Enhancing lithium-ion battery prognostics in dynamic environments via quantum-augmented feature extraction and transformer framework

Accurate State of Health (SOH) estimation for Lithium-ion batteries under dynamic discharging profiles remains a critical engineering challenge, as conventional prognostic methods predominantly rely on idealized constant current conditions, limiting their applicability in variable operational environments. To address this, a Quantum-Augmented Transformer (Q-Transformer) architecture is proposed. This framework integrates a Quantum Neural Network (QNN) preprocessing stage to exploit quantum superposition for robust feature extraction, coupled with a Transformer encoder utilizing cross-attention mechanisms to capture long-range temporal dependencies. The primary contribution lies in the development of a quantum-enhanced attention mechanism for degradation modeling, establishing a novel paradigm for leveraging quantum computational advantages in time-series prognostics. Experimental validation using the NASA battery datasets demonstrates the superior performance of our proposed architecture. Compared with BiLSTM, BiGRU, CNN and standard Transformer baselines evaluated under identical conditions, the proposed model lowers the average RMSE by 65.5% and the average MAE by 68.8% relative to the strongest classical baseline, and delivers a robust average R 2 score of 0.986.