Solar Resource Estimation Based on Correlation Matrix Response for Indian Geographical Cities

Abstract

Global environmental concerns and gradual escalation of fuel cost linked with conventional energy sources encouraged the use of renewable energy in electric power supply sector greatly. India is blessed with great environmental wealth of solar energy due to its favorable location (40ºS to 40ºN). This research work explores the viability of great solar potential of 14 Indian geographical locations by estimating Global Solar Radiation(GSR) for four summer months using Artificial Neural Network (ANN). Initially, eight parameters are chosen as input data set for ANN from a number of environmental factors influencing GSR,based on their natural dependence on it. But, less correlated inputs as training data set for ANN results lead to more sensitive outputs.So, lately, inputs to the ANN are extracted based on Spearman rank-correlation coefficient, where only positively correlated input factors are considered as the input data set for the ANN to enhance its performance.Spearman rank-correlation coefficient describes the extent of correlation between two variables using a monotonic function by utilizing rank-order of the data regardless of distribution between two data sets.  This makes it suitable not only for discrete and continuous variables but also ordinal variables (data sets including inconsistent values).A multi layered, feed forward, standard ANN model with one hidden layer and five hidden neurons corresponding to least mean square error is considered among various ANN models with different training algorithms, hidden layers and neurons, for the prediction of GSR. It is found that correlation based ANN predominatessimple ANN.