Data-driven Voltage Estimation in Microgrids: A Comparative Study of Machine Learning Algorithms and Practical Applications

Voltage state estimation is an important technology that improves network observability. A main challenge remains the limited sensing capability of low voltage (LV) feeders and the absence of a unified method to yield accurate, transferable, and fully validated estimates in microgrid environments. The scope of this work is to address this challenge by presenting a data-driven voltage estimation model that leverages machine learning (ML) techniques enhanced with an input feature optimization stage. The obtained benchmarking results demonstrated that the Extreme Gradient Boosting (XGBoost) regressor outperformed all other investigated ML models, hence rendering this model suitable for actual-environment applications. Specifically, the optimally constructed XGBoost estimator exhibited low voltage estimation errors <0.2%, given by the root mean square error (RMSE) relative to the nominal voltage value. Finally, useful information is provided for the replication of voltage state estimators that facilitate improved network observability at distribution systems.