State of Charge Estimation for Lithium-ion Batteries in Electric Vehicles by Transformer Neural Network and L1 Robust Observer

Accurately estimating the lithium-ion battery's state of charge (SOC) is of great importance to the electric vehicle (EV) operations. In this paper, an innovative duo-layered algorithm consolidates a Transformer neural-network and an L 1 robust observer is originated to estimate the SOC of an EV's battery. For the upper layer, the current, voltage, and temperature data are imported into the novel Transformer to predict the SOC. Subsequently, the lower-layer L 1 robust observer strives for smoothing the output from the upper-layer machine learning model. Such a novel SOC estimator is advantageous in two aspects. On the one hand, the Transformer outpaces other recurrent neural networks (RNNs) owing to its competency of finding the dependency between any two positions in the input and output sequences, and of acquiring richer information. On the other hand, the L 1 robust observer concomitantly achieves the peak-to-peak attenuation from the disturbance to the estimation error and the robustness against the model uncertainties. The new method is evaluated based on experimentally collected data in US06 cycle, and the result manifests its improved accuracy over a baseline method.