Yeni bir veri madenciliği tabanlı bölgesel rüzgar gücü tahmini için yukarı ölçekleme algoritması

Together with the increasing need for energy, the importance of renewable energy sources has been increasing day by day. Although wind is one of the most important alternative energy sources due to its high potential, it is not a stable source since it depends on the weather conditions. So, in order to include the power produced by the wind into electricity grid with planned manner, it must be predicted accurately beforehand. To produce a reliable wind power forecast, getting a Wind Power Plant's (WPP) power data in real time and constructing the model with past production values is an expected and optimal situation. However, this situation cannot be applicable for all WPPs in the country due to the difficulties on getting the power data of WPP in real time. Hence there is a need for accurate upscaling algorithm for generating the power forecast of such WPPs and producing a regional power forecast for a given region. These forecasts are especially so crucial for the management and planning of the electricity grid. It is very important that the system operators who manage the energy flow in the country use their energy resources correctly by using these power forecasts day in advance. In this thesis, six different statistical based models are developed to solve the regional wind power estimation problem and the results obtained are compared with some well known statistical based machine learning models in the literature such as ANN and SVM. The most important contribution of the proposed method is that it produces regional power forecasts while also generating power estimates for wind power plants in the system for which we do not have their historical power data. Another advantage of the method is that the wind potential of a candidate point where a wind farm is planned to be established can be determined by this method. In this thesis, this feature of the model is tested for 16 different candidate points from four different cities of country. In addition, regional forecasts results are tested for 9 different load distribution regions and performance results of the models are analyzed.