Wind power generation forecasting based on multi-model fusion via blending ensemble learning architecture

Because of the intermittency and randomness of wind power generation, constructing an accurate wind power generation forecasting model is of great necessity for stable operation and optimal scheduling of modern power systems. Considering the unsatisfied performance of the single learner model and the diverse learning abilities of different machine learning algorithms, XGBoost model, KNN algorithm, SVM algorithm, and CNN-BiLSTM-Attention neural network are integrated via blending ensemble architecture to construct the multi-model fusion short-term wind power forecasting model. Pearson correlation analysis is applied to reveal the interrelation between meteorological factors and wind power. Additionally, the training samples of base learners are reconstructed for ensuring all data can be utilized. The advantages of each learner are combined co-ordinately via blending ensemble learning framework. Prediction results of ensemble learning model and single learner model are compared in the same scenario. Simulation results indicate that the ensemble learning model can effectively extract potential features of input information and realize higher prediction accuracy. Considering the intermittency and randomness of wind power generation, this article constructs a wind power forecasting model with high accuracy and a robust training process. Because of the unsatisfied performance of the single learner model, XGBoost model, KNN algorithm, SVM algorithm, and CNN-BiLSTM-Attention neural network are integrated via blending ensemble architecture to construct the multi-model fusion short-term wind power forecasting model. In order to avoid the overfitting phenomenon in the training process, the data set is equalized and the model is trained by the cross-validation method. image