This paper describes ongoing research in the area of solar PV production forecasting intended to address a range of effects on the utility grid associated with high penetrations of PV. The ability to anticipate near-term –minutes ahead to hours ahead to day or multiple-day ahead-- production of the variable solar resource will be key to successfully integrating ever larger PV capacities with minimal costs. A number of forecast methodologies are surveyed and a mechanism for validating their performance is described.
1. 1
Validation of Solar PV Power Forecasting
Methods for High Penetration Grid Integration
James Bing, Senior Member, IEEE, and Obadiah Bartholomy and Pramod Krishnani
forecast, by extension, becomes a forecast of photovoltaic
Abstract- This paper describes ongoing research in the area of energy production within the spatial-temporal context of the
solar PV production forecasting intended to address a range of utility service territory. To validate this model, as well as
effects on the utility grid associated with high penetrations of PV. potentially others, NEO has developed and SMUD has
The ability to anticipate near-term –minutes ahead to hours
deployed a network of 71 solar monitoring devices covering
ahead to day or multiple-day ahead-- production of the variable
solar resource will be key to successfully integrating ever larger most of SMUD’s 2330 square kilometer service territory. The
PV capacities with minimal costs. A number of forecast project was started in June, 2010 and will continue for 2 years.
methodologies are surveyed and a mechanism for validating their Solar monitoring starting in May of 2011 and will last for at
performance is described. least 14 months.
Index Terms—photovoltaic system, demand forecasting, solar
power generation, distributed power generation, power
III. INTEGRATION CONCERNS
distribution lines, power distribution, power grids.
The increase in intermittent decentralized power
I. NOMENCLATURE production, both in the form of small distributed generation
Standard Test Conditions (STC) and utility-scale PV power plants, presents a number of
concerns to utilities such as voltage control, regulator
II. INTRODUCTION equipment duty, relay desensitization increased regulation and
reserve requirements, changes in load-following resource
T HE Sacramento Municipal Utility District (SMUD),
the country’s 6th largest utility has teamed with NEO
Virtus Engineering (NEO), a solar engineering, consulting and
types, and increased O&M costs associated with cycling
existing generation [1], [2]. As PV penetrations increase,
utilities such as SMUD see an increasing need for tools to
monitoring provider, to deploy a service-territory wide solar
help plan for and schedule other resources around solar
monitoring network for validating solar forecasting models.
intermittency. Improved solar forecasting tools and validation
Thanks to a grant from the California Public Utilities
of existing tool performance are both needed to enhance
Commission, under their California Solar Initiative RD&D
utilities’ ability to manage such intermittency from an
program, SMUD has initiated a number of projects related to
operations and a planning perspective. At current solar
integrating high penetrations of solar PV.
installation rates, SMUD expects improved solar forecasting
approaches will provide value within the next 2-3 years.
NEO Virtus Engineering has developed a solar forecasting
Further, given planning horizons associated with adding
approach that makes use of the National Weather Service’s
significant new generation or storage resources, understanding
National Digital Forecast Database (NDFD) and a cloud cover
overall forecast accuracy in the next few years will be critical
radiation model (Kasten-Czeplak and Gul-Muneer). The
to planning the right resources for managing significantly
cloud cover radiation model or CRM, in conjunction with
higher penetrations in the next decade. Recognizing this,
array geometry calculations, will provide the necessary
SMUD and NEO decided to collaborate to provide a large
conversion from percentage-of-sky-covered-by-clouds to
scale validation of a specific forecasting approach and further,
incident irradiance on the module surface. This irradiance
to assess appropriate performance metrics for judging
forecasting success that were more relevant to time periods of
SMUD and NEO Virtus Engineering would like to acknowledge the grant
interest.
funding support from the California Public Utilities Commission California
Solar Initiative RD&D program and the ratepayers of California who funded
it. We would also like to acknowledge the administration of that program by Current approaches to solar forecasting have been
Itron. benchmarked and show that solar forecasting approaches
J. M. Bing is President of NEO Virtus Engineering, Inc., 410 Great Road,
B-6, Littleton, MA 01460 (e-mail: jbing@neovirtus.com). available today have a typical RMSE on the order of 100 –
O. Bartholomy, is a Senior Project Manager in SMUD’s Energy R&D 150 W/m2 for day ahead. However, this value includes many
department, Sacramento Municipal Utility District, 6201 S Street, Sacramento, clear days, and therefore likely obscures a much poorer
CA 95817 (e-mail: OBartho@smud.org).
P. Krishnani was a Former Engineering Technician with SMUD (current
performance for days of high variability. Understanding
e-mail: krishnani.pramod@gmail.com). performance metrics for specific time-periods of interest to the
2. 2
utility will assist in understanding integration costs, forecast C. Aggregated Ground Based Solar Measurements
value, and integration resource needs. By one account the US has over 3500 publically accessible
ground based measurement sites which report hourly and daily
IV. PV PRODUCTION FORECASTING METHODS observations [6]. An approach being developed aggregates
In this early stage of development of PV power forecasting these public data, in combination with national weather
technology there are a number of competing technologies and service forecasts, to produce regional one hour and three hour
methodologies . Some of the key differences amongst the forecasts. A pilot study has been done in the greater Los
competitors are time horizon, geographical area covered, Angeles region [7].
accuracy (both absolute and over time), and cost. D. Sky Imager Technology
In recent years researchers at the University of California San
TABLE I
TYPES OF IRRADIANCE FORECASTING TECHNOLOGIES
Diego and at commercial utility scale installations have
Technology Time Coverage Comments developed a method of intra-hour, sub-kilometer forecasting
Horizo using a device known as a "Sky Imager." Reflected images of
cloud motion are translated into estimates of ground level
n
Errors associated with
irradiance and, by extension, PV production [8], [9].
Satellite 12hr to global
Satellite-based weather are
7 days greatest over short time
periods[3]
Mesoscale 12hrs to global All GFS-based models,
including NDFD, have
months similar accuracies as
quantified by RSME and
MAE, but European or
Canadian global weather
simulations tend to deliver
better RSME results.[5] Fig. 1. Sanyo – UC San Diego fisheye lens technology sky imager in La Jolla,
Nationwide coverage of California.
AggriGround 1hrs to regional
higher accuracy ground
3hrs sensors make better short
term prediction based on E. Array Scale Irradiance Sensor Networks
network sensor impact on
local sensor or local
Instrumentation such as silicon pyranometers and
prediction. thermopile pyrheliometers are routinely installed in utility
Sky Imager 30min 2 to 10km This method reduces a 50- scale PV arrays providing high granularity global horizontal
60% error compared to
to 3hrs radius persistence forecasting.[8]
irradiance (GH) and plane of array irradiance (POA). The
Array Scale 1 to 30 Array size This method shows the National Renewable Energy Laboratory (NREL) is presently
minutes
impact of cloud speed on engaged in this form of high spatial/temporal monitoring in a
the complete array and also
test bed at a measurement site in Hawaii [10]. This
shows the impact of larger
and smaller clouds on the instrumentation resource will eventually be integrated into the
PV production depending very near term energy production forecasts for these plants
on the length of time
[11].
averaging. [11]
V. REGIONAL SOLAR FORECAST VALIDATION
A. Satellite Imagery SMUD was awarded a grant from the CPUC in 2010 for a 2
year project to deploy hardware and software tools to model
Satellite imagery is an established means of estimating ground and mitigate impacts of high penetrations of PV on the
level irradiance for photovoltaic system performance distribution network. SMUD’s grant partners and
assessment [3]. More recently satellite cloud motion data are subcontractors on the project include Hawaiian Electric
beginning to be used as a means of generating short-term Company (HECO), BEW Engineering, Sunpower
forecasts (hours ahead to days ahead) of ground level Corporation, and NEO Virtus Engineering. The full scope
irradiance [4]. Predictions of ground level irradiance are then includes modeling and measuring high PV penetration
extended to forecasts of PV production. circuits, developing utility interfaces to enhance the
B. Mesoscale Weather Model Forecasts understanding of the performance of intermittent resources,
Numerical weather prediction (NWP) models are being used developing methods to utilize the smart meters to
by a number of practitioners in the field of irradiance and PV communicate with PV inverters, and finally this project,
power forecasting. The models simulate cloud fraction which deploying a network of irradiance sensors to monitor and
is then extended to ground level irradiance and, further, to PV validate solar forecasting approaches. Overall, these efforts
power production forecasts. These models typically have will benefit the utilities involved as well as all California
effective forecast windows between one half a day to a week ratepayers by identifying solutions to integrating increasing
[5]. amounts of PV onto the distribution grid.
3. 3
As part of this research and to validate forecast accuracy, A primary deliverable of this research will be the database of
irradiance measurements will be made using a combination of irradiance and temperature measurements. The monitoring
six Rotating Shadowband Radiometers (primary stations) and regime calls for one-minute records of global horizontal and
sixty six global horizontal measurements systems (secondary ambient temperature measured in locations roughly evenly
stations). This combination of primary and secondary distributed across the utility's service region. As noted
monitoring stations will be deployed on the same five previously, approximately six additional sites within the same
kilometer square grid as used by the NDFD for their skycover grid will also contain diffuse horizontal and direct normal
(cloud cover) forecasts. irradiance. The monitoring network will be deployed for
approximately 14 months so that the database will cover a full
12 months with all 71 stations deployed. Once completed this
database will be made available to researchers in the field of
solar energy forecasting.
The installation of the network was completed in the March-
June 2011 timeframe, with the development of the forecasting
models following closely behind.
The validation of the PV production forecasts will be done by
Fig. 2. Rotating Shadowband Radiometer (RSR) primary station (left) and
comparing the forecasted PV output to the actual PV output
secondary station (right). for 100 MW of systems currently being installed under
SMUD’s Feed in Tariff program. Forecast validation will
The monitored area will span 1800 square kilometers within occur for 9 individual sites as well as the aggregation of sites
SMUD’s service territory. to evaluate how well the forecast does for a utility service
area.
The validation of the irradiance forecasts will compare
measured irradiance in each grid cell to the forecast irradiance
values for that grid cell, as well as aggregations of grid-cells
representing larger areas to determine the spatial effects on
accuracy and the benefits and drawbacks of different forecast
granularities.
Figure 4 demonstrates the extreme variability in a one-day
timeframe over the monitored region. As the red line
demonstrates, averaging over this spatial region can
significantly dampen minute to minute variability, but does
not completely address intra-hour and multi-hour swings in
output. Understanding how spatial averaging and overlays of
projected PV installations will smooth forecast and variability
models will be a key component of this research.
Fig. 3. Map of SMUD Service Territory, NDFD Grid cells and Centroids with
1km buffer.
The sixty six secondary stations are measuring global
horizontal irradiance and ambient temperature. The primary
stations measure global horizontal, diffuse horizontal and
direct normal irradiance, and ambient temperature (This
pairing of primary and secondary stations mimics the format
of the National Solar Resource Database [NSRDB]). All
monitoring stations are recording one minute averages. The
monitoring stations will be located in the "nominal centroid"
of each 25 square kilometer cell in the NDFD grid. The
funded research will first establish the irradiance forecasts
over the monitored region and then will quantify the error Fig. 4 Measured 1-minute data for 62 global irradiance monitors across
between measured and forecast irradiance over the term of the SMUD service territory, including average and hourly simulated dispatch
experiment. based on average
4. 4
VI. PV POWER FORECASTING WITH NDFD DATA
The National Digital Forecast Database or NDFD is a
National Weather Service product developed in the past
decade which provides digital forecasts of a range of
meteorological parameters in both numeric and graphical
form. Grids for the continental United States are currently
available from NDFD at 5 kilometer spatial resolution. The
temporal resolution for skycover (cloud cover) and
temperature is every three hours out to 72 hours and every six
hours out to 168 hours.
The forecast data may be viewed with a web browser or
retrieved via file transfer protocol in binary form. Figure 4 is
Fig. 6. Satellite map of solar monitoring network sites in Sacramento County,
an example of display of forecast skycover for an area of California
northern California which includes Sacramento County.
NEO uses the NDFD forecast elements of skycover,
temperature, and relative humidity to model direct normal and VIII. FUTURE WORK
diffuse horizontal irradiance parameters which are then used
to simulate PV power production. Given the importance to SMUD of improving the utility and
industry understanding of solar forecasting accuracies, we
anticipate future work in this area expanding the number and
type of forecasts that the monitoring network is used to
validate. We also anticipate evaluating variability of the solar
resource across our service territory, modeling variability
impacts of high penetration PV scenarios using different
spatial distributions of solar arrays, evaluating the
performance of satellite-based vs. ground-based solar
monitoring regimes, and evaluation of optimal spacing and
temporal resolution of ground-based solar monitoring regimes
for meeting utility needs. These follow-on items will assist
SMUD and the industry in better understanding how solar
forecasts perform over a utility service territory, and how
important geographic diversity is to mitigating variability.
Ultimately this work will lead to a much clearer picture of the
costs and types of grid assets and services that will be needed
to accommodate high penetrations of solar PV.
IX. APPENDIX
A secondary benefit of the validation monitoring network
which spans SMUD’s service territory is that it provides an
unprecedented high spatial-temporal resolution measurement
of the solar resource, and by extension view of the potential
Fig. 5. Skycover variable depiction from National Weather Service NDFD PV power production contribution to an entire utility service
territory.
VII. PRELIMINARY PV POWER FORECAST RESULTS The sequence of three dimensional graphs shown in figures 7
As a result of both the timing of SMUD’s Feed in Tariff PV through 12 below is taken from the 71 station monitoring
system installations and the development timeline for the solar network. The devices measure global horizontal data for May
forecasting model, it is unlikely that sufficient generation data 14, 2011. The day is one of intermittent clouds with a peak
will be available, nor a forecast model ready, for PV power irradiance occurring both before and after the typical high
forecasting validation until February, 2012. Depending on the irradiance time of the day, solar noon, 12:06pm. The ability
availability of both of these elements, it is unclear to what to forecast and thus anticipate the equivalent PV generation
extent the PV power validation piece will be available for the carrying capacity in the geographic context of SMUD’s
final submittal of this paper. service territory could greatly facilitate the management of
both load following and regulation requirements.
5. 5
Fig. 7. Measured global horizontal irradiance (GH) levels viewed across Fig. 10. Measured global horizontal irradiance (GH) 11:29am.
SMUD service territory on May 14, 2011.
Fig. 11. Measured global horizontal irradiance (GH) 12:58pm.
Fig. 8. Measured global horizontal irradiance (GH) 8:26am.
Fig. 9. Measured global horizontal irradiance (GH) 9:28am. Fig. 12. Measured global horizontal irradiance (GH) 2:28pm.
X. ACKNOWLEDGMENT
SMUD and NEO Virtus Engineering would like to
acknowledge the grant funding support from the California
Public Utilities Commission California Solar Initiative RD&D
program and the ratepayers of California who funded it. We
would also like to acknowledge the administration of that
program by Itron.
Any opinions, findings, and conclusions or
6. 6
recommendations expressed in this material are those of the evaluation for meeting SMUD’s Renewable Portfolio Standards. He also leads
SMUD’s Climate Change and Energy Efficiency R&D programs. He earned a
author(s) and do not necessarily reflect the views of the BS in mechanical engineering from Cal Poly, San Luis Obispo and an MS in
CPUC, Itron, Inc. or the CSI RD&D Program. Transportation Technology & Policy from UC Davis, and is a registered
Professional Engineer in the state of California.
XI. REFERENCES
Pramod N. Krishnani is the Performance Monitoring
Engineer of Belectric, Inc. He is a Mechanical Engineer
[1] J. Smith "Feeder Characterization for PV Integration Assessment," working in data acquisition, monitoring, prediction and
presented at DOE Hi-Pen PV Workshop, SMUD, Sacramento, power systems design in the photovoltaic industry. Prior
California, June 13, 2011. to the work at Belectric, Mr. Krishnani has worked for
[2] M. Thomson and D. G. Infield, "Network Power-Flow Analysis for a Sacramento Municipal Utility District (SMUD) in the
High Penetration of Distributed Generation," in Proc. 2009 IEEE Power Research and Development Department from 2008 to
Engineering Society General Meeting. 2011; where he has evaluated PV performance, meteorological solar data and
[3] J. Stein, R. Perez and A. Parkins, "Validation of PV Performance developed custom PV modeling and cloud transient modeling tools. At
Models Using Satellite-Based Irradiance Measurements: A Case Study " SMUD, Mr. Krishnani has completed work related to other renewable energy
presented at American Solar Energy Society annual SOLAR 2010 like Biogas, wind energy, storage technology and Geothermal. All the work
conference proceedings. presented by Pramod Krishnani in this paper was completed during the period
[4] R. Perez, S. Kivalov, J. Schlemmer, K. Hemker Jr., D. Renné, T. Hoff, of employment at SMUD and has no affiliation with Belectric Inc. He has
"Validation Of Short and Medium Term Operational Solar Radiation simulated several study projects related to utility load regulation, impact of
Forecasts In The Us," presented at American Solar Energy Society solar and wind on utility load and simulated Utility scale PV from 500kW to
annual SOLAR 2009 conference proceedings. 300 MW. In addition to his Mechanical Engineering Credentials, Mr.
[5] R. Perez, S. Kivalov, S. Pelland, M. Beauhamois, E. Lorenz, J. Krishnani has three years of experience in project management, research and
Schlemmer, kHemker, Jr., G. Van Knowe,"Evaluation Of Numerical estimating.
Weather Prediction Solar Irradiance Forecasts In The Us," presented at Mr. Krishnani holds a MS in Mechanical Engineering from California State
American Solar Energy Society annual SOLAR 2011 conference University – Sacramento and BS and Diploma in Mechanical Engineering
proceedings. from University of Mumbai.
[6] James Hall and Jeffery Hall,"Quality Analysis Of Global Horizontal
Irradiance Data From 3500 U.S. Ground-Based Weather Stations,"
presented at American Solar Energy Society annual SOLAR 2011
conference proceedings.
[7] James Hall and Jeffery Hall,"Forecasting Solar Radiation For The Los
Angeles Basin – Phase II Report," presented at American Solar Energy
Society annual SOLAR 2011 conference proceedings.
[8] B. Urquhart, C. W. Chow, M. Lave, J. Kleissl, "Intra-Hour Forecasting
With a Total Sky Imager at The UC San Diego Solar Energy Testbed,"
presented at American Solar Energy Society annual SOLAR 2011
conference proceedings.
[9] M. Ahlstrom and A. Kankiewicz, "Solar Power Forecasting," presented
at Solar Power International conference Los Angeles, California, 2010
[10] M. Sengupta,"Measurement and Modeling of Solar And PV Output
Variability," presented at American Solar Energy Society annual
SOLAR 2011 conference proceedings.
[11] A. Kankiewicz, M. Sengupta, D. Moon,"Observed Impacts of Transient
Clouds on Utility-Scale PV Fields" presented at American Solar Energy
Society annual SOLAR 2010 conference proceedings.
XII. BIOGRAPHIES
James M. Bing (M’ 1987, SM' 2006) James Bing is
the founder and president of NEO Virtus
Engineering, Inc. (formerly New Energy Options,
Inc.) He is a professional electrical engineer
working principally in data acquisition and power
system design in the photovoltaic industry. Prior to
working in the photovoltaic industry Mr. Bing
worked a for the R&D firm of Bolt Beranek and
Newman in the area of sensor system and
technology. In addition to his engineering credentials Mr. Bing possesses
over twenty years of experience in the areas of electrical system design,
project management, estimating, and personnel supervision in the electrical
construction industry. Mr. Bing holds a BS in Electrical Engineering from the
University of Massachusetts at Lowell, as well as an AB in Anthropology
from the University of California at Berkeley.
Obadiah Bartholomy is a Project Manager in the
Energy R&D group at SMUD. He currently works on
solar R&D, including intermittency impacts and
forecasting, evaluation of high penetrations of PV on
the distribution system, utility scale solar siting,
commercial and residential solar mapping tools
including the SMUD Solarmap, and utility scale solar