Supervised Learning and Data Intensive Methods for the Prediction of Capacity Fade of Lithium-ion Batteries under Diverse Operating and Environmental Conditions

Abstract

The enactment of lithium-ion batteries (LIBs) is influenced through complex, interconnected electrochemical, electrical, thermal, and mechanical processes occurring at different spatial and temporal scales. To pave the way for a sustainable, fossil-free energy system, it is crucial to advance an inclusive understanding of mechanisms applied to capacity degradation in LIBs. Accurate estimation of capacity fade in lithium-ion batteries is crucial for ensuring their safe use, diagnostics, and prognostics, ultimately leading to increased adaptability. However, currently available models for capacity fade prediction methods lack efficiency in predicting accuracy across different algorithms. This paper presents a framework for capacity fade estimation and monitoring in LIBs using various supervised ML algorithms, including support vector regression, decision tree, K-nearest neighbor, and random forest. The study proposes use of NMC 811 (Ni0.84Mn0.06Co0.1 (H-NMC)/ Graphite +SiO), which offers high energy and power density, and compares its performance using various error models such as mean absolute error, mean absolute percentage error, root mean squared error, and mean absolute percent error at discharge rates of 1C, 2C, and temperatures ranging from 15°C to 35°C. The analysis utilizes a constant charging rate of 0.5C, covering the full state of charge range (0-100%), for a complete cycle of the cell. The obtained results are validated using a different set of cells denoted as “cell b,” which undergoes the same test parameters for comparison. Health features such as data preprocessing, analysis, charge capacity, and discharge capacity are utilized to understand the capacity degradation mechanism. Notably, this paper stands out as it employs four distinct supervised machine learning methods for capacity fade estimation, with a comparison of model accuracy, marking the first time such an approach has been used. © 2023, Central Board of Irrigation and Power. All rights reserved.