|
|||
|
|||
 |
|
 |
|
 |
|||
 |
|||
 |
|||
|
|||
 |
|||
|
|||
|
|||
|
|||
 |
 |
||
|
|||
|
|||
 |
|||
|
|||
|
|||
|
|||
|
|||
 |
|||
|
|||
|
|||
|
|||
|
|||
|
|||
 |
|
|
|
|
|
|
(c) 1-Axis tracking PV arrays
(d) Thin-film PV roof shingles
(e) Concentrating PV on 2-axis tracker (f) Building integrated PV
FIGURE 1.2 Examples of commercially available PV systems for producing electricity in a variety of applications: (a) fixed-tilt PV arrays; (b) polycrystalline PV modules; (c) fixed-tilt PV arrays; (d) thin-film PV roof shingles; (e) concentrating PV on 2-axis tracker; (f) building – integrated PV. (Courtesy of NREL Image Gallery, http://images. nrel. gov.)
Wavelength (nm)
 |
FIGURE 1.3 Spectral response functions of selected PV materials illustrating their selective abilities to convert solar irradiance to electricity. (Courtesy of Chris Gueymard.)
Voltage
FIGURE 1.4 PV system performance characteristics determined by short-circuit current (T^) and open-circuit voltage (VOc), and maximum power point (Pmax).
 |
 |
0 2 4 6 8 10 12 14 16 18 20 22
Voltage
 |
FIGURE 1.5 PV-array short-circuit current (Isc) is proportional to solar irradiance incident to the module. Open-circuit voltage is much less dependent on irradiance level.
Solar Irradiance (Watts / Square meter)
Parabolic trough collector
(b) Power tower and heliostats
MSJfc
аь.
(c) Dish Stirling engine (d) Linear Fresnel collector
FIGURE 1.7 Examples of CSP systems for converting high levels of DNI to heat and electricity (a) parabolic trough collector; (b) power tower and heliostats; (c) dish sterling engine; (d) linear Fresnel collector. (Courtesy of NREL Image Gallery, http://images. nrel. gov.)
Reflected Atmospheric
scattering
|
|
FIGURE1.8 Solar-radiation components resulting from interactions with the Earth’s atmosphere and surface provide POA irradiance to a flat-plate collector (POA — Direct + Diffuse + Ground – reflected). (Courtesy of Al Hicks, NREL.)
|
FIGURE 1.9 Time – series plot of solar-irradiance components for clear and cloudy periods as measured by pyrheliometers (A = DNI) and pyranometers (B = GHI; C =DHI), and corresponding sky images during the day, Golden, Colorado, July 19, 2012. |
|
|
 |
|
|
|
|
 |
The Eppley Laboratory, Inc. Model NIP (Normal Incidence Pyrheliometer) 1957 – 2012
|
|||||
|
|||||
|
|||||
|
|||||
|
|||||
|
|
|
|||
|
|
|
|||
Angle from Sun Center (Degrees)
|
|
||
|
|
|
|
 |
|
|
|
FIGURE 1.15 Elements of the solar-forecasting process for electric utility operational needs over a range of timescales. |
 |
 |
£} …………………………
Geqstationaryprbit
FIGURE 2.1 Geostationary and polar-orbiting satellite orbits and operational field of views.
|
FIGURE 2.3 Global map showing annual AOD 670 averaged over the year 2009, calculated from the Monitoring Atmospheric Composition and Climate (MACC) database developed by a consortium coordinated by the European Centre for Medium-Range Weather Forecasts (ECMWF). The color scale is 0.02-0.60. |
|
FIGURE 2.4 Global map showing the annual average of precipitable water for the year 2009, calculated from the NOAA/NCEP Climate Forecast System Reanalysis (CFSR) database (kg/m2). |
|
FIGURE 2.8 Sample of dynamic range for a site over the Atlantic Ocean from the visible channel of the GOES-East satellite. GOES-13 replaced GOES-12 in May 2010, resulting in a change in the dynamic range. |
|
2012 |
|
І 100 |
|
FIGURE 2.11 Dynamic range for the year 2010 in a location with frequent occurrences of snow cover (Fort Peck, Montana). |
|
FIGURE 2.12 Example of the classification output for Tartu-Toravere, Estonia, for Meteosat: (a) reflectance for the visible channel at 0.6 um, (b) classification for cloud-free land, (c) cloud-free snow, and (d) clouds. The x-axis represents day of year; the y-axis, time slot of the satellite image (bottom, morning; top, evening). |
|
FIGURE 2.13 Snapshot of the SolarGIS database: annual average DNI (kWh/m2) representing years 1994 (1999 in Asia and Australia) through 2011. |
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
FIGURE 2.17 Scatter-plot and cumulative frequency distribution of DNI data before (blue) and after (red) site adaptation for Tamanrasset, Algeria (grey): cumulative distribution of ground measurements.
(а)
Minutes Hours Days+
————————————– >
Forecast Lead Time
FIGURE 3.1 (a) Conceptual diagram of forecast skill hand-off as a function of forecast lead time
for different methods ranging from persistence to climatology. The curve with the greatest potential for advance in skill is numerical weather prediction; satellite data play a vital role here in terms of both analysis and improved parameterization. (b) Example solar-forecast methods from Fig. 3.1a, from left: persistence, surface-based trajectory, satellite-based trajectory, weather – forecast models, and climatological cloud statistics constrained by meteorological regime. Satellite information is applicable to all of these timescales.
|
FIGURE3.5 Cloud advection in a short-range solar forecast. Top: surface observation time series of solar irradiance as measured at a surface station near Fort Collins, Colorado, on June 26, 2010. Middle: clouds (blue — cold tops, yellow — warmer tops) moving across the station location (shown as a white cross). Bottom: cloud field over the solar array as viewed from the south. Over the 2100-2130 UTC time period, a break between clouds results in a rapid ramp-up of solar irradiance. |
|
Time (UTC) |
|
FIGURE 3.7 CloudSat cross-section through the eye of Hurricane Ileana in the Eastern Pacific on August 23, 2008, showing the detailed inner-core structure of the storm. |
|
N0GAPS* Pressure (mb) |
|
FIGURE 3.8 Importance of accounting for cloud height and solar geometry when forecasting solar irradiance at surface stations. Shadows may extend tens of kilometers away from the subcloud location. |
|
FIGURE 3.9 Speed and directional sheer of the atmospheric wind field—an important consideration for cloud advection that requires detailed knowledge of the vertical distribution of clouds. |
|
FIGURE 3.10 Observed and simulated cloud field (Weather Research and Forecasting (WRF) model data passed through an observational operator). |
January 1998-2002 Valid Time: 1900 UTC
500
FIGURE 6.2 Dispersion-smoothing effect occurring at 25 locations dispersed over a 4 x 4 km area (Data from the Cordelia Junction network, San Francisco Bay area, California.)
|
FIGURE 6.3 Site-pair correlation as a function of distance (D) and time interval (At) for stations in the ARM network. (From Mills and Wiser 2009.) |
|
FIGURE 6.4 Site-pair variability correlation as a function of distance derived from hourly 10 km-resolution satellite data for California (top) and the Great Plains (bottom). The top row in each case represents p as a function of distance. The bottom row expresses this relationship as a function of the ratio between D and At x implied CS, showing that the distance relationship is predictably dependent on At and CS. |
|
Б50 |
10 20 ЗО 40
At (minutes)
FIGURE 6.7 Applying equation 6.4 to estimate the effective site-pair decorrelation distance as a function of Dt and CS. The short line labeled “Virtual network” represents the preliminary estimate of this relationship based on limited evidence.
|
FIGURE 6.10 Smoothing effect at the scale of a metropolitan area comparing single-site and modeled 40 km x 40 km extended fluctuations for different timescales. |
Time
CRITICAL TIME SCALE
RR0 [(MW*arbitrary scaling factor)/1 s]
RR0 [(MW*arbitrary scaling factor)/10 s]
RR0 [(MW*arbitrary scaling factor)/30 s]
RR0 [(MW*arbitrary scaling factor)/60 s]
FIGURE 7.7 Cumulative distribution of ramp rates in power output for the 1 y period from August 1, 2011, through July 31, 2012. Ramp rates are shown at various timescales: 1 s (top left), 10 s (top right), 30 s (bottom left), and 60 s (bottom right). At each timescale, shown are the ramp rates of measured power output (thick blue line), WVM run with ground CS values (dashed green line), WVM run with NAM-cdf CS values (dashed red line), and a point sensor with no smoothing (dashed magenta line). The x-axis is the RR in MW/timescale multiplied by an arbitrary scaling factor to protect the confidentiality of the power data.
timescale
September, 2012
Sun Mon Tues Wed Thurs Fri Sat
FIGURE 7.10 RRs for the 60 MW plant: violations (red dots); total number of violations per day (bottom, bold red).
days per month
10% 15% 20% 25% 30% 35% 40% 45% 50%
RR [% of capacity / minute]
Longitude (Degrees)
FIGURE 8.1 Forecast GHI (W m-2) on April 10, 2010, at midday from the North American Mesoscale model (NAM).
|
FIGURE 9.1 TSI mounted on an inverter enclosure at a solar plant in the United States. |
|
FIGURE 9.2 (a) Canopy camera and (b) the SIO-MPL’s WSI deployed at the Department of Energy’s Atmospheric Radiation Measurement Program field site in Lamont, Oklahoma. |
|
FIGURE 9.6 HDR process on the USI showing three exposure times: (a) 5 ms, (b) 20 ms, and (c) 80 ms; (d) final composite. |
Subset of pixels
to correlate
Search Window
Position of
original subset
region with
highest correlation!
FIGURE 9.13 Normalized cross-correlation method used to compute inter-image cloud motions. The image at to-30 s (a) is broken into small tiles, each of which is cross-correlated with the corresponding search window in (b), the image at fo.
FIGURE 9.17 Sequential cloud advections for a single forecast issue with the direction of motion indicated. The cloud positions are shown for the nowcast (a), along with the 5 min (b), 10 min (c), and 15 min (d), cloud-position forecasts.
|
cloudmap |
|
shadowmap |
|
FIGURE 9.18 Ray tracing to construct a georeferenced mapping of shadows. The shadow value for a given point in the forecast domain grid is determined by tracing a ray along the solar vector and determining the cloud value at the intersection with the cloudmap. |
solar anywhere
/iNHANCCO RISOLUTION
SOLAR anywhere
/STANDARD RISOLUTION
Zoom in lo select location(s)
Standard Resolution
Hourly – 10 km
Enhanced Resolution
/2 hourly – 1 km
|
FIGURE 10.3 KSI and OVER metrics. Top: modeled and measured cumulative probability distributions and the critical value envelope around the measured distribution. Bottom: absolute difference between the two distributions. The metrics are obtained by integrating the area under the curves: KSI (lightly shaded); OVER (striped). |
|
RMSE (Wm |
|
FIGURE 10.4 Annual RMSE and forecast skill as a function of forecast time horizon. |
|
Forecast Skill |
FIGURE 10.5 Comparison of hourly forecasts and persistence versus measured GHI scatter plots for 1, 3 h ahead and 1,3 d ahead. Scatter plots provide a qualitative, visual appreciation of model performance showing that the core of forecast points are closer to the 1:1 line and exhibit fewer outlying points.
|
FIGURE 13.5 Transform between (a) lognormal and (b) Gaussian spaces and its implications. The horizontal blue, red, green, and magenta lines indicate the inverse transform from the transformed normal distribution back to the lognormal distribution for lognormal distributions of s = 0.25, 0.5, 1.0, and 1.5, respectively. When inverted from the Gaussian-transform analysis space, the transform approach finds the median in the lognormal space and thus loses all skewness information contained in the original lognormal distribution, where the vertical blue, red, green, and magenta lines indicate the respective original lognormal modes. |
|
FIGURE 13.6 Cloudy-radiance assimilation using the RAMDAS 4DVAR system for a region in central Oklahoma with a domain of 300 x 300 km (using 6 km horizontal grid spacing). The results demonstrate use of the GOES Sounder channel-1 (12 pm) on March 21, 2000, at 11:45 UTC. Blue denotes cold cloudy brightness temperatures (K) (i. e., high to middle clouds); red denotes warm brightness temperatures (K) (i. e., low clouds). The DA processing moves from left to right: (a) first guess (current model state), (b) final assimilation analysis state, and (c) GOES Sounder-channel 1 satellite observations. The original mean RMS error was 39 K; in the converged final analysis, the RMS error is 3.9 K. (Images courtesy of Manajit Sengupta.) |
|
FIGURE 14.3 Percentage maximum total revenue (R) as a function of forecast error and the ratio of RTM to DAM price for a market system with a forecast-deviation penalty of twice the maximum of the RTM or DAM (equation 14.3). The white Hue represents 0 total revenue, not including cost of operation. |
|
FIGURE 14.6 Direct cloud assimilation using a GOES cloud mask. (a) Clouds are to populate qvapor in WRF initial conditions (green); (b) May 17, 2011. |
|
FIGURE 14.9 GOES satellite imagery: (a) June 11, 2011, compared to the intraday (0-24 h, initialized on June 11, 2011, 12 UTC) (b) and day-ahead (24-48 h, initialized on June 10, 2011, 12 UTC); (c) WRF-CLDDA irradiance forecasts in San Diego, California. |
[1] Of course, this assumption, like most other model assumptions, becomes less robust for very large solar-zenith angles
[2] Current U. S. GOES satellites provide images on a half-hour basis, whereas current Meteosat satellites provide data every 15 min. The next round of GOES satellites is expected deliver data every 5 min.
[3] Concerns with calibrations at three of the Texas sites and the three component comparison inconsistencies (GHI, DNI, and DHI) skewed the results, so all the Texas sites were removed and the MBE errors were recalculated with a more consistent dataset. These errors are discussed more in the measured data section of this chapter.
[4] Direct normal rradiance (DNI) would be the relevant quantity if concentrating technologies were considered.
[5] The range of the global index is reduced as the Sun’s elevation decreases, because the relative weight of diffuse irradiance increases during clear-sky conditions.
[6] Currently available in North America.
Solar Energy Forecasting and Resource Assessment. ISBN: 9780123971777
Copyright © 2013 Elsevier Inc. All rights reserved.
[7] Kt* equals the ratio between (satellite-derived) GHI and local clear-sky global irradiance
GHIclear.
[8] The current version of SolarAnywhere uses only the satellite’s visible channel to determine cloud indices; hence, cloud motion can be determined only after sunup. As a result, N-h forecasts are only available N + 1 h after sunrise. The new version of SolarAnywhere uses the infrared satellite channels in addition to the visible channel, thus making it possible to infer nighttime cloud motion and so overcome this limitation.
[9] The achievable time resolution is defined by the ratio of cloud speed to the image’s spatial resolution (1 km), which defines the size of the cloud structures that can be captured to determine variability at a given time scale.
[10] Therefore, the results here represent a worse-case evaluation of the SolarAnywhere forecasts because, operationally, forecasts are refreshed every hour.
[11] Note that another measure of dispersion, mean absolute error, was recently recommended as the preferred methodology to report relative (percentage) errors (Hoff et al. 2012).
[12] This is because the satellite model used in this evaluation uses only the satellite’s visible channel. The next version of SolarAnywhere, which uses the satellite’s IR channels, will not have this limitation.
[13] The loss of dynamic range reflects the tendency of the NDFD forecasters to increasingly “hedge” their forecasts as the time horizon increases.
[14] Note that the IEA team did not compare the GFS global model directly to the other global models, but only its application via mesoscale models or the NDFD. Mathiesen and Kleissl (2011) have shown that the standalone GFS model should indeed perform better than through-the-filter mesoscale models, which tend to introduce unwarranted dispersion error at this stage of their irradiance-modeling development.
