|dc.description.abstract||In the power trading market, transmission system operators and other actors buy and sell power related to future production. Power production from wind- and solar farms is affected by rapid weather changes, and producers of this power often have to trade in reaction to the changes. Ongoing actual power production data for wind- and solar farms is published by the transmission system operators. These numbers indicate what volumes these actors have to trade to keep up with the weather changes, and is one of the largest cost drivers in the market. Therefore, the ability to forecast power production is highly relevant in the power trading industry.
In this master thesis we do a case study with focus on wind energy, and the main research task is to predict wind power production. We introduce three models named CCPR, UCPR and CPR-LP, where all are based on a new methodology. The methodology starts out with one or two initial forecasts, in the form of cumulative density functions. The CCPR and UCPR use one initial forecast, and the method proceeds by transforming the initial forecast through a beta transformation function, returning a calibrated final forecast. The CPR-LP uses two initial forecasts, where the methodology beta transforms a weighted sum of these. The parameters which define the beta transformation function are modelled as a function of deterministic forecasts related to the wind power production. We divide our test results into groups, based on these deterministic forecasts. UCPR is performing very well compared to the other models for large deterministic forecasts, and CCPR is performing well for small deterministic forecasts. The CPR-LP model on the other hand is preferable when considering all groups as a whole.||