S, estimation of state of charge (SoC) is a demanding task, in particular due to a large flat section in the voltage-SoC curve. This study proposes a new SoC estimator using an online parameter identification method in conjunction with a classification technique. This study investigates a new prototype Li-S cell. Experimental characterization tests are conducted under various conditions; the duty cycle – intended to represent a real-world application – is based on an electric city bus. The characterization results are then used to parameterize an equivalent-circuit-network (ECN) model, which is then used to relate real-time parameter estimates derived using a Recursive Least Squares (RLS) algorithm to state of charge using a Support Vector Machine (SVM) classifier to estimate an approximate SoC range. The estimate is used together with a conventional coulomb-counting technique to achieve continuous SoC estimation in real-time. It is shown that this method can provide an acceptable level of accuracy with less than 3% error under realistic driving conditions.