由於風速波動,風力發電總是伴隨著一些不確定因素,精確地預測風電出力對有效地電力系統運轉是很重要的。本論文提出了前一小時和前二十四小時的風力發電量預測方法。此方法是基於的複雜深度學習神經網絡技術,即具有雙高斯函數的徑向基函數神經網絡作為激活函數的捲積神經網絡。受模糊邏輯類型二神經網絡的啟發,所提出的捲積神經網絡由完全連接部分中的徑向基函數神經網絡和作為激活的雙高斯函數組成。本文模擬了一組實際的風速和風力發電量,並進行了不同方法之間的比較研究。
Wind power generation is always associated with some uncertainties as a result of the fluctuating value of the wind speed. It is well known that both developed and developing countries are seeking new energy for their economy. Wind power and other natural resources are considered as next-generation energy. They are clean without pollution. They can be renewed without limitations as they are available anywhere. Accurate predictions are important for efficient operation of power systems. This thesis presents a hybrid deep learning neural network approach for 1-hour and 24-hour ahead forecasting. Differently, this method is based on Convolutional Neural Network with a combination of Radial Basis Function Neural Network with Double Gaussian function as activation function (CNN-RBFNN-DGF). The proposed CNN is consist of a Radial Basis Function Neural Network in the fully connected part with the Double Gaussian Function as activation and is inspired by the good output of the Fuzzy Logic Type-2. A set of realistic wind speed and wind power generations measured in a wind farm was used for simulation. The deep learning neural network methods are programmed in Python, TensorFlow, and Keras. The CNN methods in this thesis also been tested for the analysis of their parameters. Comparative studies between different approaches are shown.