AI-driven lithium-ion battery health prediction: a review of SOC, SOH and RUL estimation techniques

Abstract

The need for precise lithium-ion battery (LiB) monitoring and prognostics has grown significantly with the increasing use of electric vehicles (EVs) and renewable energy system (RES) technologies. State of charge (SOC), state of health (SOH), and remaining useful life (RUL) are three critical performance metrics for ensuring battery safety, functional dependability, and effective energy utilization. Non-linear degradation patterns, operating uncertainties, and dynamic environmental fluctuations are frequently difficult for traditional electrochemical and empirical assessment methods to capture. As a result, recent research has focused on advanced data-driven techniques for comprehensive battery state estimation, notably those based on artificial intelligence (AI) and deep learning (DL). This review provides an overview of current methodologies for SOC, SOH, and RUL prediction, with a focus on modern DL architectures such as transformer networks, autoencoder-based latent modeling, neural basis expansion techniques, attention mechanisms, and hybrid optimization-driven frameworks. The article examines new advances in intelligent battery prognostics critically, emphasizing their potential for improved estimation precision, resilience, and flexibility in practical operating scenarios. In addition, it identifies future research directions such as enhanced optimization-assisted frameworks, hybrid physics-informed learning, and explainable artificial intelligence (XAI). This article intends to offer a comprehensive overview of modern battery prognostics approaches to aid in the development of dependable, efficient, and long-term battery management system (BMS) solutions for electric mobility and energy storage systems.