A novel State of Charge and State of Health estimation technique for Lithium-ion cells using machine learning based Pseudo-Random Binary Sequence method

Abstract

Lithium-Ion batteries are electrochemical storage devices with State of Charge (SOC) and State of Health (SOH) sensitive operations. The accurate estimation of these two parameters is vital for the battery's optimum usage for different applications. Electrochemical Impedance Spectroscopy (EIS) is a benchmark technique that measures the cell internal impedance for online estimation of these parameters. Recent studies have used EIS with Discrete Fourier Transform, neural networks, and other machine learning algorithms for improving the accuracy and reliability of these parameters. However, such methods also increase system complexity, time interval, and cost of the estimation system. Few researchers have also used Pseudo-Random Binary Sequence (PRBS) techniques to improve the speed of EIS estimations. However, they have lower accuracy due to frequency harmonics, and the continued requirement of simultaneous measurements of physical parameters introduces complexity in designing an online estimator. This paper proposes a novel hybrid method of estimating these parameters using a combination of a PRBS and an online piecewise linear machine learning technique. A hardware prototype is developed to validate the proposed method resulting in a cost-effective and simpler method with an estimation accuracy of 98 % at 5-sec intervals.