Data-Driven Nonlinear Identification of Li-Ion Battery Based on a Frequency Domain Nonparametric Analysis

Abstract

Lithium ion (Li-ion) batteries are attracting significant and growing interest because their high energy and high power density render them an excellent option for energy storage, particularly in hybrid and electric vehicles. In this paper, a data-driven polynomial nonlinear state-space model (PNLSS) is proposed for the operating points at the cusp of linear and nonlinear regime of the battery's electrical operation, based on the thorough nonparametric frequency domain characterization and quantification of the battery's behaviour in terms of its linear and nonlinear behaviour at different levels of the state-of-charge (SoC).Index Terms-Li-ion battery, input-output response, Nonparametric characterization, polynomial nonlinear state-space (PNLSS), nonlinear system identification. I. BATTERY MODELLINGThe pursuit for battery models with high accuracy and computational efficiency still remains a challenge. Generating a mathematical model of a Li-ion battey, e.g. needed by battery management system (BMS), that can describe the input current-to-output voltage dynamics of a battery is a challenging problem. A primary reason for this is that battery dynamics vary significantly with operating conditions. Depending on the final purpose of the developed model, one can divide battery models into the models that describe the short term behaviour e.g. state of charge, voltage response etc. and models that describe the long term behaviour of the cells, e.g. lifetime models, state of health etc. In the field of battery modelling many different battery models exist for both the short and long term behaviour of battery cells [1]. These models can be broadly classified in the following categories: 1) Equivalent circuit models (ECM), 2) Electro-chemical models, 3) Analytical and impedance based models, 4) Empirical and semi-empirical models. ECMs are structurally simple and computationally efficient due to the use of lumped-parameter circuit elements, e.g. inductors, resistors and capacitors, to represent the battery impedance, and these models frequently The authors * are with the ELEC Department and the authors ** are with MOBI research group of the Vrije Universiteit Brussel (VUB), Belgium respectively. Corresponding author can be contacted at Rishi.Relan@vub.ac.be. incorporate empirical functions to describe the relationship between SoC and open circuit voltage (OCV). ECMs thus are widely used for impedance analysis [2], SoC estimation [3] and charging control [4]. Performances of several commonly used ECMs is compared in [5]. On the other hand electrochemical models [6]-[8]usually use coupled nonlinear partial differential equations to describe ion transport phenomena and electrochemical reactions to achieve high accuracy, but incurring heavy computation load. In general electrochemical models such as pseudo two dimensional models [9], single particle models [10], and extended single particle models [11] are more accurate than ECMs. In comparison ECMs are easier to implement, but have worse accuracy than electrochemical models...

Joint Estimation of Battery Parameters and State of Charge Using an Extended Kalman Filter: A Single-Parameter Tuning Approach

“…5 are quite similar to those of the WLTP cycle up to k ≈ 9000s. As mentioned previously, from this point in time (i.e., low SoC range) the battery behaviour becomes more nonlinear, see, e.g., [30]. Therefore, the underlying parameters of the system change, which is clearly visible in the voltage behaviour in Fig.…”

An overview of grid-edge control with the digital transformation

“…As highlighted in [31], the accuracy of the data-driven models is strictly related to the amount of training data available. In the attempt to overcome this issue, online auto-adaptive parameter identification methodologies have been developed.…”

A Local Polynomial Approach to Nonparametric Estimation of the Best Linear Approximation of Lithium-Ion Battery From Multiple Datasets

“…439 of [15]). Hence, individual BLAs estimated at varying operating points can be used to develop black-box linear timevarying or parameter-varying models or the C BLA can be used as initialization for the nonlinear model structure proposed in [10].…”