Wind power forecasting based on hourly wind speed data in South Korea using machine learning algorithms

Given that wind farms have high initial investment costs and are not easy to move after installation, the amount of energy that can be produced in the desired installation area needs to be predicted as accurately as possible before installation. Four machine learning algorithms are adopted to predict power production based on the daily wind speed average and standard deviation. The actual power output is calculated from the wind data generated by the numerical weather prediction, and its temporal resolution is 1 hour. The R-square (R²) values of the models range from 0.97 to 0.98 while adopting the average value of daily wind speed as the input data, and it increases by −1 % with the additional input data of the standard deviation of wind speed. The power production is predicted based on the wind data at a relatively lower height of 10 m than the hub height, where the R² value ranges from 0.95 to 0.98. The results could provide the possibility of replacing the wind data measurement process at the hub height by that at a relatively lower height, reducing the cost of wind data measurement.