Sensitivity analysis based artificial neural network approach for global solar radiation prediction in India

Abstract

The objective of this paper is to build an artificial neural network model to predict global solar radiation (GSR) with improved accuracy using less number of best input parameters selected using sensitivity analysis. In this work, the input parameters used for training the artificial neural network (ANN) models are bright sunshine duration, maximum and minimum temperature, day length, months, extra terrestrial radiation (H0), relative humidity and geographical parameters of the locations namely the latitude and longitude. Sensitivity analysis is used to discover how the output data are influenced by the changeability of the input data. Three ANN models namely T-ANN, S-ANN and TS-ANN are proposed with most suitable input parameters selected using sensitivity analysis. The principle of this feature selection using sensitivity analysis is to improve the prediction accuracy of solar radiation models with less number of inputs. The proposed ANN model is also tested under noisy data and proved that ANN is able to perform reasonably good in GSR prediction on practical applications where the data is affected by noise caused by errors on measuring, fault of data acquisition system, recording problems, and so on.