Week Ahead Wind Power Forecast Using A Time-Series Forecasting Model

Wind farm power production is unreliable due to unpredictable weather. Wind power forecasting comes into the picture to balance supply and demand by evacuating maximum power to the grid. Several existing studies focus on forecasting for a short duration mostly three days, but it is expected to forecast a week ahead to schedule, generate a maximum of planned power and avoid penalties. This paper proposes week ahead wind power forecast based on a time-series forecasting model using ensemble learning such as Random Forest Regressor, XGBoostRegressor, CatBoostRegressor and other machine learning algorithms such as support vector regression, and linear regression and calculates the accuracy of all algorithms to make an accurate week ahead forecast. The dataset collected on a wind farm for a period of two years is taken and used for training and testing the time-series forecasting models. The dataset contains weather parameters like temperature, humidity, and pressure, and wind parameters like wind direction and speed at the wheel hub height. The best week-ahead forecasting algorithms are found to be Random Forest Regressor and CatBoostRegressor. Thus, the forecasting results will be useful in maximizing revenue and avoiding penalties for power producers, while simultaneously benefitting the power utility.