LAUSR

Wind power generation is one of the technologies of electric production which still in development in Brazil, however, it already has a great penetration in the national energy matrix, representing 13.98% of the national energy consumption in Brazil. Due to the high level of uncertainty and the chaotic fluctuations in wind speed, predictions of wind energy with high accuracy is a challenge. In this context a stacking ensemble (STACK) model is proposed to forecast the wind power generation of a turbine in a wind farm at Parazinho, RN Brazil. The proposed model combines four different algorithms as base-learners, such as, eXtreme Gradient Boosting (xgBoost), Support Vector Machine for regression with Linear Kernel (SVRLinear), Multi-Layer Perceptron with multiple layers (MLP) and K-Nearest Neighbors (K-NN), and one algorithm as metalearner Support Vector Machine for regression with Radial Basis Function Kernel (SVR-RBF). To access the performance of adopted methodology, the results of STACK are compared with the results of the base-learners. Four performance measure criteria, as well as statistical tests are adopted. As results, STACK reached better results in all performance measures. Indeed, STACK and SVR-Linear are statistically equals. According to these results, applying the STACK proposed model indeed improved the forecasting when comparing with the other algorithms tested individually. Keywords—Wind energy, forecasting, time series, machine learning, stacking ensemble.