The Use of Analog Ensembles to Improve Short-Term Solar Irradiance Forecasting

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 6 February 2014: 4:00 PM
Room C205 (The Georgia World Congress Center )
Steven H. Young, MESO, Inc., Troy, NY; and J. W. Zack

Handout (2.0 MB)

Accurate short-term wind and solar irradiance forecasts become increasingly important as the penetration of wind and solar power production on electric grids increases. Wind and PV integration presents a particular challenge for the state of Hawaii due to the isolated nature of the individual electric grids on each island and the large wind and solar penetration on several of the islands. The lack of upstream observations and the complex flow patterns resulting from the interaction of the stable northeasterly trade winds with the islands also contribute to the challenge of producing an accurate forecast. To meet the forecast need, AWS Truepower has developed the Solar and Wind Integrated Forecast Tool (SWIFT) with funding from the Hawaiian Electric Company (HECO) and the Electric Power Research Institute (EPRI).

SWIFT employs a suite of forecast methods to generate wind and solar power forecasts for a broad range of look-ahead time scales extending from intra-hour to days ahead. One of the primary approaches for solar power forecasts for hours ahead time scales is the use of a cloud advection algorithm known as the Pyramidal Image Matcher (PIM) (Zinner et al., 2008, Young and Zack, 2013) along with a scheme to statistically adjust the PIM forecast to reduce systematic errors (biases). The focus of this presentation is on the formulation and performance of the statistical error reduction scheme. This scheme is based on the analog ensemble (AE) concept (Delle Monache, et al. 2013). The AE generates probabilistic and bias-corrected deterministic forecasts clear sky factor (CSF) from a deterministic PIM forecast of CSF. CSF is defined as the ratio of observed global horizontal irradiance (GHI) to clear sky GHI.

The AE method selects a historical sample of similar cases by picking PIM forecasts from an historical sample (a form of training sample) that most closely resemble the current PIM forecast. There are several steps to this process. First, a set of observed or simulated “case-matching” variables is chosen. Second, the case-matching score components are computed. A case-matching score component is the normalized difference between a case-matching variable for the forecast case and that same variable from an historical case. Finally, the case-matching score components are combined into a case-matching score that measures the composite difference between the case-matching variables from the current case and the historical case. The observed CSFs from historical cases with the smallest case-matching scores are selected as ensemble members.

The AE method was initially applied to forecasting CSF at 7 Hawaiian surface stations that measure global horizontal irradiance (GHI) as well as 144 electrical substations on Oahu and the Big Island of Hawaii. Solar irradiance measurements are not available at the substations, so substation forecasts are verified against the CSFs derived from visible satellite image data. Conversions between CSF and GHI were accomplished using the method of Perez et al., (2002). Initial case-matching variables included average PIM-predicted CSF over a 10-km box centered at the forecast site at each forecast interval, and the average direction and amplitude of the PIM cloud displacement vectors computed from the two most recent visible satellite images over the same box.

Deterministic forecasts consisting of the AE mean and 50% probability of exceedance were compared to the raw PIM forecast, a persistence-forecast and a the mean diurnal cloudiness trend superimposed on a persistence forecast.

The conference presentation will provide an overview of the AE method and show the impact of the selection of case matching variables, size of the ensemble and other factors on AE forecast performance.

Delle Monache, L., F.A. Eckel, D.L. Rife, B. Nagarajan and K. Searight, 2013: Probabilistic Weather Prediction with an Analog Ensemble, Mon. Wea. Rev. In press. Available from: http://journals.ametsoc.org/doi/abs/10.1175/MWR-D-12-00281.1

Perez, R., P. Ineichen, K. Moore, M. Kmiecik, C. Chain, R. George and F. Vignola, 2002: A new operational model for satellite-derived irradiances: Description and validation. Solar Energy, 73, 307-317.

Young, S;, and J. Zack, 2013: Application Of A Pyramidal Image Matching Scheme For Short-Term Cloud And Irradiance Prediction in The Hawaiian Islands. SOLAR 2013, Annual conference of the American Solar Energy Society. In press. Available from http://www.proceedings.com/1773.html

Zinner, T., H. Mannstein, A. Tafferner, 2008: Cb- TRAM: Tracking and monitoring severe convection from onset over rapid development to mature phase using multi- channel Meteosat-8 SEVIRI data. Meteor. Atmos. Phys. 101, 191–210, DOI 10.1007/s00703-008-0290-y.