UC San Diego

There has recently been a global increased interest in expanding the renew- able energy portfolio of many countries. One of the most promising technologies is that of solar photovoltaics due to the high solar resource potential. However, the solar resource at ground level is highly dependent on local meteorological conditions such as aerosol content and most notably cloud fields. This renders the solar resource inherently variable which poses problems associated with the cost of dispatchable and ancillary generators and grid reliability. As a result, high accuracy forecasts are required on multiple time horizons. The theory as well as various examples of applications in the literature are given for a number of fore- casting methods including stochastic techniques, artificial neural networks, clear sky models, persistence models, numerical weather predictions as well as satellite and ground based imaging techniques. The limited range of spatial and temporal horizon is discussed for each method and the conclusion is drawn that a high fidelity solar forecast engine would need to take advantage of a number of different forecasting methods in order to span all spatial and temporal horizons of interest