A Master of Science thesis in Electrical Engineering by Fatemeh Marzbani entitled, "Short-term Wind Power Prediction," submitted in June 2014. Thesis advisor is Dr. Ahmed Osman and thesis co-advisors are Dr. Mohamed Hassan and Dr. Taha Landolsi. Available are both soft and hard copies of the thesis.
Environmental considerations in addition to energy crises have forced many countries to consider alternative energy sources; renewable energies are known as the best alternatives. Among renewable energies, wind power is the most promising energy source. The chaotic nature of the wind is a major challenge against the integration of wind power into grids. Integration of wind power results in several problems due to the fluctuations inherent in wind power, such as power quality, stability, and dispatch issues. The prediction accuracy of wind power affects its integration into power systems. Several wind power forecasting techniques have been proposed and developed. However, not all of them are able to provide sufficient accuracy. The main contribution of this thesis is to provide accurate short-term wind power prediction. A simple, yet effective adaptiveparameter regression model is developed. Specifically, the proposed approach uses a window of previous observations to obtain the model parameters that minimizes the prediction error. Regression-based models are affected by measurement errors. Thus, other models with the capability of moderating the impact of measurement errors are needed. In order to cope with such errors, two hybrid grey-based short-term wind power prediction techniques are proposed: GM(1,1)-ARMA and GM(1,1)-NARnet. These techniques are combined with ARMA models and Nonlinear Auto Regressive Neural Network (NARnet) models, respectively. GM(1,1)-ARMA and GM(1,1)-NARnet are applied to wind power data and the obtained results are compared with those obtained from ARMA, the traditional grey model, as well as the persistent model. The efficiency of both of the proposed techniques is confirmed. In contrast to the GM(1,1)-ARMA model, the GM(1,1)-NARnet model utilizes the nonlinear components of wind power during the forecasting procedure which results in more accurate prediction.