Deterministic Solar Power Forecasting Using Sky Imagery

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 6 February 2014: 3:30 PM
Room C114 (The Georgia World Congress Center )
Chi Wai Chow, Univ. of California, La Jolla, CA; and H. Yang, B. Kurtz, A. Nguyen, B. Urquhart, M. Ghonima, and J. Kleissl

Sky imaging systems have a history of providing atmospheric observation and monitoring. Because of their ability to continuously monitor cloud conditions, they have been used for solar power output forecasting (Chow et al. 2011, Urquhart et al. 2013). These imagers are effective monitoring systems for solar plant operations, providing both situational awareness and estimates of power output over a large spatial extent (Urquhart et al. 2012). The University of California San Diego (UCSD), leveraging expertise in sky imaging technology (Shields et al. 2013), has designed a sky observing system (UCSD Sky Imager or USI) (Fig. 1a) specifically for solar resource assessment work (Urquhart et al. 2013). The increased resolution and dynamic range has allowed the USI to overcome the primary shortcomings of the previous imaging system used at UCSD, which was a commercially-available Total Sky Imager.



Description: IMG_6252.JPG


Description: R:\database\USI\images\usi1-7\20130325\20130325173100_prev.jpg

Figure 1 A whole sky imager developed at University of California, San Diego (a) and its raw sky image (b)

Using the USI camera system, the short-term irradiance forecasting methodology developed in Chow et al. (2011) has been improved significantly. Sky images taken every 30 seconds are processed to determine sky cover and cloud motion vectors. Future cloud locations are computed and used to predict the global horizontal irradiance. The performance results for both cloud location forecasting and surface irradiance forecasting has been evaluated over a 31 day period. Forecasts were issued every 30 seconds for a forecast horizon of 10 minutes at 30 second intervals. Distributed pyranometer measurements over several square kilometers were available to provide spatial averages of irradiance. For the duration of this period, cloud location matching errors, which monotonically increase as a function of forecast horizon, did not exceed 30% over the sky domain considered. Irradiance forecast error ranged from 9-14% mean absolute error, and 17-23% root mean square error, where the error in W/m2 was normalized by daily average irradiance. Visual evaluation of ramp forecasting skill also indicates the forecast methodology accurately captures the cloud-induced irradiance fluctuations over the UCSD campus.

Specific improvements to the forecast procedure initially developed by Chow et al. (2011) will be highlighted. New developments include the ability to classify clouds into thin and thick (Fig 2a, Ghonima et al. 2012), the ability to determine the cloud height accurately using stereoscopic techniques (Fig 2b), and a method to estimate average cloud transimissivity over the forecast domain. Updates on the contemporary research of novel optical flow and cloud segmentation methods used to separate cloud layers and determine their individual velocities will also be described. These solar forecasting methods are currently being applied beyond the UCSD campus to locations such as a 48MW solar power facility in Nevada, and a large 10.5 MW multi-rooftop warehouse array in Redlands, CA and select results from these studies will be presented as well as results become available.



Description: C:\Users\Chi Wai Chow\Google Drive\0458.png


Description: heightmap_overlay.epsDescription: heightmap_overlay.eps

Figure 2 (a) Mapping of cloud shadows on the ground, classified as being cast by thin cloud (light gray), thick cloud (dark gray), or no shadow (blue). Ground station locations are marked by black boxes. (b) Cloud heights can be determined with two sky imagers using stereoscopic techniques.


Chow, C. W., Urquhart, B., Lave, M., Dominguez, A., Kleissl, J., Shields, J., Washom, B., 2011. Intra-hour forecasting with a total sky imager at the uc san diego solar energy testbed. Solar Energy 85, 2881-2893.

Ghonima, M. , Urquhart, B, Chow, C. W., Shields JE, Cazorla A., Kleissl J., 2012. A method for cloud detection and opacity classification based on ground based sky imagery. Atmospheric Measurement Techniques Discussion, Volume 5, pp. 4535-4569, doi: 10.5194/amtd-5-4535-2012.

Gohari, S. M. I., Urquhart, B, Kleissl J., “Comparison of solar power output forecasting performance of the Total Sky Imager and the University of California, San Diego Sky Imager”, Energy Procedia, Elsevier, 2013.

Shields JE, Karr ME, Johnson RW, Burden AR (2013). Day/night whole sky imagers for 24-h cloud and sky assessment: history and overview. Applied Optics Volume 52, Issue 8, pp. 1605–1616, doi: 10.1364/AO.52.001605.

Urquhart, B., Ghonima, M., Nguyen, D., Kurtz, B., Chow, C. W., Kleissl, J., 2013. Sky imaging systems for short-term solar forecasting, chapter appearing in Solar Energy Forecasting and Resource Assessment.

Urquhart, B., Chow, C. W., Nguyen, D., Kleissl, J., Sengupta, M., Blatchford, J., Jeon, D., 2012. Towards intra-hour solar forecasting using two sky imagers at a large solar power plant. In: Proceedings of the American Solar Energy Society. Denver, 568 CO, USA.