Mapping of solar energy potential in Fiji using an artificial neural network approach.

The concerned stakeholders have been pursuing renewable energy seriously due to its overwhelming benefits. Countries that receive less solar radiation are not lagging behind as they are working to optimize the available radiation let alone of countries that receive sufficient solar radiation over long durations such as Fiji. In view of the abundancy of this energy in Fiji, the country has been working intensely on tapping the full potential of this energy, thus proposed that by 2030; more than 50% of its energy will come from renewable energy. The accurate estimation of global solar radiation determines the reliability of performance evaluation of solar energy systems. Therefore, the key interest of this study is in respect of accurate mapping of solar radiation to aid reliable solar energy design especially in siting and sizing of photovoltaic power systems. In the light of this, this work modelled solar radiation on the earth of Fiji from common meteorological and geographical data in all locations in Fiji using Artificial Neural Networks (ANN). There are different configurations of ANN but in this study, Levenberg-Marquardt (LM) and Scaled Conjugate Gradient (SCG) were selected as the learning algorithms due to the data size, speed of computation and the success of these algorithms in solar radiation modelling. Similarly, a tangent sigmoid transfer function was used in the network. In total, twelve different configurations of ANN were considered and the best configuration was selected to predict the solar radiation potential in Fiji. Since ANN requires input data to train the network, meteorological data covering 36 years (1984-2019) and geographical data from NASA database were supplied to the network. All the locations considered were distributed evenly throughout Fiji and thus covered all the four regions and 14 provinces in Fiji. The geographical and meteorological data used to train the network are month, latitude, longitude, altitude, mean temperature, relative humidity, precipitation and solar radiation. The mean squared error of 0.118838 and correlation coefficient of 0.9402 were obtained between the ANN predicted and measured solar radiation for the entire dataset. These correlation coefficients and mean squared error showed that ANN model of solar radiation in Fiji is satisfactory and thus can be used as an alternative where solar radiation data are not available. Similarly, the network produced satisfactory solar radiation result for the locations where there are no solar radiation data. To ease solar radiation assessment of all places in Fiji, the iso-lines of the solar radiation were presented in the form of monthly maps. It is believed that this prediction will aid energy stakeholders in making best decision concerning solar energy potential in Fiji thus boosting optimal utilization of the scarce resource.