Global solar radiation prediction using recurrent neural networks

Renewable energy sources derive enormous energy from the sun’s radiation. Global Solar Radiation prediction is essential in Photo Voltaic power plants for efficient sizing and improving the performance of these systems. Some computational Intelligence methods are used in time series prediction of solar radiation based on the statistical data. A number of neural network models like Radial Basis function (RBF) and Multilayer perception (MLP) were used and these are all forward prediction methods which may result in inaccuracy of prediction. Here, Recurrent Neural Network (RNN), in which a feedback from the output layer is given as input to one of the hidden layers has been used. Input variables used for prediction are Day of the month, daily mean air temperature, Relative humidity, Air Pressure and Solar azimuth angle.RNN is being trained using Particle Swarm Optimization (PSO) and Evolutionary algorithm (EA).EA is stochastic search and optimization heuristics derived from evolutionary theory.PSO is an optimization based technique used for solving non-linear and multidimensional problems. Also, performance of these algorithms is compared by calculating Root Mean Square Error (RMSE).