Enhancing Battery Capacity Estimation in Electric Vehicles Through Continuous Domain Adaptation

Electric vehicles (EVs) play a critical role in advancing the adoption of renewable energy and reducing environmental impact. Accurate battery capacity estimation is crucial for ensuring the safety and reliability of EVs. In this paper, we formulate the problem of battery capacity estimation in EVs and propose a novel continuous domain adaptation framework, L-MPOT, to address this challenge. The framework integrates an LSTM-based feature extractor to capture domain-specific temporal features by minimizing mutual information across varying State of Charge (SoC) levels. With enhanced feature representation, our method improves continuous domain adaptation (CDA) and boosts estimation accuracy. Experimental results show that L-MPOT outperforms existing CDA methods, significantly enhancing the efficiency of EV battery monitoring.