Edge device compatible optimization considering trade-off between lithium-ion battery charging speed and temperature rise via Grover-Q-learning

Artikel i vetenskaplig tidskrift, 2026

Fast charging and temperature rise regulation in battery systems remain critical challenges for the safe and efficient application of electric vehicles. However, existing optimization methods often rely on large-scale parameter searches, limiting their deployability under real-time and embedded computational constraints. To address this issue, this paper proposes a high-efficiency and lightweight quantum reinforcement learning-based charging optimization method, termed Grover-Q-learning. By introducing an amplitude-amplification-based action selection mechanism inspired by Grover quantum search, the method efficiently concentrates on high-value charging actions while preserving the lightweight structure of classical Q-learning. A simplified electro-thermal coupled model is further adopted to reduce computational overhead. Using experimentally measured ternary lithium-ion battery data, a high-fidelity charging simulation environment is established for policy training and evaluation. Experimental results show that, compared with the conventional CC-CV strategy, the proposed method reduces charging time by 2.6% and peak temperature rise by 2.5%. Compared with classical Q-learning, it achieves 28% faster convergence with reduced training oscillations. Without physical quantum hardware, the method attains a single-step decision latency below 0.93 ms and RAM usage under 53 KB, enabling deployment on resource-constrained ARM Cortex-M–based BMS platforms.