Identification of optimal prediction error Th\'evenin models of Li-ion cells using the MOLI approach

TL;DR

This paper develops and compares system identification algorithms for optimal prediction error Thévénin models of Li-ion cells, incorporating Randles circuit and Warburg impedance to improve battery modeling accuracy.

Contribution

It introduces novel algorithms for identifying 1st and 2nd order Thévénin models and compares their performance with Randles circuit models using experimental data.

Findings

Thévénin models show comparable accuracy to Randles circuit models.

The proposed algorithms effectively estimate model parameters from experimental data.

Warburg impedance can be approximated by finite order LTI models.

Abstract

This report presents System Identification algorithms to estimate the dynamical model of Li-Oin cells. First the dependence of open circuit voltage (OCV) on the state of charge (SOC) is studied. thN battery equivalent model when a resistor is added to the circuit is stated. The discharge data is divided into segments where the internal resistance is assumed constant, and therefore SOC is constant, thence is described an LTI identification algorithm to be used to estimate the cell model in each segment. A Randles circuit is introduced to the model to describe the diffusion process. This model includes the so called Warburg impedance which is as fractional system. This impedance is discussed and it is approximatted by a finite order linear time invariant state-space model. Also, after presenting the simplified Randles circuit, is stated an identification algorithm that estimates the parameters of this model. The Th\'evenin model is presented as an alternative to the Randles circuit. An algorithm to identify a Th\'evenin model of 1st and 2nd order is enunciated. The performances of the simplified randles model and of the two Th\'evenin models models described and its respective identification algorithms, are discussed and compared using an experimental set of data.