Wind power forecasting without using historical data

Wind power, i.e., electrical energy produced making use of the wind resource, is a potential alternative to the conventional energy technologies. But its fluctuating nature requires a uniform approach to manage energy from all sources. An accurate forecast of the wind power for the forthcoming hours is crucial, so that proper planning and scheduling of the conventional generation units can be performed. The main drawback with the existing statistical models is that they require a historical dataset for forecasting. This is a hurdle for newly started farms due to the lack of historical data set. This paper models a self-adaptive Artificial Neural Networks to effectively forecast the wind power without using the historical data. The algorithms will converge after few days of operation. Neural network models such as Multi-Layer Perceptron (MLP), Radial Basis Functions (RBF), and Support Vector Machine (SVM) etc. have better prediction capabilities at the expense of high computational complexity due to the presence of hidden layer. To reduce the computational cost, this paper uses models such as Functional Link Artificial Neural Network (FLANN), Legendre Neural Network (LeNN) and Chebyshev Neural Network (ChNN). Finally these models are compared.