Voltage Sag Estimation for Distribution Systems Using Linear Machine Learning Models

Abstract

This paper proposes a voltage sag estimation approach based on linear machine learning models. The proposed approach estimates the voltage sag profile on distributions systems with limited monitored buses. The approach takes advantage of the priory knowledge about the linearity of the problem to train and compare three linear models: Least Squares (LS), Ridge Regression (RR), and Absolute Shrinkage and Selection Operator (LASSO). The approach is tested on the IEEE-34-bus distribution system, and the performance of models is validated through the Mean Square Error (MSE). The results show that the proposed linear machine learning models capture the internal relationships of the problem and estimate the voltage sag with high accuracy on test data.