State‐of‐charge estimation of lithium‐ion batteries using composite multi‐dimensional features and a neural network

Abstract

A novel method that uses composite multi-dimensional features data to estimate the state of charge (SOC) of a battery is presented to address the shortcomings of using single-dimensional feature data. Two types of data, the terminal voltage and the terminal current, which can be obtained directly by measuring, are selected as low-dimensional feature data. The open-circuit voltage (OCV), as high-dimensional feature data, cannot be directly measured, and can be used to estimate the SOC by the OCV-SOC method. Thus, in this study, the second-order RC equivalent model of a battery is used and the OCV is identified online by the forgetting factor recursive least-squares algorithm. The proposed method is implemented by first using a feed-forward neural network, followed by a time-series neural network. The dynamic stress test and urban dynamometer driving schedule discharging profiles are applied to train and test the two neural networks. The experimental results show that the proposed method can estimate the SOC more accurately than neural networks using only single-dimensional feature data. Moreover, the time series neural network can overcome the shortcomings of traditional estimation methods.