Very short-term probabilistic wind power prediction using sparse machine learning and nonparametric density estimation algorithms

In this paper, a sparse machine learning technique is applied to predict the next-hour wind power. The hourly wind power prediction values within a few future hours can be obtained by meteorological/ physical methods, and such values are often broadcast and available for many wind generators. Our model takes into consideration those available forecast values, together with the real-time observations of the past hours, as well as the values in all the power generators in nearby locations. Such a model is consisted of features of high dimensions, and is solved by the sparse technique. We demonstrate our method using the realistic wind power data that belongs to the IEEE 118-bus test system named NREL118. The modeling result shows that our approach leads to better prediction accuracy comparing to several other competing methods, and our results improves from the broadcast values obtained by meteorological/physical methods. Apart from that, we apply a novel nonparametric density estimation approach to give the probabilistic band of prediction, which is demonstrated by the 25% and 75% confidence interval of the prediction. The coverage rate is compared with that yielded from quantile regression. (c) 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

This paper applies a sparse machine learning technique to predict next-hour wind power, considering forecast values, real-time observations, and neighboring power generators. The model outperforms other methods and improves upon broadcast values obtained from meteorological/physical methods. Additionally, a novel nonparametric density estimation approach is used to provide probabilistic prediction bands.