probabilistic wind power forecasting in electricity market
TRANSCRIPT
FERC Technical Conference on Increasing Real-Time and Day-Ahead Market Efficiency Through Improved Software, June 28-30 2011
Probabilistic Wind Power Forecasting in Electricity Market Operations: a Case Study of Illinois
Zhi Zhou*, Audun Botterud*, Jianhui Wang Argonne National Laboratory, USA [email protected]; [email protected]
Ricardo Bessa, Hrvoje Keko, Jean Sumaili, Vladimiro Miranda INESC Porto, Portugal
Project website: http://www.dis.anl.gov/projects/windpowerforecasting.html
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
2
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
3
4
Project Overview: Wind Power Forecasting and Electricity Markets
Collaborators: Institute for Systems and Computer Engineering of Porto
(INESC Porto), Portugal
Industry Partners: Horizon Wind Energy and Midwest ISO (MISO)
Sponsor: U.S. Dept. of Energy (Wind and Water Power Program)
The project consists of two main parts:
Wind power forecasting – Review and assess existing methodologies
– Develop and test new and improved algorithms
Integration of forecasts into operations (power system and wind power plants) – Review and assess current practices
– Propose and test new and improved approaches, methods and criteria
Goal: To contribute to efficient large-scale integration of wind power by
developing improved wind forecasting methods and better integration of
advanced wind power forecasts into system and plant operations.
http://www.dis.anl.gov/projects/windpowerforecasting.html
Background and Motivation – U.S. Wind Power Capacity
Wind power has been rapidly integrated into the current power systems
5
Background and Motivation - Handling Uncertainties in System/Market Operation
What are the impacts on the system?
– Reliability (curtailment,..)
– Efficiency (system cost, price..)
Source of
uncertainty
Operating
Reserve
Δ Load
Δ Generating
capacity
Operating Reserves
(spin and non-spin)
Δ Wind
Power
??
??
Increase operating
reserves?
Change commitment
strategy?
- Stochastic UC
[MW]
6
Wind power
forecasting
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation and reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
7
Probabilistic forecasting with kernel density estimation
Conditional wind power probabilistic forecasting
Kernel density estimation (KDE)
8
tktX
tktktXP
tktktPxf
xpfxXpf
|
|,
|
,|
Joint or multivariate density function
of p and x
Marginal density of x
Quantile-Copula Estimator for Conditional KDE
9
Copula Definition multivariate distribution function separated in:
•marginal functions
•dependency structure between the marginal,
modeled by the copula
copula density function
KDE ESTIMATOR
Ui=FXe(Xi) and Vi=FY
e(Yi)
KDE ESTIMATOR
empirical cum. dist.
R. Bessa, et. al. “Quantile-copula density forecast for wind power uncertainty
modeling,” Proceedings IEEE Trondheim PowerTech 2011, Trondheim Norway, 2011.
Illustration of Kernel Density Forecast
10
Forecast the wind power pdf at time step t for each look-ahead time step
t+k of a given time-horizon knowing a set of explanatory variables (NWP
forecasts, wind power measured values, hour of the day)
0
0.2
0.4
0.6
0.8
1
0
5
10
15
20
Win
d S
pee
d (m
/s)
Wind Power (p.u.)
Scenario Generation and Reduction
Kernel Density Forecast (KDF) methods (e.g. Quantile-copula in the IL case study)
produce pdf forecasts of the wind power generation
Stochastic unit commitment model requires scenario representation of wind power
forecast → account for the temporal correlation of forecast errors
A large number of scenarios generated with Monte-Carlo simulation based on quantile
distribution (multivariate Gaussian error variable, covariance matrix) [Pinson et al. 09]
Three scenario reduction methods
– Random selection
– Scenario reduction method in GAMS [Gröwe-Kuska, Heitsch, et. al, 2003] (used in the IL case study)
– Scenario clustering approach [Sumaili et al. 2011]
11
A. Probabilistic forecast
(KDF)
Scenario Generation and Reduction - Illustration
12
B. Large scenario set
C. Reduced scenario set
(scenarios with different
probabilities)
Scenario Reduction Reduces Variance of Scenario Set
13
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46
Sce
nar
io V
aria
nce
[p.u
.]
Time [hour]
1000
10-SR1
10-SR2
10-SR3
0
0.005
0.01
0.015
0.02
0.025
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46
Sce
na
rio
Va
ria
nce
[p
.u.]
Time [hour]
1000
100-SR1
100-SR2
100-SR3
10 reduced scenarios: 100 reduced scenarios
SR1 - Random selection
SR2 - ScenRed GAMS
SR3 - Scenario clustering
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation and reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
14
Steps in U.S. Electricity Market Operations (based on Midwest ISO)
1100 1600 1700 Submit
DA bids
Clear DA market
using UC/ED
Day ahead:
Post-DA
Reliability UC
Submit
revised
bids
DA – day ahead
RT – real time
UC – unit commitment
ED – economic dispatch
Operating day:
-30min Operating hour Submit
RT bids
Clear RT market using
ED (every 5 min) Intraday
Reliability UC
Post operating
reserve requirements
Post results
(DA energy
and reserves)
Post results (RT
energy and reserves)
ISO/RTO
Forecast
15
A Stochastic Unit Commitment (UC) Model w/Wind Power Uncertainty
Formulation using wind power forecast scenarios (s) w/probabilities (probs):
A two-stage stochastic mixed integer linear programming (MILP) problem
– First-stage: commitment
– Second-stage: dispatch
Objective function (min
daily expected cost)
Energy balance (hourly)
Spinning Reserve balance
(hourly)
Unit commitment constraints
(ramp, min. up/down)
Z. Zhou, A. Botterud, J. Wang, R.J. Bessa, H. Keko, J. Sumaili, V. Miranda, “Application of
Probabilistic Wind Power Forecasting in Electricity Markets”, submitted 16
𝑀𝑖𝑛 𝑝𝑟𝑜𝑏𝑠𝑠
∙ 𝐹𝐶𝑡,𝑖𝑠 + 𝐶(𝑅𝑁𝑆𝑡
𝑠) + 𝐶(𝐸𝑁𝑆𝑡𝑠)
𝑡,𝑖
+ 𝑆𝐶𝑡,𝑖𝑡,𝑖
𝑔𝑒𝑛𝑡ℎ𝑒𝑟𝑚𝑎𝑙,𝑖,𝑡𝑠
𝑖 + 𝑔𝑒𝑛𝑤𝑖𝑛𝑑,𝑡𝑠 = 𝑙𝑜𝑎𝑑𝑡 − 𝐸𝑁𝑆𝑡
𝑠, ∀ 𝑡, 𝑠
𝑠𝑟𝑡ℎ𝑒𝑟𝑚𝑎𝑙,𝑖,𝑡𝑠
𝑖
≥ αsr (𝑂𝑅𝑟𝑒𝑔,𝑡 + 𝑂𝑅𝑤𝑖𝑛𝑑,𝑡𝑠 ) − 𝑆𝑅𝑁𝑆𝑡
𝑠, ∀ 𝑡, 𝑠
𝑛𝑠𝑟𝑡ℎ𝑒𝑟𝑚𝑎𝑙,𝑖,𝑡𝑠
𝑖
≥ (1 − 𝛼𝑠𝑟) (𝑂𝑅𝑟𝑒𝑔,𝑡 + 𝑂𝑅𝑤𝑖𝑛𝑑,𝑡𝑠 ) − 𝑁𝑆𝑅𝑁𝑆𝑡
𝑠, ∀ 𝑡, 𝑠 Non-spinning Reserve
balance (hourly)
Commitment Constraints (i, t)
Wang J, Botterud A, Bessa R, Keko H, Carvalho L, Issicaba D, Sumaili J, and Miranda V,
Wind power forecasting uncertainty and unit commitment, Applied Energy, in press, 2011.
Operating Reserves vs. Stochastic UC
Hourly operating reserve requirement
(spinning + non-spinning) +
Deterministic UC
17
Forecast quantiles Reduced scenario set
Stochastic UC +
scenario set
Commitment
schedule
Commitment
schedule
Real-time
dispatch Real-time
dispatch
Realized generation
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation and reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
18
Case Study Assumptions
210 thermal units: 41,380 MW
– Base, intermediate, peak units
Wind power: 14,000 MW
– 2006 wind series from 15 sites in Illinois
(NREL EWITS dataset)
– 20% of load
Peak load: 37,419 MW
– 2006 load series from Illinois
No transmission network
120 days simulation period (July 1st to
October 31st, 2006)
– Day-ahead unit commitment w/wind power
point forecast
– Real-time reliability assessment commitment
(RAC) w/ wind power scenarios
19
Case study focus is to compare:
-Operating reserves vs. stochastic UC
-Probabilistic forecasting methods
0
5000
10000
15000
20000
25000
30000
35000
40000
111
122
133
144
155
166
177
188
199
111
01
12
11
13
21
14
31
15
41
16
51
17
61
18
71
19
81
20
91
22
01
23
11
24
21
25
31
26
41
27
51
28
61
Lo
ad
/Win
d P
ow
er (
MW
)
Hour
Wind and Load in July-October 2006
Load
Wind
4.78%
20.60%
19.71%
30.95%
Generation Capacity
Combine Cycle Turbine
Gas Turbine
Nuclear
Steam Turbine
Market Simulation Set-Up
20
Day-ahead
point forecast
4hr-ahead
probabilistic forecast
Actual
wind power
UC ED ED RAC
Real-time
dispatch and price
Day-ahead
schedule and price
UC Case Study: Deterministic and Stochastic Cases
21
* This additional reserve is applied at the RAC stage only to handle wind power uncertainty. All cases use
a regular reserve, 𝑂𝑅𝑟𝑒𝑔,𝑡, equal to the largest contingency ( 1146 MW).
Case Add’l Reserve: 𝑂𝑅𝑤𝑖𝑛𝑑,𝑡* Forecast UC strategy at RAC
stage
P1 None Perfect in both DA and RT Deterministic
PF-F0 None 50% quantile Deterministic
PF-F1 Fixed: avg. 50-10% quantile 50% quantile Deterministic
PF-F2 Fixed: avg. 50-5% quantile 50% quantile Deterministic
PF-F3 Fixed: avg. 50-1% quantile 50% quantile Deterministic
PF-D1 Dynamic: 50-10% quantile 50% quantile Deterministic
PF-D2 Dynamic: 50-5% quantile 50% quantile Deterministic
PF-D3 Dynamic: 50-1% quantile 50% quantile Deterministic
SF-S0 None 10 Scenarios Stochastic
SF-S1 10% of wind scenario 10 Scenarios Stochastic
SF-S2 20% of wind scenario 10 Scenarios Stochastic
0
200
400
600
800
1000
1200
1400
1600
P1 PF-F0 PF-F1 PF-F2 PF-F3 PF-D1 PF-D2 PF-D3 SF-S0 SF-S1 SF-S2
Cost
(M
$)
Cases
Costs
Unserved load
Unserved nonspinning reserve
Unserved spinning reserve
Start-up
Fuel
Overview of total cost (4-months period )
22
Point forecast with no additional reserve too risky
Stochastic unit commitment has the lowest total costs
Dynamic reserves perform slightly better than fixed reserves
Overall, more operating reserves lead to lower costs within the same categories
Overview of generation cost (4-months period)
23
520
540
560
580
600
620
640
660
P1 PF-F0 PF-F1 PF-F2 PF-F3 PF-D1 PF-D2 PF-D3 SF-S0 SF-S1 SF-S2
Gen
erati
on
Cost
s (M
$)
Cases
Total Generation Costs
Start-up
Fuel
Stochastic UC model has slightly higher generation costs
Additional generation costs are more than offset by the reduced curtailment costs
Total curtailment on load and reserve (4-months period)
24
Same trend on curtailment of load and spinning reserve.
More load curtailments in cases with fixed reserve strategies
More spinning reserve curtailment in cases with dynamic reserve strategies
Least curtailment on both load and spinning reserve in cases with stochastic UC
Selected Over-forecasted Day (October 19th, 2006)
25
0
2000
4000
6000
8000
10000
12000
1 3 5 7 9 11 13 15 17 19 21 23
Win
d p
ow
er (
MW
)
Time (Hour)
S1
S2
S3
S4
S5
S6
S7
S8
S9
S10
4 hour ahead
Real
0
200
400
600
800
1000
1200
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Lo
ad
Cu
rta
ilm
ent
(MW
)
Time (Hour)
P1
PF-F0
PF-F2
PF-D2
SF-S2
0
500
1000
1500
2000
2500
3000
3500
4000
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Ho
urly
En
erg
y P
ric
e (
$/M
WH
ou
r)
Time (Hour)
P1
PF-F0
PF-F2
PF-D2
SF-S2
Efficiency (clearing prices) and reliability (load curtailment)
Selected Under-forecasted Day (September 22nd, 2006)
26
Efficiency (clearing prices and wind curtailment)
0
2000
4000
6000
8000
10000
12000
14000
16000
1 3 5 7 9 11 13 15 17 19 21 23
Win
d P
ow
er (
MW
)
Time (Hour)
S1
S2
S3
S4
S5
S6
S7
S8
S9
S10
4 hour ahead
Real
0
5
10
15
20
25
30
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
RT
Ho
urly
En
erg
y P
ric
e (
$/M
WH
ou
r)
Time (Hour)
PF-F0
PF-F1
PF-F2
PF-D1
PF-D2
SF-S0
SF-S1
SF-S2
0
200
400
600
800
1000
1200
1400
1 3 5 7 9 11 13 15 17 19 21 23
Win
d P
ow
er C
urta
ilm
en
t (M
W)
Time (Hour)
PF-F0
PF-F1
PF-F2
PF-D1
PF-D2
SF-S0
SF-S1
SF-S2
Outline
Background and Motivation
Wind power forecasting
- Probabilistic density forecasting
- Scenario generation and reduction
System operation with wind power uncertainty
- Two-settlement market
- Stochastic unit commitment
Test Case
- IL Power System
- System operation analysis
Conclusion and future work
27
Conclusions
Probabilistic wind power forecasts can contribute to efficiently schedule
energy and operating reserves under uncertainty in wind power generation
Dynamic operating reserves (derived from forecast quantiles)
+ Well aligned with current operating procedures
+ Lower computational burden
+ Lower cost and increased reliability
- Does not capture inter-temporal events
- Uncertainty not represented in objective function
28
Stochastic unit commitment (with forecast scenarios) + Captures inter-temporal events through scenarios
+ Explicit representation of uncertainty in objective function
+ Lower cost and increased reliability
- More radical departure from current operating procedures
- High computational burden
Conclusions
Others
– Dynamic operating reserves and stochastic UC give similar results in the IL Test Case
– Inaccurate forecasts can lead to large implications for system efficiency and reliability.
29
Comments and Questions.
30
Contacts: Zhi Zhou*, Audun Botterud*, Jianhui Wang Argonne National Laboratory, USA [email protected]; [email protected]
Ricardo Bessa, Hrvoje Keko, Jean Sumaili, Vladimiro Miranda INESC Porto, Portugal
Project website:
http://www.dis.anl.gov/projects/windpowerforecasting.html