Integrated approaches for lithium-ion battery state estimation and life prediction: A critical review of model-driven, data-driven, and hybrid techniques

Abstract

Lithium-ion batteries (LIBs) are crucial for a wide range of applications, from electric vehicles to grid storage, and require accurate state-of-charge (SOC), state-of-health (SOH), and remaining-useful-life (RUL) estimates to achieve optimal performance and safety. However, the existing prediction methods face key challenges, such as their insufficient adaptability to dynamic working conditions, heavy reliance on high-quality data, and limited model generalizability, as well as the resource constraints imposed by edge devices. In this paper, model-driven, data-driven and hybrid methods are comprehensively analysed to highlight their advantages and limitations. Model-driven approaches have strong interpretability, but they face challenges in terms of their computational complexity and adaptability to dynamic conditions. Data-driven methods perform well in terms of flexibility, real-time deployment and lightweight computing. However, these methods usually require high-quality data. Hybrid methods combine physical interpretability with data-driven real-time performance, effectively reducing their reliance on high-quality data and significantly improving their accuracy and generalizability in different operational scenarios. Future battery state management research must prioritize lightweight modelling, quantum computing for atomic-level electrochemical simulations, and meta-learning for adaptive battery management systems (BMSs) to overcome the current deployment limitations. These advancements will jointly drive BMS technology towards higher efficiency, safety and sustainability levels, bridging the gap between laboratory research and industrial applications.