An ANFIS-based Forecasting for Solar Radiation Data from Sunshine Duration and Ambient Temperature

Total solar radiation data is considered the most important parameter in renewable energy application, particularly for sizing photovoltaic (PV) system. However, these data are not always available particularly in isolated sites due to the non-availability of the meteorological stations in these sites. Fortunately, the mean temperature and the sunshine duration are always available because we can measure these parameters using a simple instrument. This paper introduces a new approach for predicting and modelling of total solar radiation data from only the mean sunshine duration and air temperature using an adaptive neuro-fuzzy inference scheme (ANFIS). This technique is suitable for time series prediction. In this study a database of daily sunshine duration, ambient temperature and total solar radiation data, which had been recorded for 10-years (1981-1990) are used. An ANFIS model has been trained based on 9-years known data from the database a set of 365 values of solar radiation data is used for testing the model. In this way, the network was trained to accept and even handle a number of unusual cases. Known data were subsequently used to investigate the accuracy of the prediction. Subsequently, the unknown validation data set produced very accurate estimation, with the mean relative error (REM) not exceeding 1% between the actual and predicted data, and the correlation coefficient obtained for the validation data set is 98%. This model has been developed for Algiers location but the methodology can be applied to any geographical area in the world.