A Three-Layer Hybrid Model for Wind Power Prediction
Jian Gao, Panitarn Chongfuangprinya, Yanzhu Ye, Bo Yang
Abstract
Accurate wind power prediction (WPP) is important for stable operation of power systems. However, the intermittent nature and high variability of wind causes many challenges. This paper proposes a three-layer WPP model considering the data from historical power measurements and numerical weather prediction (NWP) systems. The first layer uses a linear model to learn the wind power generation equation. The second layer includes several non-linear models to learn the seasonality and the inertia of wind turbines. The third layer uses stacked regression to learn a hybrid combination of predictors in the previous layer. We compared the proposed approach against the state-of-the-art algorithm as well as two neural network models. Experiment results show that our approach has the best performance.