Global horizontal irradiance modeling for all sky conditions using an image-pixel approach

Abstract

Ground images with a sky camera have become common to evaluate cloud coverage, aerosols, and energy collection. In parallel, the growth of solar energy has led to an impulse to evaluate and forecast the solar potential in a site before investments, which has increased the importance of solar power measurements. Facing that scenario, this work presents a novel sky camera model that allows to measure the global horizontal irradiance (GHI). Initially, images from a fisheye camera were stored and a pixel-based approach model was created for cloud segmentation. A total of 813 k vectors of features were used as input to the support vector machine for classification (SVC), which yielded a success rate of about 98.6% in accuracy. The Sun’s position was also segmented and an artificial neural network (ANN) regression model for GHI with 17 input features was created based on segmentation of the Sun, clouds, and sky. The training/validation stage of the ANN used 89,964 samples and the test stage reached about 97.4% in Pearson’s correlation. The RMSE was 72.3 W/m2 for GHI and the normalized RMSE, nRMSE, revealed 12.9% for GHI. That nRMSE value was comparable to or lower than other studies, despite the high fluctuations in the observed GHI.