Artificial Neural Network based Prediction of Solar Radiation for Indian Stations

Abstract

ABSTRACT The Artificial Neural Network (ANN) fitting tool is used for the prediction of solar radiation. Solar radiation data from 12 Indian stations with different climatic conditions are used for training and testing the ANN. The Levenberg-Marquard (LM) algorithm is used in this analysis. The results of ANN model are compared with measured data on the basis of root mean square error (RMSE) and mean bias error (MBE). It is found that RMSE in the ANN model varies 0.0486–3.562 for Indian region. Keywords Solar radiation, Levenberg-Marquard (LM) algorithm , Artificial neural network. knowledge. ANN has 1. INTRODUCTION Solar radiation data are required for a number of solar thermal and Solar photovoltaic applications like solar power generation, solar heating, cooking, drying and solar passive design of buildings [1–4]. The measured solar radiation data are not available for most of the sites due to high cost, maintenance of the measuring instruments. As such, various empirical models have been used to predict monthly mean daily solar radiation all over the world [5–9]. The Artificial neural networks (ANNs) are used to solve a number of scientific problems. It has the capability to approximate any continuous non-linear function to arbitrary accuracy [10]. A multi-layer feed-forward neural network can approximate a continuous function due to its robustness, parallel architecture and fault tolerance capability. In past years, ANN models are used by a number of researchers to estimate solar radiation [11–15] and concluded that ANN model are proven to be superior to other empirical regression models. Reddy [11] used Radial Basis Functions (RBF) and Multilayer Perceptron (MLP) models to predict solar radiation using data from eight stations in Oman. So¨zen [12] determined the solar-energy potential in Turkey using artificial neural networks. A Rehman and Mohandes [15] estimated function asdaily global solar radiation for Abha city in Saudi Arabia by taking air temperature, number of day and relative humidity as inputs to neural networks. The results obtained indicate that the mean absolute percentage error (MAPE) is 4.49%. M.A. Behrang et. al [17] predicted DATAdaily global solar radiation for Dezful city in Iran by using different ANN techniques based on different combination of meteorological variables (day of the year, daily mean air temperature, relative humidity, sunshine hours, evaporation and wind speed). The MAPE for the Multilayer Perceptron (MLP) network is 5.21% while this value is 5.56% for Radial Basis Function (RBF) network. Mohandes et al. [16] has used RBF network for modeling solar radiation and compares its performance with MLP model by using latitude, longitude, altitude and sunshine duration as input parameters. The average MAPE for the MLP network is 12.6 and the average MAPE for RBF networks is 10.1. In the present study, an ANN model is developed which can be used to predict solar radiation at any given location in India.