Chapter 12 - Support vector machine model for multistep wind speed forecasting

Abstract

This chapter aims to design and evaluate data-driven models based on a hybrid complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) technique and support vector machine model (SVM) to forecast multistep wind speed in Australia. Wind speed forecasting for 6-hourly, daily, and monthly horizons are developed at the four study sites of Albany, Capital, Macarthur, and Woolnorth located in wind belt regions. The outcomes show that the standalone SVM and Volterra models had similar computation efficiency and model performances, while standalone ARIMA had the worst performance for all sites and all timescales. The proposed CEEMDAN–SVM is superior to standalone models and is effectively applied in optimizing the combined model. The CEEMDAN–SVM outperformed the alternative models at all sites for 6-hourly and daily forecasting horizons and outperformed at two sites for a monthly scale. The proposed algorithms were found to be effective in high-precision wind speed forecasting.