Prediction of Direct Normal Irradiance using a Novel Multi-model Algorithm

The prediction of solar radiation is a hot issue in the field of solar energy and time series data analysis. Due to the rapidly development of photovoltaic (PV), most related works just focus on the Global Horizontal Irradiance (GHI), while the Direct Normal Irradiance (DNI) is more important for another new solar energy technology: solar thermal. The DNI is related to GHI, however, the they are not corresponding one by one. Furthermore, the change rules of DNI is very various in different meteorological conditions. For example, in the cloudy day the DNI is more likely to change sharply than the clear day. Thus, a method to deal with these different weather conditions is needed. This paper proposes a multi-model algorithm to matching this new requirement, by building a complex network, we describe the relationship between the different sample, and then using a community detection method based on modularity optimization, we cluster the samples into different classes. Finally, we deal with these classes by different extreme learning machine models. The experimental result based on the real data proven this multi-model algorithm is useful.