Energy management for EV: state-of-charge estimation in Li-ion batteries with support vector regression hybrid approach

Abstract

Lithium-ion batteries serve as pivotal energy storage solutions across diverse do- mains, including smartphones, laptops, electric vehicles (EVs), and other applications. To ensure optimal performance and prevent sudden power loss, accurate estimation of the battery's state of charge (SOC) remains paramount. This chapter introduces a novel approach, using first the support vector regression (SVR) for SOC estimation, then exploits the gh-filter to refine this estimation, thus improving their accuracy. Notably, the training phase incorporates the US06 driving cycle, while the generalization ability of the SOC model is validated with LA92 and HWFET driving cycles. The main contributions of this study are first, adopting the Turnigy Graphene 5000 mAh 65C Li-ion battery data as a foundational dataset enhances the real-world applicability of the proposed methodology; also, unlike traditional methods, our SVR-gh approach reduces computational complexity, it bypasses the need for explicit battery model development, and finally, incorporating the gh-filter adds a layer of refinement, resulting in SOC estimates that are accurate and robust against noise. Through comprehensive testing on the LA92 and HWFET drive cycles, this study showcases the efficacy of the proposed approach in various operational scenarios. The achieved accuracy and adaptability substantiate its potential for broader adoption in diverse battery-powered systems. In summary, the SVR-gh method shows good SOC estimation accuracy and acceptable convergence ability under bad initial SOC.