| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 99 === With the hiking prices of fossil raw materials, mounting gravity of global warming, and the implementation of the Kyoto Protocol, countries all over the world have worked with great vitality to develop renewable energies in recent years. Of all currently available renewable energies, wind power has received extensive attention thanks to its environmental friendliness, virtually inexhaustible abundance, and relatively simplicity in development. However, though quickly becoming the fastest-growing form of renewable energy, wind power finds its disadvantage in the inherent changeability and instability of wind resources that have caused considerable difficulties in wind farm and grid integration and dispatch of power generators. However, the impacts of this inherent disadvantage on power system operation and dispatch can be minimized with accurate wind power forecasts.
The thesis adopts four stand-alone and develops one hybrid model and applies the concept of segmentation in the theory of optimal stratification to forecast short-term wind power outputs. auto-regressive integrated moving average with extra (ARIMAX), support vector machine (SVM), feed-forward back-propagation neural network (BP-ANN), and adaptive neural fuzzy inference system (ANFIS) are used to construct stand-alone wind power forecasting models respectively while genetic algorithm (GA) is used to construct the hybrid forecasting model. Construction of wind power forecasting models is based on the wind power databases during four different periods at the Zhongtun Wind Farm in Penghu, Taiwan. As demonstrated by the forecasting results, the hybrid model outperforms all stand-alone models in terms of forecast accuracy due to its ability to highlight the distribution of weighted values in different segmentation blocks in one single forecast and to reinforce the overall structure of the forecasting model. The models adopted and developed by the study can also be applied to forecast the outputs of other wind farms and to compare the compatibility of various forecast models with individual wind farms. Research results of the thesis can provide power companies with important reference for expediting economic dispatch.
|
|---|
