Probabilistic Power Quality Level Forecasting through Quantile Regression Models

Waveform distortions are Power Quality (PQ) disturbances with harmful impacts on power system components, and planning engineers and system operators are deeply interested in new methodologies able to provide a support to limit this impact. The forecasting of PQ indices on voltage and current waveforms is nowadays considered one of the most important tools to provide this support. In this paper, probabilistic forecasting methodologies are proposed, investigated and compared to predict waveform distortion indices with different intervals. The Quantile Regression (QR) model is considered as the underlying forecasting model for predicting the considered PQ index level, and the Principal Component Analysis (PCA) is considered to mitigate the high dimensionality of the forecasting problem that could arise by exploiting data collected by dedicated PQ measurement systems. Numerical applications based on actual data and developed for different lead times give evidence of the suitability of the methodologies and the interest in the obtained results.