Probabilistic short-term wind speed forecasting using a novel ensemble QRNN

Abstract

Neural Networks are widely used as the intelligent tool for short-term wind speed forecasting. Owing to the inherent randomness and intermittency of the wind, it is hard to ensure sufficient forecasting precision when using single conventional Network models to conduct deterministic forecasting. Thanks to the recent achievements of ensemble methods, a novel ensemble Quantile Regression Neural Network (QRNN) based on Wavelet decomposition (WD) and least absolute shrinkage and selection operator (LASSO) for probabilistic short-term wind speed forecasting is developed here. The signal processing technology, feature extraction method and predicting model are ingeniously combined in the proposed ensemble QRNN. More specifically, the WD is utilized to extract the signal features into a collection of easy-to-analyze subseries. LASSO regression is combined with the QRNN to conduct the robust prediction. Finally, the kernel density estimation method is adopted to further enhance the forecasting precision. The forecasting performance between the proposed ensemble QRNN and several benchmark models are evaluated using a piece of field measured 10-min mean wind speed data. Results show that the proposed ensemble model can adequately enhance the performance of short-term wind speed forecasting, which can also quantify and reduce the uncertainty of the prediction results.