WIND POWER FORECASTING USING MACHINE LEARNING IN BHUTAN

In this research, an approach for predicting wind energy using machine learning has been explored. An indirect method has been adopted. Predicting wind speed at first using the hourly weather data and combining that predicted wind speed with the power curve of considered wind turbine prepared by the companies. This research aims to develop a generalized machine learning based wind power forecasting model for Bhutan. Thus, hourly weather data for the year 2018 and 2019 of 300kW On-grid Wind Farm at Rubesa was used to train the base model. Meanwhile, the trained base model was tested against the weather data sets for the selected sites namely Gaselo and Dagana. A Random Forest Regression machine learning algorithm was used in this research. The developed base model has five input variables which are time, temperature, global horizontal irradiance, relative humidity, and pressure, while the target is wind speed. The R- squared values, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) for the developed base model were found to be 0.88, 0.40 and 0.30 respectively. Energy output in the wind turbine was calculated via the predicted wind speed and power curve prepared by the wind turbine companies. The calculated energy output could shape the considered theoretical power curve. The power curve considered in the present research is 300kW On-grid Wind Farm at Rubesa, Wangdiphodrang.