Deep Learning Enabled State of Charge Estimation for Electric Vehicle Batteries Under Noise Effects

• Main Authors: Lipu, M.S.H., Miah, M.S., Ansari, S., Meraj, S.T., Hasan, K.
• Format: Conference or Workshop Item
• Published: 2022
• Online Access: http://scholars.utp.edu.my/id/eprint/37622/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85159055211&doi=10.1109%2fSTI56238.2022.10103230&partnerID=40&md5=2132e0cf232a2eb5037bb86afd0e7555

State of charge (SOC) is a vital parameter utilized to examine the performance of battery storage systems in electric vehicle (EV) applications. The lithium-ion batteries have been broadly utilized in the EV application for SOC estimation because of their high voltage, energy, capacity, and long life cycles. Nonetheless, there has been a concern in determining SOC for lithium-ion battery storage due to battery aging, chemical reaction, material degradation as well as noise and temperature impacts. Deep learning has demonstrated very effective in SOC estimation under variable environmental settings and dynamic load conditions. Thus, this work introduces long short-term memory (LSTM) network for SOC estimation due to its improved generalization performance and strong computational capability in highly battery nonlinear conditions. The accuracy and robustness of LSTM are verified by including random noises and bias to the measured dataset achieved through the battery experiments tests at ambient temperatures and publicly available EV drive cycles at different temperatures. In addition, the proposed LSTM is dominant to state-of-the-art machine learning (ML) techniques in terms of obtaining low root means square error (RMSE) and SOC error. The results indicate that the LSTM archives SOC error below 5 in all battery experimental tests and EV drive cycles. © 2022 IEEE.