Forecasting of Mid- and Long-Term Wind Power Using Machine Learning and Regression Models

Abstract

Environmental concerns over the past decade have driven the need to harness renewable energy resources, such as wind power generation. Forecasting wind power is beneficial to power utilities; however, predicting wind power generation has proven challenging due to wind speed variability. This paper has used two machine learning algorithms, Gradient Boosting Machine (GBM) and Support Vector Machine (SVM), along with the regression model Multivariate Adaptive Regression Splines (MARS), to predict wind-based power production over medium and long-term time frames. A comparative analysis of each forecasting method is presented with the predictions that account for all features. The critical feature among the independent variables is also determined and used for comparative analysis in this study. The preliminary case study results indicate that the SVM model performs better over other models to a greater extent for substantial uncertainty in dataset but suffers from larger computational run time.