Comparison of two multi-step ahead forecasting mechanisms for wind speed based on machine learning models

Accurate wind speed forecasts are important to the realtime optimization of wind farm operation and the scheduling of a power system. In the case of multi-step ahead forecasting, two mechanisms, namely, iterative and direct, are commonly adopted. In this paper, a comprehensive comparison study is presented on the applicability of these two methods, based on the wind speed datasets from three wind farms in China. Three representative machine learning models, linear regression (LR), multi-layer perceptron (MLP) and support vector machine (SVM) are developed, respectively. The results show that neither direct nor iterative forecasting can always outperform each other in terms of all the error measures. But in most cases, the performance of the direct forecasting is better than that of the iterative forecasting, especially when the prediction horizon is large and combined with the non-linear models (MLP or SVM).