Measurement Error Situation Awareness Method for Electronic Voltage Transformers Based on Physics-Informed Machine Learning

Capacitive voltage divider-type electronic voltage transformers (EVTs) are widely used measurement devices in high-voltage power systems, where their measurement errors directly affect smart grids safety and economic operation. To address the issues of low robustness and insufficient accuracy in existing data-driven or single-physics-driven measurement error perception models, this paper proposes a state awareness method for capacitive voltage divider-type EVT measurement errors based on physics-informed machine learning (PIML). First, considering the generalization limitations of conventional temporal features, a physics-informed augmentation mechanism is designed to construct deep feature parameters that influence electronic voltage transformer measurement errors. Second, based on the physical processes governing EVT measurement error variations, physical loss constraints are formulated to establish a physics-informed machine learning model. Subsequently, an enhanced Shapley Additive Explanations (SHAP) interpretation mechanism is proposed to provide global explanations for these deep feature parameters, culminating in the development of an EVT measurement error situational awareness model. Actual engineering data demonstrate that the proposed situational awareness method for electronic voltage transformer measurement errors achieves a prediction accuracy of 0.99, representing an average improvement of 3.12% over existing models. The ablation experiment also shows that the model integrating physical information and physical constraints can improve accuracy by at least 2.77%. The explanation mechanism enables visualization of feature contribution and importance on a ‘single-instance-to-time-period’ basis. The explanation results indicate that the thermal-load coupling effect of EVTs warrants particular attention.