Vinsamlegast notið þetta auðkenni þegar þið vitnið til verksins eða tengið í það: http://hdl.handle.net/1946/42891
Power production oversight is a crucial part in decreasing waste and increasing the utilization of the resource of the power plant. The risk of selling electricity from a hydro power plant lies in both resource waste and unforeseen cost due to an increase or decrease in power production resulting in having to buy or having bought excess or insufficient amount of electricity. The objectives of this thesis is to create two forecasts, one for day ahead forecast and another for a month ahead forecast that can accurately predict the power production of a small hydro power plant in the north of Iceland. Specifically to increase the accuracy of the forecasting of the power plant to minimize the waste of the resource and the financial risk associated with electricity production. Four prediction models were created and trained on weather data and production data from 2014-2018 and predicted the production for 2019-2021. The four prediction models are composed of two variations of tree based models, namely random forest and one optimized distributed gradient boosting model (XGBoost). One machine learning model, Long-short-term-memory recurrent neural network (LSTM) and one stacked ensemble method based on variations of the other three models. The results were measured using the coefficient of determination (R2) and root mean squared error (RMSE). The four models surpassed the forecasting method which was used to predict the power production in 2019-2021. With a larger margin of improvements for the month ahead forecast. The performance margin for the day ahead forecast was not sufficient for a recommended implementation into the day to day forecasting of the power plant, the month ahead forecast margin was however enough to recommend that the LSTM with a 24 month rolling window will be tested for the forecasting purposes of the power plant.
|Power production forecasting for a small hydro power - Steinar Grettisson - 2022.pdf||14.31 MB||Opinn||Heildartexti||Skoða/Opna|