aalto1 untyped-item.component.html

Battery degradation in stationary energy storage systems — Modelling approaches for emerging and established battery chemistries

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Engineering | Master's thesis

Department

Mcode

Language

en

Pages

106

Series

Abstract

The rapid deployment of battery energy storage systems has highlighted crucial knowledge gaps in battery degradation modelling, particularly for sodium-ion batteries (SIB) compared to well-established lithium iron phosphate (LFP) models. This work investigates degradation mechanisms across LFP and SIB chemistries, applying machine learning frameworks to improve lifetime prediction accuracy and explore cross-chemistry knowledge transfer. A comprehensive analysis evaluated five empirical models against machine learning approaches including Random Forest and XGBoost algorithms using LFP degradation data spanning 885 days across 16 calendar and 13 cyclic ageing conditions. Transfer learning methodologies were applied to adapt LFP-trained models for SIB prediction using research and industrial datasets, with parallel from-scratch modelling for comparison. Machine learning approaches demonstrated superior performance over empirical methods, with XGBoost achieving 7.6x improvement in prediction accuracy and maintaining robust performance across extreme operating conditions where empirical models failed. Transfer learning results varied significantly by degradation mechanism, succeeding for cyclic ageing but initially failing for calendar ageing, though performance was recovered through fine-tuning. The findings establish machine learning as a possibly transformative approach for battery degradation prediction, though chemistry-specific mechanisms limit the practical utility of cross-chemistry transfer learning for broader applications.

Description

Supervisor

Santasalo-Aarnio, Annukka

Thesis advisor

Smajila, Luka

Other note

Citation

Endorsement

Review

Supplemented By

Referenced By