NWP Combination Correction Model Based on Variable-weight Stacking Algorithm

Abstract

Wind power forecasting (WPF) is of great significance for the balance of power system integrated the large-scale wind power and the economic benefit of wind farm. The error of WPF comes mainly from the NWP (Numerical Weather Prediction), which is the main input of WPF. In this paper, a combination correction model based on variable-weight stacking integration algorithm is proposed. First, the correlation coefficients between wind turbines (WTs) and their corresponding NWPs are calculated, and the NWP wind speed with the highest correlation is obtained as the model input. Second, the variable-weight stacking correction model is constructed, the sub-model includes multiple linear regression model (MLR), random forest regression model (RFR), BP neural network model (NN), and support vector machine model (SVM). To prevent overfitting, the cross-validation method is used in the sub-model training. The combination of each individual correction model is achieved using a nonlinear and variable-weight strategy. To validate the proposed model, data from a wind farm in North China are used. The results show that the proposed model outperforms the other benchmarks containing four traditional single models using single point data.