An advanced wind speed multi-step ahead forecasting approach with characteristic component analysis

Abstract

Accurate wind speed prediction provides essential information for system operation and management with wind power integration. Most existing prediction methods chose to model original data directly without considering the inherent characteristics of wind speed time series. However, the properties of characteristic components can offer better information for prediction. This paper proposes to extract the characteristic components of original data using the ensemble empirical mode decomposition and sample entropy techniques. For the multi-step ahead forecasting of characteristic components, a multiple-input multiple-output (MIMO) based extreme learning machine model is constructed. The effectiveness of the proposed approach is illustrated by applying it to two real wind farm datasets. The extracted characteristic components are shown to be of much lower complexity and the MIMO strategy is found to be better than the common iterated strategy and direct strategy. Moreover, compared to the existing neural networks based methods, the proposed approach has been demonstrated to be a more effective method in both prediction accuracy and computational cost.