Online Gaussian Process Regression for Short-term Probabilistic Interval Load Prediction

We propose a hybrid probabilistic interval prediction method for short-term load forecasting. The method combines K-means clustering based feature selection approaches and online Gaussian processes regression(OGPR) to generate better prediction results. The K-means clustering algorithm based feature selection are used to select the most relevant features during a dynamical process to better capture the load characters along with time. OGRP, includes dynamically updating the hyper-parameters and training sample sets as two key features, is served as a forecasting engine to carry out load probability interval prediction. The load data from Queensland market, Australia is used to validate the model proposed. The comparative results show that the proposed approach can obtain higher quality prediction interval.