THE RELEVANCE OF THE OPTIMIZATION ALGORITHM ON THE DATA-DRIVEN ESTIMATION OF THE STATE-OF-CHARGE OF THE PANASONIC 18650PF LITHIUM-ION CELL USING DEEP FEEDFORWARD NEURAL NETWORKS

Abstract

This article presents a novel empirical study for the estimation of the State of Charge (SOC) of a lithium-ion (Li-ion) battery which uses a Deep Feedforward Neural Network (DFNN) with three hidden layers. We model a series of ten vehicle drive cycles that were applied to the Panasonic 18650PF Li-ion cell. Our results show that the choice of the optimization algorithm affects the model performance. The proposed model was able to achieve an error smaller than 1.0% over a dataset of 25o C in all drive cycles.