A Regime-Dependent Bayesian Approach to Short-Term Solar Irradiance Forecasts

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 7 January 2015: 5:00 PM
124B (Phoenix Convention Center - West and North Buildings)
Tyler C. McCandless, NCAR, Boulder, CO; and S. E. Haupt, G. S. Young, and A. J. Annunzio
Manuscript (2.4 MB)

We present a regime-dependent methodology for short-range predictions of Global Horizontal Irradiance (GHI) that is cast in the Bayesian framework. The first step in the process is to classify the cloud regime. The next step is to build artificial intelligence models for each cloud regime independently, such as using regression trees, neural networks, or autoregressive models. The final probabilistic forecast is an application of Bayes theorem to the regime-dependent forecasts. The forecasts are made for 15-minute intervals out to three hours. The GHI data used in this study are irradiance measurements from the Sacramento area and quality controlled weather observations from four of the closest METAR locations. The dataset consists of observations from January 25th, 2014 to May 27, 2014 and is randomly divided into 2/3 training and 1/3 testing. The results on the testing dataset show improved forecasts over a smart persistence forecast, which is the persistence of the clearness index applied to the daily solar irradiance curve.