Predicting Solar Diffuse and Direct Components Using Deep Neural Networks

Abstract

With the increased reliance on solar energy generation, accurate solar irradiance data is important for calculating the value of existing and planned generation. Historical and real-time data on global horizontal irradiance (GHI) is widely available, however; the diffuse horizontal irradiance (DHI) and direct normal irradiance (DNI) components of GHI value are not. Solar photovoltaic and concentrated solar power production models rely on these two components for precise calculations of generator output. In this research, we used a deep neural network model to capture the sparse connections and complex interactions between many different meteorological parameters to accurately calculate DHI and DNI from GHI. Our model proved to outperform other previous parametric and decomposition models in the literature. It is also generalizable to different locations. We have released our model as an open source library here:https://github.com/tpt5cu/solarIrradiancePredictor.