Title: Effect of training algorithms in accurate state of charge estimation of lithium-ion batteries using NARX model

Authors: Namrata Mohanty; Neeraj Kumar Goyal; V.N. Achutha Naikan

Addresses: Subir Chowdhury School of Quality and Reliability, Indian Institute of Technology, Kharagpur, West Bengal, 721302, India ' Subir Chowdhury School of Quality and Reliability, Indian Institute of Technology, Kharagpur, West Bengal, 721302, India ' Subir Chowdhury School of Quality and Reliability, Indian Institute of Technology, Kharagpur, West Bengal, 721302, India

Abstract: Accurate state of charge (SOC) estimation is required to ensure the safe and reliable operation of lithium-ion batteries (LIBs) in electric vehicles. Battery SOC does the similar function or operation as the fuel gauge in IC engine-driven vehicles indicating the energy left inside the battery to power a vehicle. This paper proposes the use of non-linear autoregressive exogenous input (NARX) model for accurate SOC estimation of LIBs and studies the effect of two training algorithms namely, Levenberg Marquardt and scaled conjugate gradient, on accuracy of proposed neural time series network model. Datasets including dynamic stress test (DST) profiles were extracted from the Center for Advanced Life Cycle Engineering (CALCE). Accuracy is evaluated in terms of mean squared error (MSE) and coefficient of determination (R-Square or R²). Levenberg-Marquardt algorithm is better than scaled conjugate gradient algorithm in providing better results for estimating SOC with a MSE of 4.61306 × 10^-6.

Keywords: SOC; state of charge; LIBs; lithium-ion batteries; electric vehicles; NARX; nonlinear autoregressive exogenous; machine learning; ANN; artificial neural network.

DOI: 10.1504/IJHVS.2023.132335

International Journal of Heavy Vehicle Systems, 2023 Vol.30 No.2, pp.232 - 254

Received: 19 Sep 2022
Accepted: 26 Jan 2023
Published online: 18 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article