Parametric study for the prediction of wind energy potential over the southern part of India using neural network and geographic information system approach

Abstract

Wind energy potential in India has so far not been evaluated state wise. Moreover, the prediction and assessment of wind potential are difficult due to the complexity of its nature. Here, a parametric study is done for the better prediction of wind potential using generalized feed-forward with back-propagation neural networks. Effect of three meteorological parameters (pressure, relative humidity, and temperature) on the wind speed prediction is studied in southern states of India. The meteorological parameters taken here are monthly mean, measured at ground station. These data were obtained at 28 sites over a period of 20 years from the IMD, Pune. Three different architectures of artificial neural network model were designed, trained, and evaluated for the prediction of wind speed. All three models have been optimized for varying neurons in the hidden layer. To evaluate the developed artificial neural network model for test locations, mean absolute percentage error and mean squared error have been calculated. It was found that the model with relative humidity as input parameter and having six neurons in the hidden layer give better prediction of the wind speed. The correlation coefficients were higher than 0.96 and the mean absolute percentage error and mean squared error of all test locations is less than 2.5 and 0.0176, respectively, which show high reliability of the model for the prediction of the wind speed within the region of study. Predicted wind speed has been analyzed and used to create monthly mean maps using geographic information system technology.