Delft-FEWS

1. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Operating the Federal Columbia River
Power System in a World of Uncertainty
Christopher R Allen
Short Term Planning and Operations
Operations Research Analyst

2. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N

3. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Agency Background
Bonneville Power Administration (BPA) is a federal nonprofit
agency based in the Pacific Northwest region of the United
States.
BPA was established in the 1932 as an agency within the U.S
Department of Energy. The agency is largely responsible for
marketing wholesale power from:
• 31 multipurpose hydro electric dams; 22 GW
• 1 nuclear plant; 1.1 GW
• several smaller sources
BPA also operates and maintains about three-fourths of the
high-voltage transmission in the region

4. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
 Basin 614,000 km2 (approximately the size of France)

5. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Grand Coulee Dam, 6.4 GW (est 1942)
Irrigation Facility; 708 m3/s

6. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Bonneville Dam, 1.08 GW (est 1937)

7. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Multipurpose dams are operate by the U.S Army Corps of
Engineers and U.S Bureau of Reclamation:
• Flood Protection
• Environmental (Endangered Species & Clean Water Acts)
• Navigation
• Irrigation
• Recreation
• Variable resource integration (wind resources)
• Reliable electric power
Operations mission is to maintain high level of benefits while
minimizing and mitigating the harm to fish and wildlife in the
face of growing uncertainty and decreasing operational
flexibility
Federal Columbia River Power System

8. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Key Regional Stakeholders
• States of Oregon, Washington, Idaho and Montana
• Canada
• Tribal Nations
• Preference Power Customers
• Private resource owners (hydro electric)
• Recreator(s)
• Irrigators
• National Oceanic and Atmospheric Administration
Planning and coordinating the operation of the FCRPS is
very complex and involves many different competing
interests.
Federal Columbia River Power System

9. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Federal Columbia River Power System
Annual Water Year (October – September)
– Average runoff 164 km3 (Range: 98-241 km3) or 5200 m3/s on
average as measured at The Dalles
– Spring runoff (April – June)
Federal storage about 37 km3 (1/4 annual runoff)
– Colorado and Missouri basins can store x2-x3 their annual runoff
Geographical variability between snowpack sources
– Significant influence on spring runoff shape (timing & magnitude)
– Minimizing spill at the confluence of Columbia and Snake River
systems is challenging during peak runoff period

10. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
October-December
Store water through October in preparation for fisheries
operation. Release flows to support fisheries objectives.
January-April
Draft system for flood control while maintaining fisheries
spawning habitat elevations. (Starting in April; spilling to support
downstream migration of juvenile fish. Energy loss 1500 aMW)
May – June
Refill the system on the Spring freshet while meeting flow
objectives for downstream fish migration.
July – August
Draft the system to augment flows supporting the downstream
migration of juvenile fish.
September – October
Operate the system to prepare for Fall fisheries operations at
Grand Coulee, Vernita Bar, and downstream of Bonneville.
FCRPS Annual Operations Cycle

11. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Power production is driven by the need to manage water to
meet objectives, which can often conflict, in the most
economic way possible while meeting load obligations.
– Long-term and short-term energy markets are used to buy and sell
energy necessary to meet load obligations and to shape energy
production to meet operational objectives
– There must be sufficient flexibility in our resources and in the
marketplace to handle uncertainties
Capacity (flexibility) is set transmission reserve obligations:
– 1600 MW Incremental (INC) reserves: capacity that is reserved
for when wind generation falls below the schedule or load
increases within an hour
– 1100 MW Decremental (DEC) reserves: generation above
minimum that is reserved for when wind generation is above the
schedule or load drops within an hour
Federal Columbia River Power System

12. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Uncertainty Today and Tomorrow
Uncertainty sources in daily operation
– Meteorological and model related:
• Meteorological (temperature, precipitation) incorporated in the
streamflow and load forecasts
• Streamflow related to BPA’s hydrological models to forecast annual
runoff and day-to-day short-term forecasts
• Loads related to BPA’s load models
– Power system
• Ancillary Services: balancing and contingency deployments
• Resource Availability: unit outages and wind resource generation
• Power Contracts: dependent upon uncertain power market prices
• Market Liquidity: sufficient depth to manage objectives
– Non-federal reservoir operations: Private Utilities and Canada
– Operational constraints: flood control, environmental, etc…
Increasing environmental requirements will likely reduce
FCRPS capability and operational flexibility in the future

13. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Lower Snake deterministic streamflow forecast
Variability in deterministic forecasts can have tremendous operational
consequences. (Inflow forecasts ranged between 2290 m3/s and
5950 m3/s in a 5-day span.)
Lower Granite Inflow Forecasts June 2 -9, 2010
50
70
90
110
130
150
170
190
210
230
6/2 6/3 6/4 6/5 6/6 6/7 6/8 6/9 6/10 6/11 6/12 6/13
KCFS
F2 F3 F4 F7

14. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Uncertainty in Reserves Deployment

15. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Responding to future uncertainty
Develop measures of Operational Flexibility
– The ability of resources to respond to changing conditions
More operational constraints = less operational flexibility
Develop strategies to quantify sources of uncertainty
-Meteo, streamflow, loads forecasts
-Reserve obligations/deployments
-Resource availability
Explore market opportunities
– The ability of the marketplace to supply sufficient amounts and
types of energy and capacity products. (New flexible resources or
innovative market solutions like Energy Imbalance Market)
BPA Technology and Innovation projects aim to develop
strategies to assess sources of uncertainty and measures
of operational flexibility

16. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Improve Short-term Decision Support Tools
Enhancement of the production system
– Flexible data-model integration platform based on Delft-FEWS
– Replacement of the current deterministic optimization model
– Optimization models: RTC-Tools (Deltares)
Investments in Technology and Innovation projects:
1)Short-Term Hydropower Production and Marketing Optimization
(HyProM) (Deltares & Fraunhofer IOSB-AST)
2)Development of a State-of-the-Art Computational Framework and Platform for
the Optimal Control of Multireservoir Systems Under Uncertainty
3)Computationally Efficient, Flexible, Short-Term Hydropower Optimization and
Uncertainty Analysis for the BPA System
Flexible data-model integration platform will enable BPA to
more easily integrate advancements of TI projects.

17. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
TIP 259: Short-Term Hydropower Production and
Marketing Optimization (HyProM)
– Probabilistic short-term optimization is a main research topic
and enables a better consideration of forecast uncertainty
from multiple sources (meteo, streamflow, load)
– Integrated short-term management of hydropower
production and marketing; 21 day forecast horizon
– Evolution from deterministic to stochastic optimization
techniques
– Development of tools for integrated management of
scenarios of streamflow, wind generation and load and
meteo-related uncertainty

18. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
TIP 259: Short-Term Hydropower Production and
Marketing Optimization (HyProM)
common
control on both
branches
time
stream flow,
load etc.
forecast time
cold front with
30% probability
more stream flow &
wind, less load, low price
probability = 30%
less stream flow & wind,
more load, high price
probability = 70%
not sensitive to
constraints, high
flexibility (green)
sensitive to constraints, low
operational flexibility (red)

19. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Scenario Trees to Model Decision Making under
Uncertainty
Decision Uncertainty Resolution Decision
Once uncertainty is resolved, it is possible to adopt the control
strategy optimal to the remaining scenario !!!
RainRain
No RainNo Rain
t
P
1 2 3 40 t
P
1 2 3 40
t
P
0 1 2 3 4 t
P
0 1 2 3 4disturbance
control
No Rain
Rain
before
uncertainty resolution
after
uncertainty resolution
t
u
0 1 2 3 4 t
u
0 1 2 3 4

20. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Scenario Trees to Model Decision Making under
Uncertainty
Decision Uncertainty Resolution Decision
Once uncertainty is resolved, it is possible to adopt the control
strategy optimal to the remaining scenario !!!
RainRain
No RainNo Rain
t
P
1 2 3 40 t
P
1 2 3 40
t
P
0 1 2 3 4 t
P
0 1 2 3 4disturbance
control
No Rain
Rain
before
uncertainty resolution
after
uncertainty resolution
t
u
0 1 2 3 4 t
u
0 1 2 3 4
Branching Point
A tree specifies when uncertainty is resolved
Control Tree
t
u
0 1 2 3 4 t
u
0 1 2 3 4
Rain
No RainNo Rain

21. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Meteo Versus Model Uncertainty (Orofino, Snake Basin)
• Largest uncertainty of prob. forecast during snow melt
• Overestimation of model uncertainty because of missing
(manual) data assimilation in hindcast experiment

22. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Streamflow Versus Load Uncertainty (Snake Basin)
• Streamflow uncertainty is significantly larger than the
load uncertainty in the period January – June
• Load uncertainty is larger in the winter half, but relatively
small

23. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Forecast Convergence Score (FCS)
• Measures the change from one forecast to another
• Probabilistic forecast is more stable and should lead to
more stable decisions

24. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
From Ensemble to Scenario Tree
similar ensemble members
and their probabilities get
aggregated (dedicated FEWS
display is on its way)

25. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Forecast Uncertainty Propagating through the system
forebay elevation
(system state)
total outflow
(control variable)

26. B O N N E V I L L E P O W E R A D M I N I S T R A T I O N
Conclusions and Outlook
– Probabilistic forecast & stochastic optimization leads to
better and more stable decisions compared to the
deterministic approach
– Scenario tree based optimization technique translates
forecast uncertainty into a distribution of control trajectories
and system states; objectives and constraints enable the
decision maker to direct the system
Next steps:
– FEWS based production system
– The integration with an energy marketing model is on its
way to add another dimension of operational flexibility by
approaching the energy market
– R&D (HYPROM) will be delivered in April 2015