The MDA Solar Power Forecast System: Sub-hourly variability and behind-the-meter generation

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 5 January 2015
Stephen D. Jascourt, MDA Information Systems LLC, Gaithersburg, MD; and D. Kirk-Davidoff, C. C. Cassidy, and T. Hartman

Handout (840.1 kB)

This is the third in our annual series on the MDA Solar Power Forecasting System, presenting topics of interest as we have climbed the learning curve in making solar power forecasts. Two years ago, we reported on our state-of-the-science irradiance forecast system utilizing the REST2 clear sky model, AERONET aerosol observations, and other public sources as well as private site data. Last year we reported on lessons learned from forecasting electricity generation utilizing generation data from a single-axis PV farm in a challenging location beset by synoptic and local storms as well as sunny-day cumulus. This year our focus will be on sub-hourly variability and on behind-the-meter generation.

Sub-hourly irradiance and electric generation variability not only is interesting itself but impacts the total generation for the hour. This arises because of the truncation nonlinearity in the inverter system outputting AC current. At high values of irradiance, exceeding around 1000 W/m2, the excess current from the PV panels beyond the rated amount is converted into higher voltage with lower current by the inverter, capping the output AC current. Thus, the power output of the PV array is limited. So a mostly sunny day with a cumulus field could average 1000 W/m2 plane-of-array irradiance over an hour as periods of 1100 W/m2 are interspersed with cloud shadows of far less, while another day could have cirrus and a rather steady 975 W/m2. The cirrus day would produce more power because of the truncation during the sunny periods on the cumulus day. We found a way to predict the character of variability down to time scales of 1-minute averages based on conditional sampling of SURFRAD data, and we found that the sampling statistics we pull from to make predictions are rather similar (albeit not identical) at all of the SURFRAD sites. This approach results in improvement in our forecast error scores for 1-hour generation.

We are predicting the aggregate generation from the million installations in Germany and have developed a methodology for the behind-the-meter sites in California. The methodology of reducing the number of sites to a tractable number and ideal simulation of how that compares under a perfect forecast to calculating all sites individually will be shown.

Supplementary URL: http://www.mdaus.com/Products-0024-Services/Weather-Services.aspx