An Adaptive and Hybrid State of Charge Estimation Method Integrating Sequence-To Learning and Coulomb Counting for Li-Ion Based Energy Storage Systems

Abstract

For safe and long-lasting operation of Li-ion batteries used in electric vehicles and electric grid applications, the State of Charge (SOC) of the battery cell must be estimated with high accuracy. However, due to the uncertainty in environmental conditions and the complex nature of battery chemistry, SOC estimation still presents a significant challenge. In this study, an adaptive and hybrid method for SOC estimation of a Li-ion battery cell is proposed. Convolutional Neural Network (CNN) based Sequence-to-point learning architecture is used to estimate the initial SOC values at specific time intervals. In order to increase the estimation accuracy, a multi-scale CNN architecture is designed, and useful features are captured. The obtained estimation values are integrated with the partial coulomb counting method to increase the accuracy. In addition, the proposed model adaptively updates the estimation weights with the help of the estimation error data obtained during the full charging of the batteries. The proposed model is tested on the LG 18650HG2 dataset. The results prove that the proposed model is 23% more accurate than benchmark models at 25°C and 55.5% more accurate at 0°C.