Optimization of Substation Transformer Tap Adjustment Control Using Digital Twin Database

In distribution networks where it is challenging to directly measure load-side voltage, frequent voltage fluctuations often occur, leading to frequent transformer tap adjustments. These frequent adjustments can shorten transformer lifespan, increase maintenance costs, and compromise the stability of the entire distribution system. This study proposes a machine learning-based method to accurately predict load-side voltage and derive optimal tap adjustments. By doing so, it aims to improve voltage stability and reduce unnecessary tap changes. The distribution network was modeled using the Real-Time Digital Simulator (RTDS) as a digital twin, and OpenDSS was used to simulate various load scenarios and generate data. Machine learning models, including Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), Random Forest, and XGBoost, were trained using this data. Real-time data management was handled by Microsoft SQL Server, ensuring systematic and secure data handling. Among the evaluated models, XGBoost demonstrated the best predictive performance, excelling at handling large datasets and preventing overfitting through regularization. The proposed method is expected to minimize unnecessary tap changes and significantly enhance both voltage stability and the efficiency of power systems.