Daily solar radiation forecasting using historical data and examining three methods

Abstract

In any renewable energy system, knowing about availability of resources is one of the main concern for scientists. In another words, challenge is when these resources are needed to be predicted. Forecasting solar radiation is one of the most noticeable unsettled case. By having a thought of forecasting solar radiation would be easier comparing to other renewable resources, because it is not related to wind speed or weather temperature. However, after this experiment it become more clear that, the amount of receiving radiation is proportional to weather alternative conditions such as movement of clouds, haze and even the density of them. In this paper, three different approaches were examined by using historical data to forecast solar radiation day by day. The three approaches are, exponential smoothing, seasonal forecast and feed forward neural network using MatLab. An important finding based on this experiment is that, forecasting radiation by using historical data is not fitting. Meaning to say that, by numbers alone, result of forecasting is getting close to random rather than being more realistic, because receiving radiation is affected by other aspects. For improving the prediction, a method is suggested in future studies.