ABSTRACT
This paper investigates interpretable statistical machine learning approaches for sensorless capacitor voltage estimation in 4T4D STATCOM systems, proposing four methods and comparing their interpretability and accuracy.
PAPER · PDF
Loading PDF...
Key findings
Proposed four interpretable methods for voltage estimation in 4T4D STATCOM systems.
Random Forest achieved the lowest RMSE, while proposed methods showed higher interpretability but higher RMSE.
Feature importance analysis identified accumulated current integral as the most important feature.
Limitations & open questions
Proposed methods have higher RMSE compared to black-box methods.
Future work suggested to enhance feature engineering and validate on real-world hardware.