Daily array yield prediction of grid-interactive photovoltaic plant using relief attribute evaluator based Radial Basis Function Neural Network

The progress of renewable energy is becoming an important source for meeting energy requirements of India. With the plentiful availability of solar energy, Grid-Interactive Solar Photovoltaic (GISPV) plants are becoming important in most part of the country. Due to varying climatic condition it is important to predict the daily array yield of GISPV plant. In this paper, new algorithm relief attribute evaluator is used to find most influencing variables from solar radiation (SR) and back surface module temperature (BSMT) to predict the daily array yield of 190-kWp GISPV power plant using Radial Basis Function Neural Network (RBFNN) for 26 different Indian cities. The ranks given by relief attribute evaluator are 0.00775 for SR and 0.00139 for BSMT, showing SR is relevant variables for daily array yield prediction. For analysis two Radial Basis Function Neural Network (RBFNN-1, RBFNN-2) models are developed for the prediction of daily array yield for the 190-kWp GISPV power plant. SR and BSMT are used as input parameters for the RBFNN-1 model and SR is used as input for the RBFNN-2 model. The root mean square error (RMSE) for RBFNN-1 is 0.2642 kWh/kWp/day and for RBFNN-2 the RMSE is 0.2910 kWh/kWp/day. The results comparison shows that RBFNN predicts daily array yield better than the polynomial regression model. The RBFNN-2 model is used to predict daily array yield for 26 different Indian cities and it is found that total daily average daily array yield varies from 3.50 kWh/kWp/day to 7.94 kWh/kWp/day which can be used to estimate power production for solar photovoltaic power plants. The predicted total array yield by RBFNN-2 model is validated with calculated value and RMSE is found to be 2.295 kWh/kWp/day showing RBFNN-2 can be used to predict daily array yield for different sites in India.