Review of Lithium-Ion Battery Capacity-Based State of Health Estimation Algorithms

Abstract

Accurate estimation of the state of health (SOH) of lithium-ion batteries is essential for ensuring the safety, reliability, and longevity of electric vehicles, battery energy storage systems, and other energy applications. This paper presents a comprehensive review of capacity-based SOH estimation algorithms, focusing on methods suitable for real-time battery management system (BMS) implementation. Existing approaches are categorized into model-based and data-driven methods. Model-based techniques, including equivalent circuit models, electrochemical models, and filtering algorithms such as extended Kalman filters, unscented Kalman filters, and particle filters, provide strong interpretability and compatibility with embedded systems. Data-driven approaches, including machine learning, deep learning, transfer learning, and feature-driven statistical methods, offer improved accuracy and adaptability by learning degradation patterns directly from operational data. Methods that fuse physical modelling with data-driven learning are reviewed within both categories, according to which component forms the structural core of the estimator. The paper compares these approaches using key criteria relevant to BMS deployment, including estimation accuracy, robustness, interpretability, computational complexity, and implementation feasibility. Emerging trends such as physics-informed learning, cloud-edge collaboration, and digital twinning are identified as promising directions for developing scalable, adaptive, and practical next-generation SOH estimation algorithms.