IEEE Power & Energy Magazine - November/December 2015 - 51

©ISTOCKPHOTO.COM/LUKAS_ZB, PHOTO COURTESY OF NREL

By Aidan Tuohy, John Zack, Sue Ellen Haupt,
Justin Sharp, Mark Ahlstrom, Skip Dise,
Eric Grimit, Corinna Möhrlen, Matthias Lange,
Mayte Garcia Casado, Jon Black,
Melinda Marquis, and Craig Collier
solar power forecasting can be differentiated from wind forecasting in that much of the variability and
uncertainty is related to visible cloud movement. This can be forecast in the short term by technologies
such as ground-based sky imaging and satellite imaging systems; these are described here, along with
potential drawbacks related to cloud formation and dissipation.
The underlying "clear-sky" solar PV output is easily calculated based on the position of the solar
panels relative to the sun, so the main challenges lie in predicting the actual irradiance at the solar panels
given the influence of clouds, aerosols, and other atmospheric constituents, and also PV panel efficiency,
which is temperature dependent. Adding to these challenges, large amounts of solar are being installed
on the distribution system, often behind the meter (BTM), which means that transmission system operators see only load netted with the solar generation rather than the output of the solar system.
In this article we outline current methods used to produce solar PV power forecasts, focusing on
aspects unique to solar forecasting. We also explore the nuances related to BTM solar forecasting and
how these interact with load forecasting. In addition, we summarize the current performance of solar
forecasting and then end by discussing improvements on the horizon that will increase operators' ability to accurately forecast solar PV generation.

Current Solar Forecasting Techniques
Forecasting solar output involves a variety of methods based on the time frame being forecast, the data
available to the forecaster, and how the forecast is to be used. Past issues of the magazine have included
articles on wind forecasting covering many basic forecasting techniques also important for solar (see
the "For Further Reading"section). These methods, which are broadly categorized in Figure 1 according
to the time horizon in which they generally show value, include numerical weather prediction (NWP)
and the use of model output statistics, as well as statistical learning methods, climatology, and ensemble
techniques that blend different kinds of forecasts.
To achieve the greatest accuracy, forecasters supply environmental inputs, such as irradiance and temperature, to models that transpose the irradiance into the incident plane of array (i.e., by taking the beam
direction and solar panel orientation into account) and then convert the irradiance to power. Several methods can be used for this purpose; these are described below beginning with those that perform best on the
shortest look-ahead time frames.

Methods, Challenges,
and Performance

Time Series Prediction with
Statistical Learning Methods

Direct observation of irradiance (such as from pyranometers
or other measurement devices) can be used with time series
statistical learning methods to project subsequent conditions.
Historical records of site irradiance are used to train these
prediction methods, while real-time measurements identify
current condition on which to base the forecasts. Methods used include artificial neural networks,
regression models, autoregressive models, support vector machines, and Markov chains, as well as
composite methods, such as using genetic algorithms to optimize a neural network. These methods
work best for the intrahour time horizon, but they may have some value out to two to three hours or
more, especially when used in combination with other methods.

Sky Imagers
Sky imagers are digital cameras that produce high-quality images of the sky from horizon to horizon, which are used for detecting clouds, estimating cloud height above ground, and calculating cloud
november/december 2015

ieee power & energy magazine

51
http://www.ISTOCKPHOTO.COM/LUKAS_ZB
Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2015
