Multistage Wind-Electric Power Forecast by Using a Combination of Advanced Statistical Methods

Abstract

A multistage advanced statistical method has been proposed for the real-time wind-electric power generation forecast of wind power plants (WPPs) based on a combination of artificial neural network (ANN) and support vector machine (SVM) models. In the first stage, output data of wind speed and wind direction from different numerical weather prediction (NWP) models are chosen among a set of grid points in the neighborhood of each WPP to train the ANN and SVM models. The best grids are then selected from those NWP grid data giving the minimum training error, and used for training and testing the developed wind-electric power forecast models. In the second stage, for each NWP data, ANN and SVM models are applied separately. The forecast errors are corrected by applying model output statistics (MOS) at the third stage. Different 48-h ahead forecasts of wind-electric power are then combined at the fourth stage by appropriate weighting factors to obtain an intermediate 48-h ahead forecast of the electrical power generated from wind. In the final stage, these forecast data are recombined to give an ultimate forecast. The proposed model is tested on 25 WPPs satisfactorily. The performance of the proposed multistage cascaded statistical model is compared with the available benchmark models and actual wind-electric power generation data. It has been shown that the proposed model performs better than the reference models in terms of short-term forecast accuracy, especially for WPPs in complex terrains with a scattered wind regime.