PhyMamba Enhances Battery Health Prognostics with Physics Integration

Accurate long-horizon forecasting of battery health from Battery Management System (BMS) signals is critical but challenging due to varying operating conditions and sensor noise. Existing methods often struggle with these complexities and may require intrusive measurements to identify internal aging parameters. A new study proposes PhyMamba, a two-stage physics-modulated Mamba framework designed to overcome these limitations. In its first stage, a lightweight Mamba encoder processes BMS signals to generate a latent representation. This representation is then transformed via an aging parameterization module into physics-informed aging features, all without needing explicit identification of internal aging parameters. The second stage employs a customized Mamba forecasting backbone for multi-cycle prediction. Here, physics principles are tightly integrated to regulate the model's internal temporal updates, ensuring a degradation-consistent evolution. Experiments on three public datasets across multiple forecast horizons show that PhyMamba achieves the best aggregated performance, reducing mean error by 31.8% compared to diverse baselines, while also offering an optimized accuracy-efficiency trade-off for practical deployment.