Probabilistic Wind Speed Forecasting based on Minimal Gated Unit and Quantile Regression

Abstract

High-quality wind speed forecasting (WSF) is of great significance to the power system planning and operation. In this paper, a hybrid method based on Minimal Gated Unit (MGU) and Quantile Regression (QR) is proposed for 2-hour WSF. Firstly, abnormal data is filtered by using the operating mode of SCADA system and Linear Interpolation algorithm is used for missing data imputation. Secondly, this paper embeds conditional quantile as an internal unit of MGU network. We estimate the parameters of MGU network under different quantile conditions, and calculate output under each quantile condition to obtain probabilistic WSF. At last, SCADA data collected from three wind turbines is applied to test the model performance. Both point and interval evaluation criteria are applied to evaluate the performance of models. The results show that the proposed model can obtain multi-step WSF with both point and interval prediction, compared with typical methods, it has higher accuracy in interval predictions and lower computation cost.

Multi‐step wind speed and wind power forecasting using variational momentum factor and deep learning based intelligent neural network models

“…A new data‐driven ensemble method for numerical wind speed forecasting has been proposed in a work by Zhao et al 40 A new multistep forecasting strategy is proposed based on a wavelet decomposition pre‐processing module, nonlinear autoregressive artificial neural network and nonlinear autoregressive exogenous artificial neural network composite prediction module, and support vector machine classifier post‐processing module 41 . On reviewing the previous studies made in multi‐step wind speed forecasting as presented above and also in the works 42–67 carried out by numerous researchers in this field, the following limitations are inferred and are as elucidated below. Occurrences of global and local optima problems, leading to the saturated learning process of the machine learning algorithms 14–23,27–34 …”

“…Increased mean square error (MSE) proving still the forecasting has to be carried out 48–52 Overall accuracy shall be increased 53–60 Few techniques involve huge computations 24–26 …”

“…[48][49][50][51][52] • Overall accuracy shall be increased. [53][54][55][56][57][58][59][60] • Few techniques involve huge computations. [24][25][26] • Difficult to maintain the stability of the algorithms.…”