Selection of most relevant input parameters using WEKA for artificial neural network based solar radiation prediction models

Abstract

Abstract The prediction of solar radiation is important for several applications in renewable energy research. Solar radiation is predicted by a number of solar radiation models both conventional and Artificial Neural Network (ANN) based models. There are a number of meteorological and geographical variables which affect solar radiation prediction, so identification of suitable variables for accurate solar radiation prediction is an important research area. With this main objective, Waikato Environment for Knowledge Analysis (WEKA) software is applied to 26 Indian locations having different climatic conditions to find most influencing input parameters for solar radiation prediction in ANN models. The input parameters identified are latitude, longitude, temperature, maximum temperature, minimum temperature, altitude and sunshine hours for different cities of India. In order to check the prediction accuracy using the identified parameters, three Artificial Neural Network (ANN) models are developed (ANN-1, ANN-2 and ANN-3). The maximum MAPE for ANN-1, ANN-2 and ANN-3 models are found to be 20.12%, 6.89% and 9.04% respectively, showing 13.23% improved prediction accuracy of the ANN-2 model which utilizes temperature, maximum temperature, minimum temperature, height above sea level and sunshine hours as input variables in comparison to the ANN-1 model. The WEKA identifies temperature, maximum temperature, minimum temperature, altitude and sunshine hours as the most relevant input variables and latitude, longitude as the least influencing variables in solar radiation prediction. The methodology is also used to identify the solar energy potential of Western Himalayan state of Himachal Pradesh, India. The results show good solar potential with yearly solar radiation variation as 3.59–5.38 kWh/m2/day for a large number of solar applications including solar power generation in this region.