EIS equivalent circuit model prediction using interpretable machine learning and parameter identification using global optimization algorithms

• Published in: Electrochimica acta Vol. 418; p. 140350
• Main Authors: Zhao, Zhaoyang, Zou, Yang, Liu, Peng, Lai, Zhaogui, Wen, Lei, Jin, Ying
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 20.06.2022; Elsevier BV
• Subjects: Algorithms; Classification; Electrochemical impedance spectroscopy; Equivalent circuit model; Equivalent circuits; Global optimization; Global optimization algorithm; Interpretable machine learning; Machine learning; Model accuracy; Parameter identification; Interpretable machine learning; Equivalent circuit model; Parameter identification; Classification; Global optimization algorithm; Electrochemical impedance spectroscopy
• ISSN: 0013-4686; 1873-3859
• DOI: 10.1016/j.electacta.2022.140350

Among seven multiclassification machine learning (ML) models taking optimized hyperparameters found by grid search, AdaBoost achieved the known highest equivalent circuit model (ECM) prediction accuracy, 0.571, and had a prediction basis that was consistent with a common chemical knowledge—the slowest step usually is vital in the whole electrochemical process. Twenty global optimization algorithms (GOA)s were assessed on simulated and experimental impedance spectra belonging to nine different ECMs, which proved that GOAs obtained nearly the same identification accuracy as the artificial identification under no interference of obvious abnormal points. ML combining with GOA provides a new possibility to automatically process EIS. [Display omitted]