Effect of Cloud Observations and Uncertainties on Solar Irradiance Very-Short-Term Forecast

Wednesday, 17 December 2014
Hugo T. Carreira Pedro and Carlos F Coimbra, University of California San Diego, La Jolla, CA, United States
Cloud cover is the most important factor affecting the amount of solar irradiance at any given time of the day at the ground level. Predicting accurately the extent, motion, formation, dissipation, and transmittance of ever-changing clouds is a complex and somewhat unrealistic task for solar forecast applications, even for small temporal (few minutes ahead) and spatial scales (few kilometers). Nevertheless, motivated by the the increasing interest on solar energy, much effort has been put into predicting solar irradiance based on sky-images. These models often involve many free parameters, and as a result, their accuracy is penalized when the they are not robust to the uncertainty in the many parameters involved.

In this work we address some of these issues by implementing several robust solar irradiance forecast models based on cloud cover information retrieved form sky-images. These can be either local, high-resolution images captured from sky-imagers or low resolution, regional-wide satellite images. We employ several algorithms to process the images and incorporate that information into forecast algorithms for global and direct irradiance for horizons ranging from 15 minutes out to several hours into the future. One of the algorithms is a sector method that detects the direction of motion of potentially sun-blocking clouds and propagates them into the future. Another model is a k-nearest-neighbor algorithm that uses features extracted from sky-images to identify past instances that can be used to predict the future. A second tier machine learning model is applied to incorporate the extracted information from sky-images with other meteorological and irradiance measurements to produce the final forecast output. The forecast performance is compared against other models that do not use cloud cover information. The performance analysis focuses on the periods of high cloud variability that result in large and sudden ramps in the solar irradiance.