Project22:(“Developmentand(Tes6ng(of(New(Tools ... · Project22:(“Developmentand(Tes6ng(of(New(Tools”((DevelopingandTesng ImprovedToolsforPower SystemPlanningandOperaon under*Uncertainty!

Project 22: “Development and Tes6ng of New Tools”

Developing and Tes/ng Improved Tools for Power

System Planning and Opera/on under Uncertainty

Ray Zimmerman, Carlos Murillo-‐Sánchez,

Alberto Lamadrid, Daniel Muñoz-‐Álvarez, Tim Mount, Bob Thomas

CERTS Review, Cornell University

August 4-‐5, 2014

Overview

•  Background – moNvaNon, context, status •  MATPOWER – updates and direcNon •  MOPS – MATPOWER OpNmal Power Scheduler •  Tes/ng MOPS – simulaNon framework, tests •  New expansion planning tool •  AC convergence of MOPS

2

Tools Overview

3

MATPOWER

E4ST* based on DC version

Single Period SuperOPF (1st gen)

AC and DC versions

MOPS** DC version

MulN-‐period SuperOPF (2nd gen)

DC version + AC prototype

extensible OPF architecture high performance solvers

single-period explicit contingencies

stochastic cost endogenous reserves

environmental costs optimal investment

unit commitment

multi-period storage

flexible demand ramping

* E4ST – Engineering, Economic, Environmental Electricity Simulation Tool, formerly SuperOPF Planning Tool. ** MOPS – MATPOWER Optimal Power Scheduler, based on Multi-period SuperOPF with Unit Commitment (3rd generation)

new expansion planning tool AC and DC versions

parallel decomposition integer investment decisions

zonal elastic fuel supply finer time granularity

MOPS-‐AC AC prototype

parallel decomposition

2nd order update info

tesNng and simulaNon frameworks

MOPS** DC version

Background -‐ MoNvaNon •  Tool Development

–  Improve on the so\ware tools in current use for planning and operaNon of electric power systems, especially in light of industry trends:

•  uncertainty (renewables, environmental regulaNon, etc.) •  new technologies (storage, demand side parNcipaNon, microgrids, etc.)

•  Tool TesNng –  Demonstrate and measure the benefits of our approach over current system

operator pracNce. •  Compare stochasNc approach to tradiNonal determinisNc approach using MOPS. •  Compare receding horizon structure to day-‐at-‐a-‐Nme planning using MOPS.

•  Impacts –  Small improvements in efficiency can have significant economic and reliability

impacts. –  Open source tools have far-‐reaching impact on research beyond this program.

4

Background -‐ Context •  Tool Development

–  MOPS •  design does not preclude AC network model •  co-‐opNmizing energy, endogenous reserves for conNngencies and load-‐following •  internalized ramping costs •  scenario tree recombinaNon, preserves mulNstage structure of decisions

–  new expansion planning tool (basis for E4ST v2) •  AC or DC network model •  binary investment/reNrement decisions and costs •  elasNcity of fuel supply •  more Nme granularity

•  Tool TesNng –  performance tested in context of original problem with non-‐anNcipaNvity, not

approximaNons made to make problem tractable (mulN-‐stage decisions vs. two-‐stage)

5

Background -‐ Status •  Tool Development

–  MATPOWER •  released versions 5.0 and 5.1 in past year •  expect version 6.0 with MOPS in a few months •  moving toward public open source project with open development paradigm

–  E4ST •  new web site (Schulze project) to host data, tools, results •  version 1 of core solver so\ware stable

–  public distribuNon requires addiNonal cleanup and documentaNon

–  MOPS •  being acNvely used for tesNng •  integraNon into MATPOWER 6, requires more documentaNon, cleanup of code •  receding horizon requires further modificaNons (parNally completed)

–  next gen planning tool, foundaNon for E4ST v2 •  prototype complete, being used in tesNng •  integrate into MATPOWER or separate E4ST v2 project when ready

6

Background -‐ Status •  Tool Development (conNnued)

–  AC version of MOPS •  current prototype not ready for prime Nme due to convergence challenges •  explore Newton-‐based coordinaNon updates to address convergence problems

•  Tool TesNng –  integrated simulaNon plaeorm for MOPS tesNng, two-‐seflement, receding

horizon, etc. •  original afempt at grand unified simulator bogged down in details •  two seflement simulator running •  receding horizon delayed

–  two stage framework, stochasNc vs. determinisNc •  lots of obstacles in design of 118 bus comparisons, calibraNon of inputs, structure of simulaNon •  comparisons almost complete, expect paper submission in Aug or Sep

–  receding horizon •  required addiNonal changes to MOPS (nearly complete) •  requires more general simulator, pursuing gehng a student to help with this •  addiNonal challenges expected as we move to generaNng wind scenarios on the fly

7

Overview


8

MATPOWER Free, open-‐source power system simulaNon environment with extensible OPF and interfaces to state-‐of-‐the-‐art solvers.

9

hfp://www.pserc.cornell.edu/matpower/ –  used worldwide in teaching, research,

industry –  momentum & impact conNnues to grow

•  1062 cita/ons of 2 main MATPOWER papers*

•  664 cita/ons of MATPOWER so\ware/manual*

–  serves as foundaNon for all tools in this project

R. D. Zimmerman, C. E. Murillo-‐Sánchez, and R. J. Thomas, “MATPOWER Steady-‐State OperaNons, Planning and Analysis Tools for Power Systems Research and EducaNon,” Power Systems, IEEE Transac6ons on, vol. 26, no. 1, pp. 12-‐19, Feb. 2011. * Google Scholar, 8/3/15

Annual MATPOWER Downloads

10

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Annual MATPOWER Downloads by Version

5.1

5.0

4.1

4.0

3.2

3.0

2.0

1.0

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Annual MATPOWER Downloads by Version

5.1

5.0

4.1

4.0

3.2

3.0

2.0

1.0

CumulaNve MATPOWER Downloads

11

0

10000

20000

30000

40000

50000

60000

70000

80000

90000

100000

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Cumula/ve MATPOWER Downloads by Version

5.1

5.0

4.1

4.0

3.2

3.0

2.0

1.0

0

10000

20000

30000

40000

50000

60000

70000

80000

90000

100000

2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Cumula/ve MATPOWER Downloads by Version

5.1

5.0

4.1

4.0

3.2

3.0

2.0

1.0

MATPOWER Releases – v5.0 •  Version 5.0b1 – released July 1, 2014

–  conNnuaNon power flow –  applicaNon of SDP relaxaNons of PF equaNons –  extensible opNons architecture –  tools for detailed reporNng of case data, connecNvity, manipulaNng islands –  PSS/E RAW import capability –  new and updated support for 3rd party solvers

•  Version 5.0 (final) – released Dec 17, 2014 –  enhanced PSS/E RAW import, more robust, support for more versions –  so\ limits on DC OPF branch flows –  more user sefable parameters for default interior point solver –  performance enhancements –  bug fixes

12

•  Version 5.1 (final) – released Mar 20, 2015 –  new license

•  switched to more permissive 3-‐clause BSD license from GPL v3

–  new case files •  four new case files represenNng parts of European high voltage grid •  models ranging from 89 to 9421 buses (largest system distributed with MATPOWER)

–  new documentaNon •  on-‐line funcNon reference at hfp://www.pserc.cornell.edu/matpower/docs/ref/

–  new features •  unified interface for 3rd party mixed integer solvers for MILP/MIQP •  support for PARDISO as linear solver used by interior point algorithm for AC OPF •  new and updated support for 3rd party solvers, including OPTI Toolbox (CLP, GLPK, IPOPT),

IPOPT-‐PARDISO •  network reducNon toolbox (Tylavsky, Zhu)

–  performance enhancements –  other refinements and bug fixes

13

MATPOWER Releases – v5.1

New License BSD or MIT style license •  permissive •  short and to the point •  lets people do what they want, as

long as they –  provide afribuNon –  don’t hold you liable

•  leaves open commercializaNon opNons

•  E.g. FreeBSD, jQuery, Rails

GPL license •  copyle\ •  legally complex •  requires distribuNon of any

modificaNons or derivaNves to be under same terms

•  designed to keep research results from transiNoning to proprietary products

•  E.g. Linux, Git, WordPress

14 Sources: •  Why you should use a BSD style license for your Open Source Project, http://www.freebsd.org/doc/en_US.ISO8859-1/articles/bsdl-gpl/article.html •  http://choosealicense.com/

PARDISO – Large-‐scale AC OPF •  Bofleneck in primal-‐dual interior point solvers (IPOPT, Knitro, MIPS) is the Newton

update step, i.e. solving Ax = b •  PARDISO – hfp://www.pardiso-‐project.org

–  thread-‐safe, high-‐performance, robust, memory efficient so\ware for solving large sparse linear systems of equaNons on shared/distributed-‐memory mulNprocessors

•  Largest AC OPF model solved is 3000-‐bus Polish model with 63-‐conNngencies –  roughly equivalent to 193,000-‐bus AC OPF, plus extras –  size previously limited by RAM (64 GB on 12-‐core machine)

•  When using MIPS, switching from Matlab’s built-‐in x=A\b to PARDISO results in … –  30x speedup on 12-‐core machine –  order of magnitude decrease in RAM requirement

•  IPOPT-‐PARDISO – currently MATPOWER’s fastest AC OPF solver –  solves this case in ~18 minutes on my laptop (2014 MacBook Pro) –  developers looking for large-‐scale systems to test their solver, MATPOWER integraNon –  IPOPT-‐PARDISO distributed on PARDISO site under “MATPOWER libraries” –  MATPOWER is helping to drive advances in high-‐performance linear system solvers

15

STAC

16

STAC

17

MATPOWER Development – v6.0-‐dev •  ZIP load model

–  experimental feature based on contributed code •  performance enhancements

–  large speedups when running many small problems

•  MOPS integraNon (Ray) –  code cleanup

•  GNU Octave compaNbility •  splihng program opNons from input data structures •  adding UC to supporNng code for data input, auto-‐generaNon of sensible default data

–  documentaNon, manuals and in code –  automated tests –  tutorial examples

•  MOPS development – beyond MATPOWER 6 (Carlos) –  probabilisNc iniNal state features needed for receding horizon applicaNon –  AC prototype convergence improvement

18

MATPOWER Project DirecNons •  First steps to address need for a sustainable long-‐term plan. •  Apply for 3 years of funding through NSF SI2 (So\ware Infrastructure for

Sustained InnovaNon) program. –  Goals include to “support the crea/on and maintenance of an innovaNve, integrated, reliable,

sustainable and accessible soQware ecosystem providing new capabiliNes that advance and accelerate scien/fic inquiry and applicaNon at unprecedented complexity and scale.”

–  Division of Electrical, CommunicaNons and Cyber Systems “is parNcularly interested in proposals which provide wider, more flexible access to more advanced general algorithms in the areas of electronic and photonics device simulaNon (accounNng for quantum many body effects), computaNonal intelligence, nonlinear op/miza/on or energy system design.”

•  Move MATPOWER to true public open source project/open dev paradigm –  public code repository, mulNple commifers –  public bug tracking facility, user/developer forums, improved web-‐site –  core project documents defining project, goals, policies, how to contribute –  streamlined “on ramps” for users and developers –  effecNve process in place for incorporaNng contribuNons and feedback,

including from other CERTS projects 19

Overview


20

economicdispatch

add DCnetworkmodel

economicdispatch

DC OPF

add DCnetworkmodel

economicdispatch

add ACnetworkmodel

DC OPF

economicdispatch

AC OPF

add ACnetworkmodel

DC OPF

economicdispatch

add zonal reserves

AC OPF

DC OPF

economicdispatch

EDw/reserves

DC OPFw/reserves

AC OPFw/reserves

add zonal reserves

AC OPF

DC OPF

economicdispatch

add contingencies

EDw/reserves

DC OPFw/reserves

AC OPFw/reserves AC OPF

DC OPF

economicdispatch secure

ED

secureDC OPF

secureAC OPF

add contingencies

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

secureAC OPF

addstochastic

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

secureAC OPF

addstochastic

stochastic secure

ED

stochasticsecureDC OPF

stochasticsecureAC OPF

stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

secureAC OPF

stochastic secure

ED



stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

secureAC OPF

stochastic secure

ED



stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF

economicdispatch

MATPOWER

stochastic secure

ED



stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

secureAC OPF

AC OPF

DC OPF

economicdispatch

1st gen SuperOPF

secureED

secureDC OPF

secureAC OPF

stochastic secure

ED



stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

stochastic secure

ED


stochasticED

stochasticDC OPF

EDw/reserves

DC OPFw/reserves DC OPF

economicdispatch

MOPS

single deterministicsystem state

multiple probabilisticsystem states

secureED

secureDC OPF

secureAC OPF

stochastic secure

ED



stochasticED

stochasticDC OPF

stochasticAC OPF

EDw/reserves

DC OPFw/reserves


DC OPF


ED

secureDC OPF

stochastic secure

ED


stochasticED

stochasticDC OPF

EDw/reserves

DC OPFw/reserves DC OPF

economicdispatch

MOPS

MOPS ConNnuous Single Period Problems

21

Tutorial Example System •  3-‐bus triangle network •  generators

–  2 idenNcal 200 MW gens at bus 1, diff reserve cost –  500 MW gen at bus 2 –  all 3 have idenNcal quadraNc generaNon costs

•  load 450 MW at bus 3, curtailable @ $1000 •  branches

–  300 MW limit, line 1–2 –  240 MW limit, line 1–3 –  300 MW limit, line 2–3

•  adequacy requirement –  reserve requirement

•  150 MW limit

–  conNngencies •  generator 2 at bus 1 •  line 1–3

•  wind –  100 MW unit at bus 2 –  3 samples of normal distribuNon around 50 MW

240 MW

bus 1 bus 2

bus 3

200 MW 200 MW 500 MW

450 MW

300 MW

300 MW

100 MW

gen 1 gen 2 gen 3

22

Single Period ConNnuous Examples

23

Economic Dispatch

Gen1 2 3 4

MW

0

50

100

150

200

250

300Generation & Reserves

Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Flow Constraint Shadow Prices


24

Economic Dispatch w/reserves

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90



25

DC OPF

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90



26

DC OPF w/reserves

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90



27

secure DC OPF

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90



28

stochastic DC OPF

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90



29

secure stochastic DC OPF

Gen1 2 3 4

MW

0

50

100

150

200

250


Bus1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90

100Nodal Prices

Line1 2 3

$/MW

0

10

20

30

40

50

60

70

80

90


continuousdispatch

singleperiod

addintegercommitvars

continuousdispatch

singleperiod

with integercommitment

continuousdispatch

singleperiod

addstartup/shutdowncosts


continuousdispatch

singleperiod

with integercommitment + transition

costs


continuousdispatch

singleperiod

add multiple periods


costs


continuousdispatch

singleperiod

multi-period


costs


continuousdispatch

singleperiod

add min

up/downtimes

multi-period


costs


continuousdispatch

singleperiod

with integer commitment+ transition costs

+ min up/down times

multi-period


costs


continuousdispatch

singleperiod

add ramping


+ min up/down times

multi-period


costs


continuousdispatch

singleperiod

multi-period+ ramping


+ min up/down times

multi-period


costs


continuousdispatch

singleperiod

full UC + dispatchproblem

full look-aheaddispatch problem



+ min up/down times

multi-period


costs


continuousdispatch

singleperiod


addstorage


+ min up/down times

multi-period


costs


continuousdispatch

singleperiod


multi-period+ storage

multi-period+ ramping+ storage


+ min up/down times

multi-period


costs


continuousdispatch

singleperiod

MOPS Mixed Integer and MulN-‐Period Problems

30

Unit Commitment Examples

31

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

Base : No Network

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400


32

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

+ DC Network

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400


33

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

+ Startup/Shutdown Costs

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400


34

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

+ Min Up/Down Time Constraints

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400


35

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

+ Ramping Constraints/Ramp Reserve Costs

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400


36

Period1 2 3 4 5 6 7 8 9 10 11 12

Uni

t Com

mitm

ent

Gen 3 @ Bus 2

Gen 2 @ Bus 1

Gen 1 @ Bus 1

+ Storage

Period1 2 3 4 5 6 7 8 9 10 11 12

Gen

erat

ion,

MW

0

100

200

300

400

Period1 2 3 4 5 6 7 8 9 10 11 12N

odal

Pric

e, $

/MW

h

0

50

100

Net

Loa

d, M

W

0

100

200

300

400

MOPS Status

•  Well on way to release in MATPOWER 6.0 –  Need several more months to finalize code cleanup and documentaNon

•  Beyond first public release –  modificaNons for probabilisNc starNng point, required by receding horizon

–  implement fix for stochasNc cost distorNon related to interacNon between commitment status and conNngency outages

–  conNnued work on convergence of AC prototype

37

Overview


38

TesNng and SimulaNon Framework

39

modeling devices, networks,

storage, uncertainty, markets

problem formulaNon

data

algorithms

solvers

Simula/on Environment defines 6me structure, data/informa6on flow pa\erns between …

▸  mul6ple decision stages (e.g. day-‐ahead UC, 5-‐min dispatch/pricing) ▸  sequen6al solves of a given stage ▸  actual opera6on

TesNng MOPS •  MOPS is a general purpose opNmizaNon tool

–  like an OPF –  can be used in many contexts

•  To test MOPS, we need to define a context –  create a plan given a model of forecasted uncertainty –  update/execute that plan as uncertainty is revealed through Nme –  measure performance (e.g. cost, other metrics) –  Monte Carlo comparisons

•  QuesNons –  How to execute the plan?

•  what is locked in, what’s free to change? –  OperaNng policy vs. market design?

40

TesNng MOPS

Stochas/c Secure UC+OPF

•  mulNple scenarios for

demand and renewable availability

•  explicit conNngencies for security

Determinis/c UC+OPF

•  single scenario with

expected demand and renewable availability

•  zonal reserve requirements for security

41

Two Seflement Framework

•  1st seflement –  solves a mulN-‐period plan resulNng in day-‐ahead commitment decisions and reserve allocaNons

•  2nd seflement –  solves single-‐period problem to determine energy dispatch and conNngency reserve allocaNon subject to

•  UC decisions from 1st seflement •  dispatch from previous period 2nd seflement •  newly revealed uncertainty

–  currently using 2nd seflement to approximate actual operaNon

42

TesNng Structure

•  Given: –  historical temp, wind, demand up to operaNng day (any selected day of interest)

–  ARIMA model of temp, wind, demand that can generate potenNal realizaNons of the operaNng day

•  For each approach: 1.  Solve 1st seflement problem for the day (based on

uncertainty predicted by the ARIMA model). 2.  Select N realizaNons of the day generated by ARIMA

model, for each solve 2nd seflement problems sequenNally for each hour, subject to 1st seflement.

43

118-‐bus Test System

44

DC Network Example number of …

buses 118

convenNonal generators 42

wind farms 12

grid-‐level storage units 0

curtailable loads 99

periods in horizon, |T| 24

scenarios per period, |Jt| 5

conNngencies per scenario, |Ktj| – 1 7

variables in resulNng MIQP 582,990

constraints in resulNng MIQP 1,536,006

45

Typical Wind Trajectories Site 3

Hours0 5 10 15 20 25

Perc

enta

ge o

f Max

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Profiles vs. Trajectories, Site 3

46

47 Period

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Gen 42 @ Bus 116, coalGen 41 @ Bus 113, coalGen 40 @ Bus 111, coalGen 35 @ Bus 100, coalGen 33 @ Bus 91, coalGen 32 @ Bus 90, coalGen 31 @ Bus 89, ngGen 30 @ Bus 89, ngGen 29 @ Bus 87, coalGen 28 @ Bus 80, coalGen 27 @ Bus 77, ngGen 26 @ Bus 76, ngGen 25 @ Bus 73, coalGen 24 @ Bus 72, coalGen 23 @ Bus 69, ngGen 22 @ Bus 69, ngGen 21 @ Bus 66, ngGen 20 @ Bus 66, ngGen 19 @ Bus 65, ngGen 18 @ Bus 65, ngGen 17 @ Bus 62, ngccGen 16 @ Bus 61, ngccGen 15 @ Bus 59, ngccGen 14 @ Bus 54, ngGen 13 @ Bus 49, coalGen 12 @ Bus 49, coalGen 11 @ Bus 46, coalGen 10 @ Bus 42, coalGen 9 @ Bus 40, coalGen 8 @ Bus 27, coalGen 7 @ Bus 26, coalGen 6 @ Bus 25, coalGen 5 @ Bus 24, coalGen 4 @ Bus 12, coalGen 2 @ Bus 8, coalGen 1 @ Bus 6, ng

Unit Commitment - Stochastic

48 Period


Unit Commitment - Deterministic

49 Period


Unit Commitment - Both

StochasticDeterministicBoth

50 Period

0 5 10 15 20 25

MW

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000Committed Capacity vs. Expected Net Demand

StochasticDeterministicExpected Net Demand

51

Stochastic Deterministic

Expe

cted

Cos

ts ($

)

#106

0

0.5

1

1.5

2

2.5

3Expected System Costs

FuelNo LoadUnit CommitmentDA ReserveRT ReserveLoad Not Served

Expected Cost Comparison

52

Stochas/c Determinis/c Difference

fuel $1,386,000 $1,564,000 9%

no load $449,000 $440,000 0%

UC $5,000 $9,000 0%

DA Reserve $17,000 $51,000 2%

RT Reserve $8,000 $45,000 2%

LNS $66,000 $650,000 30%

Total $1,931,000 $2,760,000 43%

53 Standard Deviation ($) #105

0 1 2 3 4 5 6

Aver

age

Cos

t ($)

#106

0

0.5

1

1.5

2

2.5

3

3.5

4Performance of Dispatch Policies

stochasticdeterministic

Two-‐Seflement Challenges •  Began with the idea that 1st seflement contracts for commitment, energy,

reserves and ramping would provide “look-‐ahead view” to constrain single-‐period 2nd seflement problem –  too restricNve –  resulted in shedding load when unused capacity was available

•  unused capacity hadn’t been contracted –  consequence of

•  myopic (single-‐period) second seflement problem •  simplified uncertainty model

•  Moved to having second seflement take only UC from first seflement results –  full range of generator available for dispatch and reserves –  conNnue to guard against conNngencies –  ignores first seflement ranges that guarantee ramping feasibility

54

QuesNon

•  How should we use the informaNon from the results of coarser, longer horizon “look-‐ahead” plan to guide the updaNng of that plan by a subsequent finer grain, but shorter horizon problem with new informaNon? – context of tesNng environment – market design context

55

Overview


56

Comparison to Current E4ST New Formula/on E4ST

AC or DC network model DC network model only

binary variables for investment/reNrement decisions conNnuous variables for investment/reNrement decisions

single Nme-‐linked opNmizaNon for enNre horizon independent sequenNal opNmizaNons

temporally co-‐opNmized investment/reNrement decisions independent sequenNal investment/reNrement decisions

fine Nme-‐granularity on investment decisions, yearly steps coarse Nme-‐granularity on investment decisions, decade steps

technology-‐specific invest-‐to-‐deploy delays uniform invest-‐to-‐deploy delay (granularity of investment cycles)

explicit zonal operaNng reserves availability factors as proxy for operaNng reserves, etc.

linear elasNc zonal fuel supply funcNons, possibly with delay exogenous fuel prices

poten6al to include ramping and UC via typical trajectories* operaNons consists of single independent hours

iteraNve soluNon of model decomposiNon direct soluNon of single large model**

highly parallelizable limited opportuniNes for parallel computaNon

approximate soluNon with small non-‐zero duality gap exact soluNon**

explicit hydro constraints, etc. not yet implemented* total output constraints for hydro, emissions, RPS

* future enhancement ** for each independent investment cycle

57

Integer Deployment States

58

τ

τ+1

3

2

1

τ

3

2

1

τ+1

τ+2 τ+2

t-1 t 0

0

0

Investment not approved U =0

Under construction U =0

Online U =1

Retired U =0

Operation cost

τ = Construction delay

SoluNon

•  Lagrangian relaxaNon-‐based coordinaNon and separaNon of inner minimizaNons

•  Two sets of minimizaNons –  independent (AC) OPF problems

•  one per hour type per interval of planning horizon –  independent dynamic programs

•  one per generator or project

•  Highly separable, highly parallelizable •  PotenNal challenge: LR convergence

59

Preliminary Test

60

Test Setup •  Colombian system + several addiNonal projects

–  86 buses –  187 exisNng generaNon units

•  zonal reserve requirement, 5 zones –  15% of local demand in each reserve zone

•  potenNal new projects –  2 large hydro –  18 combined cycle gas –  5 large coal

•  5 hour types •  24 years •  inelasNc fuel supply •  3% yearly discount rate

61

Online Status of Plants

62 Year

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24new coalnew coalnew coalnew coalnew coalnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew hydronew hydroexisting syncgensgen 160 : hydrogen 155 : hydrogen 154 : nggen 153 : hydrogen 142 : nggen 126 : hydrogen 123 : nggen 119 : nggen 118 : coalgen 116 : nggen 115 : coalgen 108 : nggen 106 : ngccgen 105 : nggen 104 : nggen 103 : nggen 102 : nggen 101 : coalgen 100 : ngccgen 96 : ngccgen 95 : ngccgen 94 : coalgen 85 : coalgen 84 : nggen 83 : nggen 82 : nggen 74 : ngccgen 62 : ngccgen 58 : ngccgen 57 : ngccgen 56 : ngccgen 55 : ngccgen 51 : ngccgen 50 : ngccgen 49 : ngccgen 48 : ngccgen 47 : ngccgen 38 : coalgen 31 : coalgen 30 : coalgen 29 : coalgen 28 : nggen 27 : nggen 18 : ngccgen 17 : ngccgen 16 : nggen 15 : ngccgen 14 : coalgen 13 : ngccgen 12 : ngccgen 11 : ngccgen 10 : ngccgen 6 : ngccmany existing hydro units

Expansion Plan

Dual variables

63

Preliminary Results

•  most dual variables have sefled –  need to explicitly compute duality gap

•  new hydro plants and some combined cycle gas units selected for construcNon

•  all natural gas units reNred immediately –  NG is expensive, but units are required to cover hydro shortages during El Niño events

–  requires explicit hydro constraint and possibly explicit modeling of uncertainty to capture properly

64

Status

•  implementaNon of basic formulaNon (presented last year) completed

•  new elasNc linear fuel supply funcNon –  feature implementaNon completed –  generators belong to (possibly overlapping) “fuel zones” in which they compete for fuel in given Nme horizon/season

•  producNon costs can now include fixed cost proporNonal to installed capacity of project

•  test study for Colombian system is undergoing calibraNon of model for fuel supply funcNon

65

To Do •  Add seasonal output restricNons

–  hydro constraints •  currently requires manually adjusNng hydro generaNon cost

–  emissions caps –  RPS standards

•  Coordinate with current E4ST users to close any other gaps and plan transiNon

•  Future –  Change scenarios from being “typical hours” to being “typical

trajectories” to model ramping requirements –  Explore ways to include transmission expansion and uncertainty.

66

Overview


67

DecomposiNon of AC MOPS

68

single large LP/QP

Convergence of AC MOPS con6nuous case

•  current LR coordinaNon with subgradient-‐based lambda update –  slow and unreliable, especially when some states have negaNve prices

•  improved update strategy to accelerate convergence –  augment Lagrangian with squares of AC flow equaNons

•  introduces 2nd order derivaNves of injecNons –  should improve homing capability of lambda updates

69

Second Order Update Strategy

•  Perform lambda update from sensiNvity analysis of FOC of augmented Lagrangian –  similar to method of mulNpliers –  requires solving huge linear system

•  order-‐of-‐magnitude speed-‐ups in solving Ax = b (PARDISO) •  lambda update based on 2nd order info seems doable scale-‐wise

•  Status –  under implementaNon (currently coding Hessians)

70

Support for Other CERTS Projects

•  Bill Schulze – conNnued E4ST applicaNon/enhancements

•  Ben Hobbs – integraNon with E4ST •  Lindsay Anderson – integraNon of chance constrained approach with SuperOPF/MOPS

•  Zhifang Wang – integraNon of syntheNc power grid modeling capability into MATPOWER

•  HyungSeon Oh – AC OPF enhancements •  Kory Hedman – stochasNc unit commitment

71

Summary of FY15 •  Complete current MOPS tes/ng on two-‐seYlement structure and

submit paper to IEEE TransacNons. •  Complete MOPS integra/on and release MATPOWER 6. •  Submit NSF grant to move MATPOWER project to public open

development paradigm as first step toward sustainable long-‐term plan for MATPOWER.

•  Finish building simulator, compleNng receding horizon comparisons. •  Integrate new generaNon expansion planning tool (E4ST v2) into

MATPOWER suite. •  Explore a Newton-‐based coordinaNon scheme to resolving AC MOPS

convergence issues. •  Support other projects using MATPOWER/SuperOPF tools and

frameworks.

72

QuesNons?

73

Documents

Project22:(“Developmentand(Tes6ng(of(New(Tools ... · Project22:(“Developmentand(Tes6ng(of(New(Tools”((DevelopingandTesng ImprovedToolsfor*Power* SystemPlanningandOperaon under*Uncertainty!

Project22:(“Developmentand(Tes6ng(of(New(Tools ... · Project22:(“Developmentand(Tes6ng(of(New(Tools”((DevelopingandTesng ImprovedToolsforPower SystemPlanningandOperaon under*Uncertainty!