GREENING THE GRID

An Introduction to Wind and Solar

Power Forecasting

Bri-Mathias Hodge, Ph.D.

National Renewable Energy Laboratory
April 21, 2016

ENHANCING CAPACITY FOR LOW EMISSION

DEVELOPMENT STRATEGIES (EC-LEDS)
Learning Objectives

• Recognize how wind and solar forecasting enhances

power system operations

• Understand how wind and solar forecasts are produced

• Distinguish approaches to implementing forecasting

systems and collecting necessary data

• Identify policy and other actions to support the

implementation of forecasting systems

2
POWER SYSTEM BASICS

3
Power System Objective
Frequency

Supply electric power to

customers
– Reliably
– Economically

Consumption and production

must be balanced
continuously and
instantaneously

Maintaining system frequency is one of the fundamental

drivers of power system reliability
4
Slide credit: B. Kirby
Integrating Variable Renewable Energy (VRE)
Increases Variability and Uncertainty
All power systems (regardless in total power output compared to the clear sky cond
of VRE penetration) Output
changes in from
poweraoutput
5MWduefixed panel
to the in dive
geographic
Gujarat, India
– Variability: Load varies
throughout the day, conventional
generation can often deviate from
schedules
– Uncertainty: Contingencies are
unexpected, load forecast errors
are unexpected
Wind and solar generation
– Variability: Wind and solar
generator outputs vary on
different time scales based on the
intensity of their energy sources
(wind and sun)
– Uncertainty: Wind and solar Source: Hummon et al. (2014). NREL/TP-7A40-60991

generation cannot be predicted

with perfect accuracy 5

Figure 19. Clear sky and synthetic PV power outp

Balancing the System Takes Place at Multiple Timescales

System Load (MW)

0 4 8 12 16 20 24
Time of Day (hr)
seconds to minutes minutes to hours day
Regulation Load Following Economic
dispatch

Days
6
Unit Commitment
Different Sources of Flexibility Help to Address
Variability and Uncertainty

7
Different Sources of Flexibility Help to Address
Variability and Uncertainty

Low capital cost options,

but may require
significant changes to the
institutional context • Numerous options for increasing
flexibility are available in any
power system.

• Flexibility reflects not just physical

systems, but also institutional
frameworks.

• The cost of flexibility options

varies, but institutional changes
may be among the least
expensive.

8
IMPACT OF WIND AND SOLAR
FORECASTING ON POWER
SYSTEM OPERATIONS
9
What is Forecasting?
In this webinar, we use the term forecasting primarily to refer to the near-
term (usually up to day-ahead) prediction of electricity generation from
wind and solar power plants.
Example: Texas’ wind power production forecasts

Actual wind Most recent short-

power term wind power
forecast (updated
hourly)

Day-ahead wind power

forecast

Source: Electricity Reliability Council of Texas short-term wind power forecast

Load forecasting refers to the prediction of electricity demand. 10

Importance of Wind and Solar Forecasting
Example: Solar power under-forecast
• High penetrations of Forecast improvement

variable RE increase
the variability and
uncertainty associated
with power system
operation
• Integrating wind and
solar forecasts into
scheduling and dispatch
operations reduces
uncertainty, helping to
Each figure shows the dispatch stack for one day for the
lower costs and Day-ahead forecast (top row) and Real-time markets
improve reliability (bottom row)
Source: Brancucci Martinez-Anido et al. (2016). Solar Energy 129.

11
 How Do System Operators Use Forecasts? Part 1

• Long-term forecasts (1 week+)

– Estimates of “typical” generation used
for resource and O&M planning
Days

• Day-ahead unit commitment

– Day-ahead forecast, along with
uncertainty band, is fed into scheduling
and market decisions
Time of Day (hr)

• Intra-day adjustments
– Meteorologist flags changes to real-time
traders
– Reconfigure peaking plant schedules

12

Adapted from Justin Sharp

 How Do System Operators Use Forecasts? Part 2

• Hour-ahead scheduling and trading

– Meteorologist provides high/low uncertainty
band
– Use revised forecasts to optimize resources,
markets, transmission; trading with neighbors Time of Day (hr)

• Intra-hour dispatch minutes to

hours
– Value of forecast shifts to control room
– Operators move other generators up/down in
response to fluctuations
– Assess reserves
• Are reserves sufficient to last until next
dispatch interval?
• Can we handle ramps?
• Are peaking resources needed? 13

Adapted from Justin Sharp

How Do System Operators Use Forecasts? Part 2
• A renewable energy ramp is a significant, sustained change in output due to
changing resource conditions (i.e., wind speed, solar irradiance).
• There is no standard definition of what constitutes a ramp in renewable
energy output, and ramps that are important in one system may be trivial in
another.
Wind power forecast
250.00
False positive: True positive:
200.00 Ramp forecasted, Ramp forecasted,
Megawatts (MW)

but no ramp ramp occurred.

150.00
occurred.
100.00

50.00

0.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Operating hour
Day-ahead forecast Actual generation

14
Forecasting Leads to Economic and Operational
Benefits

• Improved unit commitment and dispatch efficiency

– Utilization of least expensive units
– Less “mileage” on operating units
– Less starting of gas turbines and other fast acting units

• Reduced reserve levels

– Regulation reserve
– Flexible/load following reserve

• Decreased curtailment of RE generation

15
The Value of Forecasting: Xcel Energy Case Study

• Leading utility wind provider in

the United States, and top 10
for solar.
– 15% of total energy supply
from wind in 2014
– Up to roughly 70%
instantaneous wind
penetration
– 5,794 MW wind capacity
installed

• Partnered with two national Outcomes:

laboratories to develop a state- – Reduced average forecast error from
of-the art forecasting model, 16.8% in 2009 to 10.10% in 2014
which is maintained by a third – Saved ratepayers US $49.0 million
party over the 2010-2014 period

16
Sources: Staton 2015; NCAR 2015; Xcel Energy 2015
 Factors that Influence Forecasting Benefits

Physical Drivers Institutional Drivers

• The variability of the VRE • System operational
resource practices, market design
• Network size (e.g., scheduling, dispatch)
• Generation resource mix • Regulations and
incentives/penalties
• Penetration level of wind
and/or solar • Forecast timescale,
accuracy, and reliability
• Operator confidence in the
forecast
• How the forecast is used

Forecasting is not a silver bullet! It must be integrated with other

flexibility solutions that favor VRE integration.
17
Adapted from Justin Sharp
 More Frequent Decisions Reduce Uncertainty

While more frequent forecasts provide greater accuracy, they are

only useful to the system operator up to the timeframe in which
actions can be taken in response to the forecast.
18
Source: Harnessing Variable Renewables -- A Guide to the Balancing Challenge, International Energy Agency, 2011
 How are Forecasts Used in System Operations?
Examples from North America
Balancing Authority Type of Forward Unit Intra-day Transmission Reserves Manageme Generation/
variable RE Commitment Unit Congestion nt of Hydro Transmissio
forecasted (Day-ahead, Commitment Management or Gas n Outage
week-ahead, Storage Planning
etc.)
Alberta Electric System Wind X X
Operator
Arizona Public Service Wind X X X

Bonneville Power Wind X X X

Administration (BPA)
California Independent Wind and solar X
System Operator (CAISO)
Glacier Wind Wind X X

Idaho Power Wind X X X X

Northwestern Energy Wind X X X

Sacramento Municipal Solar X

Utility District*
Southern California Wind* and X X X**
Edison* solar
Turlock*** Wind

Xcel Energy Wind and solar X X X X X

* Also participants in the CAISO’s Participating Intermittent Resource Program

** For hydro only, not natural gas 19
*** Uses forecast for trading, optimization, marketing, and compliance with BPA scheduling directives
Source: Porter and Rogers, 2012. Survey of Variable Generation Forecasting in the West.
HOW ARE FORECASTS
PRODUCED?

20
 How are Wind and Solar Forecasts Produced?

Environmental
Prediction Numerical Weather
National

Prediction (NWP) Models

(National Meteorological
Agencies)
Measured Data

Weather data Wind and solar plant data

(Meteorological stations) (Generating units)

Power Prediction Systems

(In-house, vendors) Power Prediction

21

Adapted from Justin Sharp

Forecasting Methods

• Physical (dynamical) methods

– Inputs weather data (temperature, pressure, surface roughness,
obstacles) into numerical weather prediction (NWP) models to
create terrain-specific weather conditions

• Statistical methods Persistence Forecast (Hour-ahead) Actual generation

250.00
– Uses historic and real-time
generation data to statistically 200.00
correct results derived from

Wind output (MW)

NWP models. 150.00

– Persistence forecasting:
100.00
uses the last observation as
the next forecast. 50.00

• Ensemble forecasting 0.00

– Aggregates results from 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Operating hour
multiple different forecasts
22
Producing Forecasts: Timescales, Methods
Type of
Time Horizon Methods
Forecast

Intra-hour 5-60 min Statistical, persistence

Blend of statistical and

Short term 1-6 hours ahead
NWP models
Generation

NWP with corrections

Medium term Day(s) ahead
for systematic biases

Week(s), Seasonal, 1 year Climatological

Long term
or more ahead forecasts, NWP

Ramp forecasting Continuous NWP and statistical

Decision
support

Day ahead, hour-ahead,

Load forecasting Statistical
intra-hour.
23
What Impacts Forecast Quality?

Examples:
• Meteorological data – e.g., density and frequency of
observations
• NWP models – e.g., data assimilation, parameterization
– Resolution—e.g., ability to represent terrain features that impact
RE resource
• Operational information for wind and solar generators –
e.g., turbine or panel availability, curtailment
• Power conversion algorithms

24
 Example: Impact of Terrain and Spatial Resolution
of Model
Terrain can have a significant impact on wind speed and
direction
Altitude (m)
 Example: Impact of Terrain and Spatial Resolution
of Model
Mountain Range appears as
a smooth dip at 27 km spatial
resolution

Contours of rolling terrain

Altitude (m)

not represented at 27 km resolution

27 km

Model Resolution

27 km

Illustrative example adapted from Justin Sharp, Sharply Focused

 Example: Impact of Terrain and Spatial Resolution
of Model

Details of peaks and valleys

not represented at 9 km
resolution
Altitude (m)

9 km

Model Resolution

9 km
27 km

Illustrative example adapted from Justin Sharp, Sharply Focused

 Example: Impact of Terrain and Spatial Resolution
of Model

Peak and valley represented

at 3 km resolution
Altitude (m)

3 km
Model Resolution
3 km
9 km
27 km

Illustrative example adapted from Justin Sharp, Sharply Focused

CONSIDERATIONS FOR
IMPLEMENTING FORECASTING
SYSTEMS
29
Different Roles for Centralized vs. Decentralized
Forecasts
Centralized Forecasting (by Decentralized Forecasting
the system operator) (by the generator)
• Enables the use of • Used by off-takers when
forecasting in unit making offers
commitment and dispatch • Helps project operators
• Requires mechanisms to optimize operation and
obtain data from generators maintenance
and encourage data quality • Informs operators of
• Allows greater consistency potential transmission
and reduces uncertainty at congestion
the system level • Limited scope can decrease
utility
Centralized forecasting by the system operator, supported
by generator-level forecasts from the plant operator, is
widely considered a best practice approach. 30
Who Accrues the Benefits of Improved Forecasting
(and Bears the Risks of Poor Forecasting)?
Utilities Variable Generation Plant Owners

Consumers Independent System Operators (ISOs)

Those who bear the risks (financial, reliability) have the

31
strongest interest in improving wind and solar forecasting
 Understanding How Forecasts Can Impact Decisions

• Who will use forecasts, and how will they use them?
– Market operations/scheduling
– Transmission and distribution system operations?

• What time intervals and horizons are needed?

– Hourly, 15 minute, day ahead, day of, hour ahead, etc.

• Point estimates Day-ahead forecast

(deterministic forecasts) 300

can provide false sense of 250

certainty Megawatts (MW) 200

150

• Estimates of forecast 100

uncertainty (probabilistic 50

0
methods) can be extremely 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
-50
useful if the system has a Hour
good way of using the Point forecast Probabalistic forecast
additional information 32
 Monitoring and Verification is an Essential
Component of Forecasting

Purposes of forecast verification:

1. Monitor forecast quality - how accurate are the
forecasts and are they improving over time?
2. Improve forecast quality - the first step toward getting
better is discovering what you're doing wrong.
3. Compare the quality of different forecast systems
and a baseline - to what extent does one forecast
system give better forecasts than another, and in what
ways is that system better?
4. Financial verification – ensuring that generator
reporting matches actual conditions

33

Adapted from Justin Sharp

Common Forecast Metrics
Forecast error is the difference between predicted and real-time
generation from non-dispatchable VRE resources

• Mean bias error (MBE)

– Indicates whether the model is systematically under- or over-
forecasting
• Mean absolute error (MAE)
– Measures the average accuracy of forecasts without considering
error direction
• Root mean square error (RMSE)
– Measures the average accuracy of forecasts without considering
error direction and gives a relatively high weight to large errors

All metrics are wrong, but some are useful

Error rates are not static; they vary based on time of year, extent of spatial or
34
geographic aggregation, among many other factors
Where are Improvements Needed?

Curtailment Dispatch Regulation Dispatch Unit Start Up

Forecast error (MW)

The system was designed to accommodate small amounts of

35
uncertainty. Large forecast errors are the most costly.
Data Collection Strategies for System Operators

• Policy mandates
Xcel Energy’s Model PPA includes
– May be implemented for utility- provisions for forecasting data collection
scale and distributed
generation
– FERC Order 764
• Interconnection or market
requirements set by federal
and state government, utilities,
and RTO/ISO
• Power purchase agreements
• Penalties/rewards
• Partnerships with
meteorological agencies
• Third-party vendors
http://www.xcelenergy.com/staticfiles/xe/Corporate/Corp
orate%20PDFs/Model_Power_Purchase_Agreement.pdf
36
What Data is Needed to Set up a Forecasting System?

Static Dynamic
• Plant location (Latitude, • Real-Time Generation
Longitude) • Availability Data
• Installed Capacity • Park Potential (potential
• Historic Data (training total output based on
data) available resources at the
wind/solar farm level)
• Meteorological Data

37
Options for Procuring Forecasts

• Third-Party Vendors • In-house Forecasting

– Vendor uses proprietary power – Staff meteorologists/analysts
prediction models to estimate develop power prediction
generation. models and are responsible
– Requires wind and solar plant data for assembling and validating
from generators or the system meteorological and plant data.
operator. – Can allow flexibility for custom
– In-house meteorologists still play a and state-of-the-art
role in reviewing forecasts and approaches that reflect
identifying critical periods. system-specific concerns.
– Typical forecasts are selling for – Will be significantly more
USD $200/project/month to expensive to develop and
$2000/project/month (source: Justin require significantly more
Sharp) computing power and
expertise to maintain than
vendor forecasts.

Working with vendors can be an inexpensive, introductory way to

get experience with forecasting. This can be a valuable first step. 38
ACTIONS TO SUPPORT
FORECASTING SYSTEMS

39
Example Areas for Early Actions

• Update interconnection standards, power purchase

agreements to enable data gathering

• Work with national meteorological institutes to improve

underlying weather data or access to it

• Facilitate training of operators on meteorology, how to

interpret forecasts, and work with vendors

• Support vendor trials and development of a smooth IT

interface between forecast vendors and users

40
Key Takeaways

• Forecasting facilitates the integration of variable renewable energy

to the grid by reducing uncertainty and improving the efficiency of
operations at multiple timescales.

• Better information is only valuable when it leads to better decisions

– Understanding areas where forecasting improves decision-making is a
first step in considering how to implement forecasting systems
– Interpreting forecasts is a critical element of effective implementation

• Centralized and decentralized have unique value; in general,

moving toward centralized forecasting is most effective in reducing
uncertainty at the system level.

• There is no one-size-fits-all approach to collecting data and

procuring and monitoring forecasts. Power systems should tailor
their forecasting programs to their unique context and needs.
41
QUESTIONS AND DISCUSSION

42
Contacts and Additional Information

Webinar Panel
Bri-Mathias Hodge, Ph.D. Jennifer Leisch, Ph.D.
National Renewable Energy United States Agency for
Laboratory International Development
Email: Email: [email protected]
[email protected]

Greening the Grid

greeningthegrid.org
Email: [email protected]

43
Learn more at greeningthegrid.org

44