State estimation of lithium-ion batteries via physics-machine learning combined methods: A methodological review and future perspectives

Abstract

Lithium-ion batteries (LIBs) have become indispensable in modern energy storage applications. However, accurate and reliable state estimation, such as state of charge (SOC), state of health (SOH), and other critical variables, remain significant challenges, especially as LIBs are being pushed to their performance limits in advanced applications. Traditional methods can be broadly categorized into physics-based (PB) and machine learning (ML) methods. Each approach has its strengths but also inherent limitations. In recent years, integrating PB and ML methods has emerged as a promising solution to address these challenges, combining the physical interpretability of PB models with the adaptability and efficiency of ML techniques. This integration has demonstrated remarkable improvements, reducing estimation errors by approximately half compared to traditional methods. This review systematically categorizes these combined methods into three main strategies—serial, parallel, and hybrid—and further analyzes their applications in LIB state estimation, focusing on key variables such as voltage, SOC, SOH, state of temperature (SOT), and other states. Additionally, this review discusses key challenges in real applications and presents future outlooks. By synthesizing the current state of knowledge, this work provides valuable guidance specifically tailored for electric vehicle engineers and energy storage researchers facing real-world deployment challenges, offering potential benefits in terms of cost reduction and efficiency improvement in battery management systems. Ultimately, the accuracy, efficiency, and reliability of LIB state estimation can be advanced through hybrid methods, bridging the gap between academic research and industrial applications.