Memory-Aware Hybrid Quantum Neural Network for robust state of health estimation of lithium-ion batteries

Accurate State of Health (SoH) estimation of lithium-ion batteries is critical for safe and reliable energy storage management, yet static estimators degrade under low-temperature and irregular operating conditions due to distributional mismatch between training and deployment; hybrid variational quantum models share this vulnerability when conditioned on static cycle features. This study proposes the Memory-Aware Hybrid Quantum Neural Network, a hybrid quantum–classical architecture that replaces static cycle inputs with a 𝑘 = 5 sliding window of recent discharge capacity observations, conditioning a 5-qubit, 45-parameter variational circuit on local degradation history without increasing circuit complexity. Evaluated through a four-way ablation on six NASA Ames PCoE batteries spanning clean 24 °C and noisy 4 °C conditions, the proposed model achieves positive 𝑅 2 on all six batteries, including 𝑅 2 = 0 . 7 9 0 and 𝑅 2 = 0 . 4 3 0 on the two noisy low-temperature cells where all static baselines yield negative 𝑅 2 , reducing average RMSE by 67.6% relative to the static quantum baseline. Cross-laboratory validation on eight Oxford Battery Degradation Dataset cells (Kokam LiCoO 2 /graphite pouch cells cycled at 40 °C) yields positive 𝑅 2 across all cells with six of eight exceeding 𝑅 2 > 0 . 9 9 , confirming generalization across cell format, chemistry, and operating temperature without architectural modification. These results establish memory-aware temporal conditioning as a parameter-free mechanism for strengthening variational quantum models in battery health monitoring applications.