| Abstract [eng] |
The aim of the work is to investigate the electricity production forecasts, hour and a few hours ahead, for the existing wind farm, using electricity production existing and history information, including LHMT wind speed forecast influence on electricity production forecasts. In the project using artificial neural networks, random forests and linear regression electricity production forecast is made an hour and a few hours ahead. Firstly, only data is used for forecasting. The forecasting methods are trained with wind farm electricity production history, the production, which was an hour ago, two, three hours and more. Trained forecasting methods with past wind farm electricity production, predicts result – the next hour electricity production. RMSE (Root Mean Square Error) error is calculated between predicted and actual wind farm electricity production values and forecasting methods are compared. In the next stage, forecasting methods are trained with the past wind farm production (like in a first part) together with wind speed and its direction forecast. Wind forecast is calculated for a wind farm area 100 meters high and provided by Lithuanian hydro meteorological service under the Ministry of the Environment. Trained forecasting methods predicts the next hour electricity production from the past production and next hour wind forecast. As in the first part, RMSE error is calculated between predicted and actual wind farm electric power production and forecasting methods compared again. Compared RMSE errors will be an answer, that influence does the wind forecast to predicting wind farm electricity production. The work includes not only compared forecasting method, but also how well each of the methods predicts the electricity production at the appropriate wind speed, or the corresponding hour of the wind forecast. |
