A new fine-grained classification strategy for solar daily radiation patterns

This paper deals with classification of solar radiation daily patterns into four classes, referred to as clear sky, intermittent clear sky, completely cloud sky and intermittent cloud sky, by using an original features-based classification strategy. The problem is relevant both for analysis and modeling purposes of this kind of time series. An original pair of indices is introduced, referred to as the area ratio Ar and intermittency I. Extraction of these features from solar radiation time series is based on an original strategy, based on the so-called Typical Day, which allows the estimation of the solar radiation that is expected to be measured in a given recording site, avoiding the use of complicate expressions requiring solar altitude, albedo, atmospheric transparency and cloudiness. It is shown that the proposed features-based classification outperforms a traditional neural network classifier which operates on the high dimensional solar radiation patterns.