Short-Term Wind Power Prediction Approach Based On Bayesian Optimization and Ensemble Learning

In wind energy studies, predicting the short-term generation amount for power plants and determining production offer to be placed on market play an important role. this study hourly estimation of a turbine located in Turkey with installed 3600 kW has been made. Estimation results were evaluated seasonal annual basis. New hybrid models have developed prediction, consisting Bayesian Optimization (BO), Support Vector Regression (SVR), Gaussian Process (GPR), Decision Tree (DT), stacking, bagging algorithms. proposed prediction approach, it is aimed reduce errors by combining different regression algorithms BO method ensemble Unlike other was used first time hyperparameter selection selected as basic learner study. optimized decision tree (BO-DT) lowest error values among base learners, gaussian process (BO-GPR) combined stacking. The efficiency learning measured statistical measurement methods Normalized Absolute Mean Error (NMAE), Root Squares (NRMSE), determination coefficient (R 2 ). According results, created BO-DT took average NRMSE, NMAE, R criteria 11.045%, 4.880%, 0.899, respectively, model best performance terms both results.