The (Rocky) Path to 80 percent Renewables STEP Seminar Series Princeton University April 14, 2014 Warren B.
Download
Report
Transcript The (Rocky) Path to 80 percent Renewables STEP Seminar Series Princeton University April 14, 2014 Warren B.
The (Rocky) Path to 80 percent Renewables
STEP Seminar Series
Princeton University
April 14, 2014
Warren B. Powell
Hugo P. Simao
PENSA Laboratory
Princeton University
http://energysystems.princeton.edu
Slide 1
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 2
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 3
Hydroelectric production
Biomass production
Wind in the U.S.
99.9 percent from renewables!
Wind &
Solar
Battery
Storage
Fossil
Backup
750 GWhr battery!
20 GW
News flash – Oct 29, 2013
Dealing with uncertainty
Available at energysystems.princeton.edu
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 12
© 2013 Warren B. Powell
Energy from wind
Wind farms on the PJM system
Energy from wind
Wind power from all PJM wind farms
1 year
Jan
Feb
March April
May June July Aug Sept
© 2010 Warren B. Powell
Oct
Nov
Dec
Slide 17
Energy from wind
Wind from all PJM wind farms
30 days
© 2010 Warren B. Powell
Slide 18
Modeling wind
Forecast vs. actual for a single wind farm
Actual
Forecasted
Wind energy in PJM
Forecast vs. actual for all wind farms in PJM
Forecasting wind
Rolling 24-hour forecast of PJM wind farms
Solar energy
Princeton solar array
Solar energy
Princeton solar array
PSE&G solar farms
Solar output over entire year (all farms)
Sept
Oct
Nov
Dec
Jan
Feb
March
April
May
June
July Aug
Solar from PSE&G solar farms
Solar from a single solar farm
Solar from PSE&G solar farms
Within-day sample trajectories
Solar from PSE&G solar farms
Solar from all PSE&G solar farms
Rainfall
Foz do Iguaçu (Brazil) – 2011 through 2013
2011
2012
2013
Commodity prices
The price of natural gas
» Reflects global and local economies, competing global
commodities (primarily oil), policies (e.g. toward
CO2), and technology (e.g. fracking).
$120 /mmBTU!
$4 /mmBTU
LMPs – Locational marginal prices
Locational marginal prices on the grid
$58.47/MW
LMPs – Locational marginal prices
Locational marginal prices on the grid
$977/MW !!!
LMPs – Locational marginal prices
Locational marginal prices on the grid
$328/MW !
Locational marginal prices on the grid
$52/MWhr
Uncertainty
It is important to separate:
» Predictable variability
PJM load
• Diurnal cycles
• Large weather patterns
• Major human events (Super
bowl)
Aggregate solar
» Stochastic uncertainty
• Temperature deviations from
forecast
• Late/early arrival of a storm
• Generator failures
• Wind shifts
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 37
Wind energy in PJM
Total PJM load plus actual wind (July)
53 wind farms
Wind energy in PJM
Total PJM load plus actual wind (July)
Wind ~ 37 percent of total load
100GW
101,000 MWhr battery
$50 billion!!
Solar energy
Solar from all PSE&G solar farms
Solar energy
Total PJM load plus factored solar (July)
Solar ~ 15 percent of total load
Combining wind and solar
Mixture of wind and solar to meet July load
815,000 MWhr battery
$989 billion!!
Combining wind and solar
260,000 MWhr battery
$130 billion!!
Mixture of wind and solar to meet July load
100GW
Solar energy
The California “duck”
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 45
© 2013 Warren B. Powell
The timing of decisions
Day-ahead planning (slow – predominantly steam)
Near-term planning (fast – gas turbines)
Real-time planning (economic dispatch)
The timing of decisions
The day-ahead unit commitment problem
Noon
Midnight
Midnight
Midnight
Midnight
Noon
Noon
Noon
The timing of decisions
The day-ahead unit commitment problem
12pm
1 2 3
12am
Noon to midnight:
Slow generators committed the day before
Plan, but do not commit to fast generators
The timing of decisions
The day-ahead unit commitment problem
Midnight to midnight:
Plan and commit to slow generators
Plan, but do not commit to fast generators
The timing of decisions
The day-ahead unit commitment problem
Midnight to, say, 4am the next day
Plan, but do not commit to any generator
Solved only to minimize end-of-day truncation
error.
The timing of decisions
Intermediate-term unit commitment problem
2:00 pm
1:00 pm
1:45 pm
1:15 pm
1:30
3:00 pm
2:15 pm
2:30 pm
The timing of decisions
Intermediate-term unit commitment problem
2:00 pm
1:00 pm
1:45 pm
1:15 pm
1:30
3:00 pm
2:15 pm
2:30 pm
The timing of decisions
Intermediate-term unit commitment problem
2 pm
4 pm
1:45pm – 2pm
All generators committed the day before or the ½ hour before
The timing of decisions
Intermediate-term unit commitment problem
2:00pm – 2:30pm
Plan and commit to fast generators whose notification+
startup times fall within this window
The timing of decisions
Intermediate-term unit commitment problem
2:30pm – 4:00pm
Plan, but do not commit
The timing of decisions
Intermediate-term unit commitment problem
t1
t3
2
Turbine 2
Turbine 3
Turbine 1
The timing of decisions
Real-time economic dispatch problem
2pm
1pm
1:05 1:10 1:15 1:20 1:25 1:30
The timing of decisions
Real-time economic dispatch problem
2pm
1pm
1:05 1:10 1:15 1:20 1:25 1:30
Slow
generators
committed
thebefore
day before
SlowSlow
generators
Slow
generators
committed
committed
the day
thebefore
day
generators
committed
thebefore
day
Fast
generators
committed
½ hour
before
Fast generators
Fast generators
committed
committed
the ½
hour
before
½the
hour
before
Fast generators
committed
the
½the
hour
before
Optimize
within
andreserve
spinning
reserve
mar
Optimize
within
Optimize
operational
within
operational
andoperational
spinning
andreserve
spinning
margins
reserve
margins
Optimize
within
operational
and spinning
margins
Lecture outline
80 percent from wind and solar?
The uncertainties of energy
A spreadsheet model
The PJM grid and planning process
SMART-ISO – Modeling the PJM energy markets
© 2010 Warren B. Powell
Slide 60
Lecture outline
SMART-ISO – Modeling the PJM energy markets
Modeling and designing robust policies
Calibration and LMPs
Modeling offshore wind
Mid-Atlantic Offshore Wind Integration Study
© 2010 Warren B. Powell
Slide 61
Designing a policy
Dealing with uncertainty
» We have to design policies to manage the different
forms of uncertainty.
» We do this by looking for robust policies, which are
rules for making decisions.
» We write our optimization problem in the form:
T
t
min E C St , X ( St )
t 0
“policy”
(rule for making a decision)
Day-ahead, hour- “simulator” where
ahead and real-time St 1 S M St , X ( St ),Wt 1
decisions
Averaging over
multiple samples
Designing a policy
The challenge
» We need to design an implementable policy that is more
robust to the uncertainty of rainfall, wind, solar,
temperature, and market behavior.
» Our approach has been to start with the process already in
use by PJM, and make the simplest changes that produce a
more robust policy.
» We draw on our research on a wide range of applications
where we have to make decisions under uncertainty.
» We will do our best to make our policies look
sophisticated and complicated, but don’t be fooled. We
are building on industry-standard methods from the U.S.
Designing a policy
1) Policy function approximations (PFAs)
» Lookup tables, rules, parametric functions
2) Cost function approximation (CFAs)
» X CFA ( St | ) arg min x X
t
C
( St , xt | )
t ( )
3) Policies based on value function approximations (VFAs)
» X tVFA ( St ) arg min x C ( St , xt ) Vt x Stx ( St , xt )
t
4) Lookahead policies
» Deterministic lookahead:
X tLA-D (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
åg
t '-t
C(Stt ' , xtt ' )
t '=t+1
» Stochastic lookahead (e.g. stochastic trees)
X tLA-S (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
p(w ) å g
å
w
ÎWt
t '=t+1
t '-t
C(Stt ' (w ), xtt ' (w ))
Lookahead policies
Lookahead policies peek into the future
The lookahead model
» Optimize over deterministic lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
Lookahead policies peek into the future
The lookahead model
» Optimize over deterministic lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
Lookahead policies peek into the future
The lookahead model
» Optimize over deterministic lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
Lookahead policies peek into the future
The lookahead model
» Optimize over deterministic lookahead model
. . . .
t
t 1
t2
t 3
The base model
Designing a policy
1) Policy function approximations (PFAs)
» Lookup tables, rules, parametric functions
2) Cost function approximation (CFAs)
» X CFA ( St | ) arg min x X
t
C
( St , xt | )
t ( )
3) Policies based on value function approximations (VFAs)
» X tVFA ( St ) arg min x C ( St , xt ) Vt x Stx ( St , xt )
t
4) Lookahead policies
» Deterministic lookahead:
X tLA-D (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
åg
t '-t
C(Stt ' , xtt ' )
t '=t+1
» Stochastic lookahead (e.g. stochastic trees)
X tLA-S (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
p(w ) å g
å
w
ÎWt
t '=t+1
t '-t
C(Stt ' (w ), xtt ' (w ))
Popular with national labs pitching their super computers
Lookahead policies
Probabilistic lookahead
» Here, we approximate the information model by using a
Monte Carlo sample to create a scenario tree:
» We can try to solve this as a single “deterministic”
optimization problem. This is a direct lookahead policy.
Lookahead policies
We can then simulate this lookahead policy over
time:
The lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
We can then simulate this lookahead policy over
time:
The lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
We can then simulate this lookahead policy over
time:
The lookahead model
. . . .
t
t 1
t2
t 3
The base model
Lookahead policies
We can then simulate this lookahead policy over
time:
The lookahead model
. . . .
t
t 1
t2
t 3
The base model
Designing a policy
1) Policy function approximations (PFAs)
» Lookup tables, rules, parametric functions
2) Cost function approximation (CFAs)
» X CFA ( St | ) arg min x X
t
C
( St , xt | )
t ( )
3) Policies based on value function approximations (VFAs)
» X tVFA ( St ) arg min x C ( St , xt ) Vt x Stx ( St , xt )
t
4) Lookahead policies
» Deterministic lookahead:
X tLA-D (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
åg
t '-t
C(Stt ' , xtt ' )
t '=t+1
» Stochastic lookahead (e.g. stochastic trees)
X tLA-S (St ) = arg minC(Stt , xtt ) +
xtt , xt,t+1,..., xt,t+T
T
p(w ) å g
å
w
ÎWt
t '=t+1
t '-t
C(Stt ' (w ), xtt ' (w ))
SMART-ISO: Calibration
Any dynamic model consists of two fundamental
equations:
» The decisions (determined by a policy)
xt X ( St )
» The dynamics (determined by the physics of the problem)
St 1 S M St , xt ,Wt 1
» We have initially focused on replicating the PJM policy
xt X
PJM
( St )
Once we calibrate our model, then we can start looking for
a better policy.
Designing a policy
Some observations
» These four classes of policies formalize what the ISOs
are already doing…
» … but it is likely that their policies will need to be
retuned to handle higher levels of variability.
» Below we report on our progress over the last four
years developing a detailed model of PJM’s energy
markets and power grid.
» We will use this model to tune their policies to handle
much higher levels of wind
Lecture outline
SMART-ISO – Modeling the PJM energy markets
Modeling and designing robust policies
Calibration and LMPs
Modeling offshore wind
Mid-Atlantic Offshore Wind Integration Study
© 2010 Warren B. Powell
Slide 78
SMART-ISO: Calibration
Historical generation mix during 22-28 Jul 2010
Pumped hydro
Comb. cycle+gas
Steam
Nuclear
SMART-ISO: Calibration
Simulated generation mix during 22-28 Jul 2010
Pumped hydro
Comb. cycle+gas
Steam
Nuclear
SMART-ISO: Calibration
Real-time LMPs during 22-28 Jul 2010
Lecture outline
SMART-ISO – Modeling the PJM energy markets
Modeling and designing robust policies
Calibration and LMPs
Modeling offshore wind
Mid-Atlantic Offshore Wind Integration Study
© 2010 Warren B. Powell
Slide 82
SMART-ISO: Offshore wind study
Mid-Atlantic Offshore Wind
Integration and Transmission
Study (U. Delaware & partners,
funded by DOE)
29 offshore sub-blocks in 5
build-out scenarios:
»
»
»
»
»
1: 8 GW
2: 28 GW
3: 40 GW
4: 55 GW
5: 78 GW
GW
Modeling wind
» Steadier than onshore? Where???
Modeling wind
The power from wind:
1
P B Av3
2
v Wind speed (in m/sec)
A Area of rotor blades in m3
Density of air ( 1.225kg/m3 )
B Power coefficient
fraction of wind converted to mechanical energy
.593 (the Betz limit)
» The cubic relationship means small changes in speed
translate to large changes in power.
Onshore & offshore wind farms
We were given access to data on the wind power
generated by onshore wind farms within PJM
Proposal: Use onshore data to calibrate a stochastic model of
forecasting errors. Then use this model to create a simulated
“actual” for offshore.
Simulating onshore wind
Actual (observed) time series (all farms in the Plains):
Forecasted power from wind
Actual
Simulating onshore wind
Histogram of the prediction error (observed/simulated
– forecast) for all farms in the Plains.
Observed
Simulated
Modeling wind
Offshore wind – Buildout level 4
Modeling wind
Offshore wind – Buildout level 4
Modeling wind
Offshore wind – Buildout level 4
Lecture outline
SMART-ISO – Modeling the PJM energy markets
Modeling and designing robust policies
Calibration and LMPs
Modeling offshore wind
Mid-Atlantic Offshore Wind Integration Study
© 2010 Warren B. Powell
Slide 94
SMART-ISO: Offshore wind study
Additional grid capacity needed from ACPF (Jul10, w/ reserves):
The grid limits the use of
offshore wind to around 3
percent. All remaining
analysis is done with an
unconstrained grid.
SMART-ISO: Offshore wind study
Power shortages
140000
Simulated Power - 22-28 Jul 2010 - Buildout 4 - No reserves
120000
80000
60000
40000
Simulated Total Power
Actual Total Demand
20000
0
1
36
71
106
141
176
211
246
281
316
351
386
421
456
491
526
561
596
631
666
701
736
771
806
841
876
911
946
981
1016
1051
1086
1121
1156
1191
1226
1261
1296
1331
1366
1401
1436
1471
1506
1541
1576
1611
1646
1681
1716
1751
1786
1821
1856
1891
1926
1961
1996
MW
100000
5-min Time Intervals
Day Ahead
Actual Hour-ahead
SMART-ISO: Offshore wind study
How do we get rid of
these shortages?
SMART-ISO: Offshore wind study
Ramping Reserves (GW)
10
9
8
*
7
GW
6
5
4
3
2
1
0
Jan-10
Apr-10
Smallest reserve that would
produce a run where all load
is covered.
Jul-10
Oct-10
* - Generators w/ any minimum
operational capacity
SMART-ISO: Offshore wind study
Scheduling up- and down- ramping
Need to schedule down-ramping to handle times
when the wind unexpectedly rises.
Need to schedule up-ramping to handle times when
the wind unexpectedly drops.
SMART-ISO: Offshore wind study
140000
SMART-ISO - Unconstrained Grid - 22-28 Jul 2010
Wind Buildout 4 - No ramping reserves
120000
80000
60000
Actual Demand (Exc)
Simulated (Used) Wind
Simulated Storage Power
Simulated Fast Power
Simulated Slow Power
40000
20000
0
1
36
71
106
141
176
211
246
281
316
351
386
421
456
491
526
561
596
631
666
701
736
771
806
841
876
911
946
981
1016
1051
1086
1121
1156
1191
1226
1261
1296
1331
1366
1401
1436
1471
1506
1541
1576
1611
1646
1681
1716
1751
1786
1821
1856
1891
1926
1961
1996
MW
100000
5-min Time Intervals
SMART-ISO: Offshore wind study
140000
SMART-ISO - Unconstrained Grid - 22-28 Jul 2010
Wind
WindBuildout
Buildout44- -Ramping
No ramping
reserves
reserves
9GW
120000
80000
60000
Actual Demand (Exc)
Simulated (Used) Wind
Simulated Storage Power
Simulated Fast Power
Simulated Slow Power
40000
20000
0
1
36
71
106
141
176
211
246
281
316
351
386
421
456
491
526
561
596
631
666
701
736
771
806
841
876
911
946
981
1016
1051
1086
1121
1156
1191
1226
1261
1296
1331
1366
1401
1436
1471
1506
1541
1576
1611
1646
1681
1716
1751
1786
1821
1856
1891
1926
1961
1996
MW
100000
5-min Time Intervals
SMART-ISO: Offshore wind study
SMART-ISO: Offshore wind study
SMART-ISO: Offshore wind study
SMART-ISO: Offshore wind study
22-28 Jul 2010 - Unconstrained Grid - Buildout 4 - Rsrv 9GW
Avail Wind: 13.7%, Used Wind: 11.1%
60000
50000
Simulated (Used) Wind
Simulated (Avail) Wind
Available wind
30000
Used wind
20000
10000
0
1
36
71
106
141
176
211
246
281
316
351
386
421
456
491
526
561
596
631
666
701
736
771
806
841
876
911
946
981
1016
1051
1086
1121
1156
1191
1226
1261
1296
1331
1366
1401
1436
1471
1506
1541
1576
1611
1646
1681
1716
1751
1786
1821
1856
1891
1926
1961
1996
MW
40000
5-min Time Intervals
SMART-ISO: Offshore wind study
40.0
Available energy from wind
35.0
Percent of total demand
30.0
25.0
20.0
15.0
10.0
5.0
0.0
Jan-10
Apr-10
Jul-10
Oct-10
SMART-ISO: Offshore wind study
90.00
Used wind as percent of available
80.00
No reserves
With reserves
70.00
60.00
50.00
40.00
30.00
20.00
10.00
0.00
Jan-10
Apr-10
Jul-10
Oct-10
SMART-ISO: Offshore wind study
Observations
The integration of renewables will require moving through
a series of plateaus:
» Grid capacity - The current grid does not have the capacity to
handle significant levels of off-shore wind.
» Reserve capacity - Uncertainty in forecasts requires significant
levels of reserves, and increased use of gas turbines.
» Storage – Grid level battery storage can smooth both diurnal cycles
as well as stochastic volatility.
» Demand response – We can reduce the load on the network, but
notification times are important.
» New generation technologies – Fast ramping generators enable
faster response.
» New markets – More responsive markets are needed to take
advantage of fast notification times. It does not help to reduce
ramping times from 12 hours to 8 hours if we still plan generation
a day in advance.