Residual Error Modelling based Hybrid Wavelet-ConvLSTM Deep Learning Model to Estimate SOC of EV Batteries

Abstract

State of charge (SOC) estimation has been an essential requirement for battery management system as seen in recent years especially for Electric Vehicles (EVs). The SOC stipulates the battery's amount of charge left so as to indicate how long more it can be used to drive an EV. Hence, accurate and uninterrupted monitoring of lithium-ion batteries (LiBs) is of utmost priority in order to maintain EVs efficient energy usage and battery life cycle. Although there has been a long research on estimating SOC, there are no accurate and easy electronic equipments available that can measure the SOC. Deep learning algorithms have shown high accuracy and reliable SOC estimation due to their ability to process large amount of data with ease under noisy conditions. This article proposes an efficient approach to estimate SOC by incorporating residual prediction error based wavelet-convolutional long-short term memory (ConvLSTM) deep learning model. Publicly available lithium ion battery dataset was used in this article for experimentation and validation of the results. Due to stochastic nature of data, multilevel wavelet decomposition has been analysed for improving accuracy of estimation. The proposed method results well in terms of accuracy and generalization capability.