A Multi-Faceted Approach Towards Solar Forecasting

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 7 January 2015: 4:00 PM
124B (Phoenix Convention Center - West and North Buildings)
Edwin Campos, Argonne National Laboratory, Argonne, IL; and E. Constantinescu, J. Wang, Z. Zhou, A. Botterud, D. Cook, H. F. Hamann, and S. Lu

We describe a series of modern irradiance forecasting technologies to support the solar energy sector. These combine deep machine learning with big data processing technologies to optimally blend guidance from complementary forecasting systems.

For initial training of the machine-learning algorithm and validation of the results, the study leverages a massive and rich data set. These include Geostationary satellite retrievals, numerical weather predictions, radiative transfer calculations, statistical blending outputs, measurements from the Atmospheric Radiation Measurement - Climate Research Facility in the Southern Great Plains, as well as measurements from the Argonne Weather Observatory and from the Argonne Solar Photovoltaic Plant in the Chicago area.

We generate hourly irradiance and power forecasts, from zero to 47 hours lead times, daily, for the spring, summer and fall months in 2013. Forecasting skills are analyzed as a function of forecasting lead times and predominant atmospheric conditions.

The technology promises improvements in accuracy of solar forecasting as measured by existing metrics. The final goal will be to integrate this forecasting technology into the operation of Independent System Operators and solar plants to demonstrate and prove tangible economic benefits.