A real-time learning-assisted charging strategy for lithium-ion batteries in electric vehicles

Abstract

The effective, secure, and adaptive charging of lithium-ion batteries in electric vehicles remains a significant challenge. This paper introduces a Real-time Learning-Assisted Charging Strategy, a new hybrid control framework that combines Constant Current–Constant Voltage charging with pulse current modulation and smart, real-time learning feedback. Unlike traditional hybrid or adaptive algorithms that rely on predetermined transition thresholds, the proposed system continuously learns from actual cell responses, including voltage, current, temperature, and State of Charge. This allows it to adaptively adjust parameters such as pulse amplitude, rest time, and voltage hold phases, enabling accurate thermal control and maximum energy transfer during charging. Experimental verification using an eight-cell 6000 mAh NMC pack demonstrates that the method achieves a charging efficiency of up to 98 %, a charge time of 42 min, and a thermal deviation of less than ±0.3 °C. In parallel, MATLAB/Simulink simulations confirm the performance trend and further predict a 21 % reduction in total charging time and a 37 % increase in cycle life under idealized conditions, while maintaining a thermal deviation of less than 4 °C. Additionally, it maximizes long-term capacity retention (85 % after 500 cycles) in the experimental study and increases projected cycle life by 37 % through simulation compared to the traditional CC–CV approach. These results indicate that the proposed method not only improves control but also serves as an optimization framework driven by learning, bridging the gap between model-based predictions and real-time experimentation. This approach provides a scalable, reliable, and intelligent foundation for next-generation Electric Vehicle Battery Management Systems, prioritizing both efficiency and safety.