Electronic compensation networks used in high–accuracy inductive voltage dividers (IVDs) can introduce temperature– and aging–dependent drift sources, complicating long–term uncertainty evaluation. This paper presents a 1 kV, 10–tap, compensation–free three–stage IVD that combines multi–stage excitation, a closed shielded core, and equipotential coaxial–cable windings to suppress excitation, magnetic–coupling, and capacitive errors. An integrated reference winding is embedded in the third–stage coaxial–cable winding, enabling a simplified bootstrap self–calibration procedure without an additional auxiliary IVD. At 50 Hz, the calibrated ratio error and phase displacement are within 0.08 μV/V and 0.08 μrad, respectively, for all taps, with expanded uncertainties of 0.04 μV/V and 0.06 μrad. Wideband measurements up to 3 kHz further show that the proposed IVD can serve as a voltage–ratio reference for harmonic voltage measurement and calibration.
With the rapid development of electric vehicles and large-scale electrochemical energy storage systems, accurate State of Charge (SOC) estimation for lithium-ion batteries faces significant challenges under complex and dynamic operating conditions, often resulting in insufficient precision and robustness. Although recent data-driven approaches, such as Long Short-Term Memory (LSTM) networks, have shown promising results, they still suffer from underutilization of physical prior knowledge and limitations in capturing multi-scale temporal features. To address these issues, this paper proposes a novel physics-informed data-driven SOC estimation framework that integrates physical features with a Multi-Scale Temporal Feature Ensemble Network (MTFEN). First, based on the Thevenin equivalent circuit model and Forgetting Factor Recursive Least Squares (FFRLS), the open-circuit voltage is extracted as a dominant physical feature and validated through Pearson Correlation Coefficient (PCC) and Maximal Information Coefficient (MIC) analysis. This feature is fused with raw measurements (current, voltage, and temperature) to form an augmented input vector. The MTFEN is then constructed by parallelly integrating a Temporal Convolutional Network (TCN), Bidirectional Long Short-Term Memory (BiLSTM), and Multihead Probsparse Attention (MPSA) to simultaneously capture local dynamic features, long-term dependencies, and global temporal correlations. Finally, an L2-regularized Kernel Ridge Regression (KRR) meta-learner is employed for end-to-end fusion estimation. Experimental results on LG 18650-HG2 batteries under various driving cycles demonstrate that the proposed method achieves a minimum Root Mean Square Error (RMSE) of 0.7272% and a Mean Absolute Error (MAE) of 0.6190%, significantly outperforming several state-of-the-art baselines. The proposed framework provides a promising technical solution for high-precision and robust online SOC estimation in battery energy management systems.