Robust Optimization of the Self- scheduling and Market Involvement - - PowerPoint PPT Presentation

robust optimization of the self scheduling and market
SMART_READER_LITE
LIVE PREVIEW

Robust Optimization of the Self- scheduling and Market Involvement - - PowerPoint PPT Presentation

Ricardo Lima Center for Uncertainty Quan4fica4on in Computa4onal Science & Engineering ricardo.lima@kaust.edu.sa Antonio J. Conejo Integrated Systems


slide-1
SLIDE 1

Robust Optimization of the Self- scheduling and Market Involvement for an Electricity Producer

Ricardo ¡Lima ¡

Center ¡for ¡Uncertainty ¡Quan4fica4on ¡in ¡Computa4onal ¡Science ¡& ¡Engineering ¡ ricardo.lima@kaust.edu.sa ¡

Antonio ¡J. ¡Conejo ¡

Integrated ¡Systems ¡Engineering–Electrical ¡and ¡Computer ¡Engineering, ¡The ¡Ohio ¡State ¡University, ¡OH, ¡USA ¡

Augusto ¡Q. ¡Novais ¡

¡

slide-2
SLIDE 2

Lima, R. M.

SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡ 2 ¡

2006 ¡-­‑ ¡PhD ¡ Department ¡of ¡Chemical ¡Engineering ¡ Faculty ¡of ¡Engineering, ¡University ¡of ¡Porto, ¡ Portugal ¡ An ¡integrated ¡strategy ¡for ¡simula4on ¡and ¡op4miza4on ¡of ¡chemical ¡processes ¡ Salcedo, ¡R. ¡L. ¡ ¡and ¡Barbosa, ¡D. ¡

slide-3
SLIDE 3

Lima, R.M.

SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡ 3 ¡

2006-­‑2008 ¡– ¡Post-­‑doc ¡ Ignacio ¡E. ¡Grossmann ¡ Carnegie ¡Mellon ¡University, ¡PA, ¡USA ¡ Department ¡of ¡Chemical ¡Engineering ¡ ¡ Op4mal ¡synthesis ¡of ¡p-­‑xylene ¡separa4on ¡processes ¡based ¡on ¡crystalliza4on ¡ ¡ Process ¡synthesis, ¡complex ¡MINLP ¡problems ¡

slide-4
SLIDE 4

Lima, R.M.

SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡ 4 ¡

2008 ¡– ¡2011 ¡– ¡Researcher ¡ Ignacio ¡E. ¡Grossmann ¡ PPG ¡Industries ¡ Carnegie ¡Mellon ¡University, ¡PA, ¡USA ¡ Department ¡of ¡Chemical ¡Engineering ¡ ¡ Planning ¡and ¡long-­‑term ¡scheduling ¡of ¡single ¡stage ¡mul4-­‑product ¡con4nuous ¡ lines ¡with ¡a ¡complex ¡recycling ¡structure ¡

slide-5
SLIDE 5

Lima, R. M.

SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡ 5 ¡

2011 ¡– ¡Co-­‑Fund ¡Marie ¡Curie ¡Fellowship ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡Laboratório ¡Nacional ¡de ¡Energia ¡e ¡Geologia, ¡I.P. ¡(LNEG), ¡Lisbon, ¡Portugal ¡ ¡ Project: ¡Planning ¡and ¡Scheduling ¡of ¡Op4mal ¡Mix ¡

  • f ¡Renewable ¡Sources ¡ ¡in ¡Sustainable ¡Power ¡

Systems ¡

  • Op4miza4on ¡models ¡and ¡solu4on ¡methods ¡ ¡
  • Interdisciplinary ¡work ¡

¡

slide-6
SLIDE 6

Lima@KAUST

  • Research ¡Scien4st, ¡joined ¡SRI-­‑UQ@KAUST ¡on ¡October, ¡2014 ¡
  • Working ¡with ¡Omar ¡Knio ¡and ¡Ibrahim ¡Hoteit ¡
  • Op4miza4on ¡under ¡uncertainty ¡
  • Merge ¡Uncertainty ¡Quan4fica4on ¡with ¡Op4miza4on ¡ ¡
  • Focus ¡on ¡high ¡impact ¡applica4ons ¡

SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡ 6 ¡ Decomposi )on ¡ Solver ¡ Model ¡

Uncertainty ¡ SRI-­‑UQ ¡@ ¡KAUST ¡ Applica4on ¡

slide-7
SLIDE 7

Motivation and objectives

  • Research ¡developments ¡and ¡challenges ¡

– Developments ¡in ¡two ¡stage ¡adap4ve ¡Robust ¡Op4miza4on ¡(RO) ¡ ¡

  • Bertsimas ¡and ¡Slim, ¡2003 ¡
  • Bertsimas ¡et ¡al., ¡2011 ¡
  • Bertsimas ¡et ¡al., ¡2013 ¡
  • Thiele ¡et ¡al., ¡2009 ¡
  • Problem ¡features ¡

– Complementarity ¡of ¡energy ¡sources: ¡hydro ¡and ¡wind ¡ – Uncertainty ¡due ¡to ¡renewable ¡energy ¡sources ¡ – Deregula4on ¡of ¡electricity ¡markets ¡

  • Scheduling ¡problems ¡to ¡minimize ¡opera4onal ¡costs ¡
  • Maximize ¡profit ¡by ¡their ¡interac4on ¡with ¡the ¡electricity ¡market ¡

– Electricity ¡pool ¡ – Forward ¡contracts, ¡Swaps ¡ ¡ Develop ¡an ¡op4miza4on ¡framework ¡based ¡on ¡RO ¡to ¡support ¡the ¡ decision ¡making ¡of ¡electricity ¡producers ¡in ¡a ¡market ¡environment. ¡ ¡

7 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-8
SLIDE 8

RO Overview

  • Aims ¡to ¡find ¡a ¡robust ¡solu4on ¡for ¡a ¡problem ¡under ¡uncertainty ¡ ¡

– Where ¡by ¡robust ¡it ¡is ¡meant ¡that ¡such ¡solu4on ¡is ¡the ¡op4mal ¡for ¡ the ¡worst ¡condi4ons ¡within ¡an ¡uncertainty ¡set ¡describing ¡the ¡ uncertainty ¡

  • RO ¡advantages ¡
  • 1. Under ¡specific ¡condi4ons ¡leads ¡to ¡computa4onal ¡tractable ¡

problems ¡

  • 2. Results ¡can ¡be ¡very ¡reliable, ¡since ¡worst ¡case ¡situa4ons ¡are ¡

considered ¡

  • 3. It ¡does ¡not ¡require ¡a ¡distribu4on ¡of ¡probabili4es ¡
  • RO ¡disadvantages ¡
  • 0. ¡
  • 1. Crude ¡representa4on ¡of ¡the ¡uncertainty ¡ ¡
  • 2. Solu4ons ¡can ¡be ¡very ¡conserva4ve ¡

Meaningful ¡uncertainty ¡sets ¡for ¡ RO ¡-­‑> ¡ ¡Big ¡Data ¡available ¡ Control ¡the ¡conserva4sm ¡level ¡

8 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-9
SLIDE 9

Problem definition

  • Mixed ¡power ¡genera4on ¡system ¡opera4ng ¡in ¡an ¡electricity ¡

market ¡

  • Combinatorial ¡scheduling ¡problem ¡
  • Constraints ¡on ¡the ¡technical ¡opera4on ¡of ¡the ¡units ¡

Electricity ¡ ¡ Market ¡

Day-­‑ahead ¡ Contracts ¡

9 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-10
SLIDE 10

Decision framework

Self-­‑scheduling ¡ Fix ¡commitment ¡of ¡the ¡thermal ¡unit ¡ Forward ¡contrac)ng ¡ Sign ¡selling ¡or ¡buying ¡contracts ¡ ¡ 1 ¡week ¡ Thermal ¡plant ¡dispatch ¡ Power ¡output ¡levels ¡subject ¡to ¡commitment ¡ Pump-­‑storage ¡hydro ¡plant ¡dispatch ¡ Power ¡output ¡levels ¡ Power ¡consump4on ¡to ¡pump ¡ Sell ¡and ¡buy ¡electricity ¡in ¡the ¡pool ¡ Resolu4on: ¡1 ¡hour ¡ Fix ¡0-­‑1 ¡variables ¡– ¡on/off ¡status ¡of ¡thermal ¡unit ¡ Decide ¡buy ¡or ¡sell, ¡power ¡and ¡price ¡ 1st ¡stage ¡decisions ¡ 2nd ¡stage ¡decisions ¡

10 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-11
SLIDE 11

Problem statement

  • Given ¡

– Electricity ¡producer ¡with ¡a ¡porkolio ¡of ¡genera4on ¡units ¡

  • Opera4ng ¡constraints ¡of ¡the ¡units ¡
  • The ¡system ¡can ¡be ¡operated ¡as ¡a ¡virtual ¡power ¡system ¡

– The ¡producer ¡can ¡interact ¡with ¡the ¡market ¡

  • Buy ¡or ¡sell ¡through ¡forward ¡contracts ¡and ¡the ¡pool ¡

– The ¡4me ¡horizon ¡of ¡1 ¡week, ¡with ¡the ¡resolu4on ¡of ¡1 ¡hour ¡ – Forward ¡contracts ¡format ¡ – Electricity ¡price ¡forecasts ¡and ¡error ¡limits ¡ – Wind ¡power ¡forecast ¡and ¡error ¡limits ¡

  • Determine ¡

– Power ¡genera4on ¡schedule ¡by ¡unit ¡ – Hourly ¡electricity ¡sold ¡and ¡bought ¡in ¡the ¡pool, ¡and ¡by ¡contracts ¡

  • Maximize ¡

– Opera)onal ¡profit ¡

11 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-12
SLIDE 12

2-stage adaptive RO framework

Determinis4c ¡model ¡ 2-­‑stage ¡adap4ve ¡RO ¡ R(x,z) ¡is ¡the ¡adap)ve ¡recourse ¡profit ¡ Maximized ¡for ¡the ¡worst ¡case ¡ realiza4on ¡of ¡the ¡stochas4c ¡parameters ¡ within ¡W ¡

12 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

Mul4-­‑period ¡MILP ¡problem, ¡ x, ¡y ¡con4nuous ¡variables ¡ z ¡binary ¡variables ¡ Uncertainty ¡on ¡wind ¡power ¡and ¡ electricity ¡prices ¡

slide-13
SLIDE 13

Comparison of 2-Stage Formulations

2-­‑stage ¡adap4ve ¡RO ¡

13 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

Q(x, ξ(w)) = max

y

cy(w)T y s.t. Fy = h(w) − G(w)x By(w) ≤ d(w) − A(w)z Hy(w) ≤ g(w) y(w) ≥ 0 R(x, z) = EξQ(x, ξ(w)) 2-­‑stage ¡Stochas4c ¡Programming ¡

slide-14
SLIDE 14

2-stage adaptive RO framework (cont.)

Recourse ¡problem ¡ Inner ¡of ¡the ¡recourse ¡problem ¡ Assuming ¡strong ¡duality, ¡the ¡dual ¡of ¡IR ¡is ¡given ¡by ¡ Convex ¡LP ¡problem ¡ DIR ¡has ¡a ¡feasible ¡region ¡independent ¡of ¡the ¡first ¡stage ¡decision ¡ ¡variables ¡ Next ¡step: ¡Merge ¡the ¡outer ¡problem ¡of ¡the ¡Recourse ¡with ¡the ¡Dual ¡DIR ¡

14 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-15
SLIDE 15

2-stage adaptive RO framework (cont.)

Reformulated ¡recourse ¡problem ¡

2-­‑stage ¡adap)ve ¡RO ¡

This ¡is ¡a ¡nontrivial ¡op)miza)on ¡problem ¡because ¡of ¡the ¡bi-­‑level ¡structure ¡ Difficult ¡to ¡solve ¡with ¡a ¡standard ¡solver ¡

15 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-16
SLIDE 16

(Dual) Constraint Generation Algorithm

(Thiele ¡et ¡al., ¡2009; ¡Zhang ¡and ¡Guan, ¡2009; ¡Jiang ¡et ¡al. ¡2010; ¡Zugno ¡and ¡Conejo, ¡2013) ¡

16 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-17
SLIDE 17

Primal Constraint Generation Algorithm

Master ¡Problem ¡ Recourse ¡Problem ¡ Introduce ¡a ¡copy ¡of ¡ ¡ the ¡primal ¡variables ¡y ¡

17 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-18
SLIDE 18

Uncertain Polyhedral Sets

  • Uncertainty ¡is ¡described ¡by ¡polyhedral ¡sets: ¡built ¡around ¡a ¡nominal ¡value ¡
  • Forecast ¡value ¡
  • Forecast ¡error ¡-­‑> ¡lower ¡and ¡an ¡upper ¡bound ¡
  • This ¡is ¡an ¡alterna4ve ¡approach ¡to ¡a ¡scenario ¡framework ¡built ¡from ¡a ¡

probability ¡distribu4on ¡

Wind ¡power ¡ Electricity ¡pool ¡prices ¡

50 100 150 200 24 48 72 96 120 144 168 Wind power (MW) Time (h) Forecast Lower and upper bounds 20 40 60 80 100 120 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast Lower and upper bounds

18 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-19
SLIDE 19

Risk management

  • The ¡solu4on ¡is ¡at ¡one ¡of ¡the ¡extreme ¡points ¡of ¡the ¡convex ¡set ¡
  • May ¡lead ¡to ¡over ¡conserva4ve ¡solu4ons ¡
  • In ¡RO ¡risk ¡management ¡is ¡implemented ¡by ¡budget ¡constraints ¡

Γ ¡– ¡Budget ¡parameter ¡ zt

+, ¡zt

  • ­‑ ¡-­‑ ¡0-­‑1 ¡variables ¡

High ¡Γ – ¡high ¡number ¡of ¡periods ¡exhibit ¡devia4ons ¡from ¡ ¡ ¡ Low ¡Γ – ¡low ¡number ¡of ¡periods ¡exhibit ¡devia4ons ¡from ¡ ¡

Risk ¡prone ¡approach ¡ Conserva)ve ¡approach ¡

Wind ¡power ¡uncertainty ¡set ¡ Electricity ¡pool ¡prices ¡uncertainty ¡set ¡

19 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-20
SLIDE 20

Characterization of the subproblem

20 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-21
SLIDE 21

Characterization of the subproblem (cont.)

αt ¡– ¡dual ¡variable ¡ λt ¡– ¡electricity ¡price ¡in ¡the ¡pool ¡

21 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-22
SLIDE 22

Linearization of the subproblem

Lineariza4on ¡of ¡ ¡ Defini4on ¡of ¡ ¡ Subs4tu4on ¡ Lineariza4on ¡ Based ¡on ¡ ¡

22 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-23
SLIDE 23

Remarks

  • 1. Master ¡and ¡Sub-­‑Problem ¡are ¡MILP ¡problems. ¡
  • 2. The ¡Sub-­‑Problem ¡is ¡always ¡bounded ¡for ¡any ¡first ¡stage ¡

decisions ¡(complete ¡recourse) ¡if ¡the ¡op4on ¡to ¡buy ¡energy ¡ from ¡the ¡pool ¡is ¡considered. ¡

  • 3. If ¡the ¡MILP ¡Sub-­‑Problem ¡is ¡not ¡solved ¡to ¡op4mality ¡then ¡

I. The ¡LB ¡is ¡not ¡computed ¡with ¡the ¡best ¡solu4on ¡of ¡the ¡ Sub-­‑Problem ¡found, ¡but ¡with ¡the ¡best ¡MILP ¡LB, ¡ ¡

  • II. The ¡integer ¡solu4on ¡obtained ¡is ¡s4ll ¡a ¡valid ¡bound ¡ ¡

23 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-24
SLIDE 24

Results

  • Computa)onal ¡experiments ¡

– Case ¡1 ¡ ¡

  • 1 ¡thermal ¡unit ¡named ¡G1 ¡
  • 1 ¡pumped-­‑storage ¡hydro ¡unit ¡
  • 1 ¡wind ¡farm ¡

– Case ¡2 ¡

  • 1 ¡thermal ¡unit ¡named ¡G2 ¡
  • 1 ¡pumped-­‑storage ¡hydro ¡unit ¡
  • 1 ¡wind ¡farm ¡

– 2 ¡Algorithms: ¡Dual ¡and ¡Primal ¡ – 3 ¡Instances ¡of ¡electricity ¡prices: ¡EP1, ¡EP2, ¡EP3 ¡ – Risk ¡management: ¡5 ¡values ¡for ¡the ¡budget ¡parameter ¡

  • Models ¡implemented ¡in ¡GAMS, ¡on ¡a ¡computer ¡with ¡an ¡Intel ¡Core ¡i7@3.07GHz ¡CPU, ¡64 ¡bits, ¡and ¡8Gb ¡
  • f ¡RAM. ¡The ¡MILP ¡problems ¡are ¡solved ¡with ¡CPLEX ¡12.5. ¡

¡

24 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-25
SLIDE 25

Case 1 – Computational results

Maximum ¡CPU ¡4me ¡set ¡to ¡1500s ¡and ¡0.1% ¡gap. ¡

25 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-26
SLIDE 26

Case 2 – Computational results

Maximum ¡CPU ¡4me ¡set ¡to ¡1500s ¡and ¡0.1% ¡gap. ¡

26 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-27
SLIDE 27

Convergence profiles: Dual vs Primal

Case ¡2, ¡Γ ¡=150, ¡EP3 ¡

16.90% ¡gap ¡

27 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-28
SLIDE 28

Convergence profiles: Dual vs Primal

Case ¡2, ¡Γ ¡=150, ¡EP1 ¡ Both ¡algorithms ¡do ¡not ¡converge ¡

0.91 ¡% ¡gap ¡ The ¡Primal ¡Constraint ¡Genera4on ¡Algorithm ¡cannot ¡close ¡the ¡gap ¡ MILP ¡Sub-­‑Problem ¡is ¡not ¡solved ¡to ¡op4mality ¡

28 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-29
SLIDE 29

Case 1, EP3 – Scheduling and Market Results

Γ ¡= ¡10, ¡Risk ¡prone ¡approach ¡ Γ ¡= ¡100, ¡Conserva)ve ¡approach ¡

20 40 60 80 100 120 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast Lower and upper bounds 100 200 300 400 500 600 700 800 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power (MW) Time (h) PUMP HYDRO WIND THERMAL 100 200 300 400 500 600 700 800 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power (MW) Time (h) 100 200 300 400 500 600 700 800 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power (MW) Time (h) PUMP HYDRO WIND THERMAL 100 200 300 400 500 600 700 800 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power (MW) Time (h) CONTRACT SELL CONTRACT BUY POOL SELL POOL BUY

A ¡more ¡conserva)ve ¡approach: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

29 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-30
SLIDE 30

Risk management results: budget parameter

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Case ¡1 ¡ ¡ Case ¡2 ¡ ¡

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡total ¡energy ¡

30 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-31
SLIDE 31

Case 2, EP1 – Perfect information for Wind

Γ ¡= ¡10, ¡Risk ¡prone ¡approach ¡ Γ ¡= ¡168, ¡Conserva)ve ¡approach ¡

00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time PUMP HYDRO WIND THERMAL 00 50 100 150 200 250 300 350 400 450 500 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power [MW] Time CONTRACT SELL CONTRACT BUY POOL SELL POOL BUY 00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time PUMP HYDRO WIND THERMAL 00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time

A ¡more ¡conserva)ve ¡approach: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Different ¡profile ¡for ¡ ¡hydro ¡genera4on ¡

31 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-32
SLIDE 32

Risk management results: budget parameter

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Case ¡1 ¡ ¡ Case ¡2 ¡ ¡

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

32 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-33
SLIDE 33

Case 2 - Budget parameter vs contract selection

Γ ¡= ¡10, ¡Risk ¡prone ¡approach ¡ Γ ¡= ¡168, ¡Conserva)ve ¡approach ¡

50 100 150 200 250 300 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power (MW) Time (h) PUMP HYDRO WIND THERMAL CONTRACT SELL 50 100 150 200 250 300 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power (MW) Time (h) PUMP HYDRO WIND THERMAL CONTRACT SELL

33 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-34
SLIDE 34

Conclusions and final remarks

  • Robust ¡op4miza4on ¡framework ¡
  • 1. The ¡Sub-­‑Problem ¡has ¡full ¡recourse ¡as ¡long ¡as ¡the ¡producer ¡

has ¡the ¡op)on ¡to ¡buy ¡electricity, ¡this ¡simplifies ¡the ¡

  • algorithm. ¡
  • 2. The ¡two ¡variants ¡of ¡the ¡constraint ¡genera4on ¡algorithm ¡

have ¡a ¡similar ¡performance ¡with ¡excep4on ¡for ¡some ¡cases ¡ where ¡the ¡Primal ¡version ¡is ¡beUer. ¡

  • 3. Some ¡MILP ¡Sub-­‑Problems ¡are ¡not ¡solved ¡to ¡op)mality ¡

I. The ¡constraint ¡genera4on ¡algorithm ¡does ¡not ¡close ¡the ¡gap ¡

  • II. The ¡convergence ¡profile ¡seems ¡to ¡indicate ¡that ¡it ¡has ¡
  • btained ¡the ¡op4mal ¡solu4on ¡

34 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-35
SLIDE 35

Conclusions and final remarks (cont.)

  • Risk ¡management ¡

1. Uncertainty ¡only ¡in ¡electricity ¡prices ¡ ‒ More ¡conserva4ve ¡approaches ¡lead ¡to ¡lower ¡profits ¡(as ¡expected) ¡ ‒ Selec4on ¡of ¡forward ¡contracts ¡to ¡hedge ¡against ¡the ¡vola4lity ¡of ¡the ¡ pool ¡ ¡ 2. Uncertainty ¡on ¡electricity ¡prices ¡and ¡wind ¡power ¡ ‒ More ¡conserva4ve ¡approaches ¡lead ¡to ¡lower ¡profits ¡(as ¡expected) ¡ ‒ It ¡is ¡difficult ¡to ¡foresee ¡and ¡isolate ¡the ¡rela4on ¡between ¡the ¡ conserva4sm ¡level ¡and ¡the ¡contract ¡selec4on ¡and ¡pool ¡involvement ¡ ¡ Lima, ¡R.M., ¡Novais, ¡A.Q., ¡Conejo, ¡A.J. ¡(2015), ¡Weekly ¡self-­‑scheduling, ¡forward ¡ contrac4ng, ¡and ¡pool ¡involvement ¡for ¡an ¡electricity ¡producer. ¡An ¡adap4ve ¡robust ¡

  • p4miza4on ¡approach, ¡EJOR, ¡240, ¡457-­‑475. ¡

¡

35 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-36
SLIDE 36

Acknowledgements

  • The ¡research ¡leading ¡to ¡these ¡results ¡has ¡received ¡funding ¡

from ¡the ¡European ¡Union ¡Seventh ¡Framework ¡Programme ¡ (FP7/2007-­‑2013) ¡under ¡ ¡grant ¡agreement ¡n. ¡PCOFUND-­‑ GA-­‑2009-­‑246542 ¡and ¡from ¡the ¡Founda4on ¡for ¡Science ¡and ¡ Technology ¡of ¡Portugal ¡under ¡the ¡Welcome ¡II ¡programme. ¡

36 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-37
SLIDE 37

References

Electricity ¡Markets ¡and ¡self-­‑scheduling ¡

  • A. ¡Conejo, ¡M. ¡Carrion, ¡J. ¡Morales, ¡Decision ¡Making ¡Under ¡Uncertainty ¡in ¡Electricity ¡Markets, ¡

Springer ¡US, ¡2010. ¡

  • L. ¡Garces, ¡A. ¡J. ¡Conejo, ¡Weekly ¡self-­‑scheduling, ¡forward ¡contrac4ng, ¡and ¡offering ¡strategy ¡for ¡

a ¡producer, ¡IEEE ¡T. ¡Power ¡Syst. ¡25 ¡(2) ¡(2010) ¡657-­‑666. ¡ Robust ¡op)miza)on ¡

  • A. ¡L. ¡Soyster, ¡Convex ¡programming ¡with ¡set-­‑inclusive ¡constraints ¡and ¡applica4ons ¡to ¡inexact ¡

linear ¡programming, ¡Oper. ¡Res. ¡21 ¡(1973) ¡1154-­‑1157. ¡

  • A. ¡Ben-­‑Tal, ¡A. ¡Nemirovski, ¡Robust ¡convex ¡op4miza4on, ¡Math. ¡Oper. ¡Res. ¡23 ¡(4) ¡(1998) ¡

769-­‑805. ¡

  • D. ¡Bertsimas, ¡M. ¡Sim, ¡Robust ¡discrete ¡op4miza4on ¡and ¡network ¡fows, ¡Math. ¡Program. ¡98 ¡

(1-­‑3) ¡(2003) ¡49-­‑71. ¡

  • D. ¡Bertsimas, ¡D. ¡B. ¡Brown, ¡C. ¡Caramanis, ¡Theory ¡and ¡Applica4ons ¡of ¡Robust ¡Op4miza4on, ¡

SIAM ¡Rev. ¡53 ¡(3) ¡(2011) ¡464-­‑501. ¡

  • A. ¡Thiele, ¡T. ¡Terry, ¡M. ¡Epelman, ¡Robust ¡linear ¡op4miza4on ¡with ¡recourse, ¡Department ¡of ¡

Industrial ¡and ¡Opera4ons ¡Engineering, ¡University ¡of ¡Michigan, ¡TR09-­‑01, ¡March ¡2009 ¡(2009) ¡

37 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-38
SLIDE 38

References (cont.)

Robust ¡op)miza)on ¡applied ¡to ¡energy ¡systems ¡

  • M. ¡Zhang, ¡Y. ¡Guan, ¡Two-­‑stage ¡robust ¡unit ¡commitment ¡problem, ¡hsp://www.op4miza4on-­‑
  • nline.org/DB_FILE/2009/10/2427.pdf, ¡Op4miza4on ¡Online ¡(2009). ¡
  • R. ¡Jiang, ¡M. ¡Zhang, ¡G. ¡Li, ¡Y. ¡Guan, ¡Two-­‑stage ¡robust ¡power ¡grid ¡op4miza4on ¡problem, ¡hsp://

www.op4miza4on-­‑online.org/DB_FILE/2010/10/2769.pdf, ¡Op4miza4on ¡Online ¡(2010). ¡

  • L. ¡Zhao, ¡B. ¡Zeng, ¡Robust ¡unit ¡commitment ¡problem ¡with ¡demand ¡response ¡and ¡wind ¡energy, ¡

hsp://www.op4miza4on-­‑online.org/DB_FILE/2010/11/2784.pdf, ¡Op4miza4on ¡Online ¡(2010). ¡

  • D. ¡Bertsimas, ¡E. ¡Litvinov, ¡X. ¡A. ¡Sun, ¡J. ¡Zhao, ¡T. ¡Zheng, ¡Adap4ve ¡Robust ¡Op4miza4on ¡for ¡the ¡

Security ¡Constrained ¡Unit ¡Commitment ¡Problem, ¡IEEE ¡T. ¡Power ¡Syst. ¡28 ¡(1) ¡(2013) ¡52-­‑63. ¡

  • M. ¡Zugno, ¡A. ¡J. ¡Conejo, ¡A ¡robust ¡op4miza4on ¡approach ¡to ¡energy ¡and ¡reserve ¡dispatch ¡in ¡

electricity ¡markets, ¡Tech. ¡rep., ¡Technical ¡University ¡of ¡Denmark, ¡Department ¡of ¡Applied ¡ Mathema4cs ¡and ¡Computer ¡Science ¡(2013). ¡

  • J. ¡Benders, ¡Par44oning ¡procedures ¡for ¡solving ¡mixed-­‑variables ¡programming ¡problems., ¡
  • Numer. ¡Math. ¡4 ¡(1962) ¡238{252. ¡
  • L. ¡Baringo, ¡A. ¡J. ¡Conejo, ¡Oering ¡Strategy ¡Via ¡Robust ¡Op4miza4on, ¡IEEE ¡T. ¡Power ¡Syst. ¡26 ¡(3) ¡

(2011) ¡1418-­‑1425. ¡ ¡

38 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-39
SLIDE 39

RO Overview

  • Aims ¡to ¡find ¡a ¡robust ¡solu4on ¡for ¡a ¡problem ¡under ¡uncertainty ¡ ¡

– Where ¡by ¡robust ¡it ¡is ¡meant ¡that ¡such ¡solu4on ¡is ¡the ¡op4mal ¡for ¡ the ¡worst ¡condi4ons ¡within ¡an ¡uncertainty ¡set ¡describing ¡the ¡ uncertainty ¡

  • RO ¡advantages ¡
  • 1. Under ¡specific ¡condi4ons ¡leads ¡to ¡computa4onal ¡tractable ¡

problems ¡

  • 2. Results ¡can ¡be ¡very ¡reliable, ¡since ¡worst ¡case ¡situa4ons ¡are ¡

considered ¡

  • 3. It ¡does ¡not ¡require ¡a ¡distribu4on ¡of ¡probabili4es ¡
  • RO ¡disadvantages ¡
  • 0. ¡
  • 1. Crude ¡representa4on ¡of ¡the ¡uncertainty ¡ ¡
  • 2. Solu4ons ¡can ¡be ¡very ¡conserva4ve ¡

Meaningful ¡uncertainty ¡sets ¡for ¡ RO ¡-­‑> ¡ ¡Big ¡Data ¡available ¡ Control ¡the ¡conserva4sm ¡level ¡

39 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-40
SLIDE 40

Models of the generation systems

  • Pump ¡storage ¡hydro ¡plant ¡
  • Wind ¡farm ¡

– Wind ¡power ¡parameterized ¡over ¡uncertain ¡convex ¡sets ¡

Water ¡mass ¡balances ¡ Power ¡genera4on ¡ Power ¡consump4on ¡ Bounds ¡on ¡opera4on ¡ Terminal ¡constraint ¡

40 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-41
SLIDE 41

Models of the generation systems

  • Thermal ¡plant ¡

¡

Bounds ¡on ¡power ¡output ¡ Ramp ¡rates ¡limits ¡ ¡

  • n ¡power ¡output ¡

Minimum ¡up ¡4me ¡ Minimum ¡down ¡4me ¡

P ¡(MW) ¡ t ¡(h) ¡ P ¡(MW) ¡ t ¡(h) ¡ Forbidden ¡

41 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-42
SLIDE 42

Models of the generation systems

  • Thermal ¡plant ¡

¡

Total ¡cost ¡ Startup ¡cost ¡ Logic ¡rela4ons ¡for ¡0-­‑1 ¡Var ¡ Shutdown ¡cost ¡

42 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-43
SLIDE 43

Market involvement

  • Pool ¡

– Buy ¡and ¡sell ¡energy ¡from ¡the ¡pool ¡with ¡a ¡price ¡for ¡each ¡hour ¡ – Price ¡taker ¡– ¡the ¡producer ¡does ¡not ¡influence ¡the ¡day-­‑ahead ¡price ¡

  • Forward ¡contracts ¡

– Hedge ¡against ¡the ¡electricity ¡pool ¡prices ¡vola4lity ¡

10 20 30 40 50 60 70 80 90 100 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast for the pool 10 20 30 40 50 60 70 80 90 100 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast for the pool Contract price 50 52 54 56 58 60 62 64 66 68 70

  • 200
  • 150
  • 100
  • 50

50 100 150 200 Price ($/MWh) Power (MW) Buying block 1 Buying block 2 Buying block 3 Selling block 3 Selling block 2 Selling block 1

Data ¡collected ¡from ¡the ¡Iberian ¡Electricity ¡Market ¡– ¡4th ¡Week ¡02/2012 ¡

43 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-44
SLIDE 44

Remarks

  • 1. The ¡master ¡problem ¡(PF) ¡is ¡built ¡over ¡a ¡relaxa4on ¡of ¡the ¡RO ¡

Problem ¡ ¡

  • 2. Any ¡feasible ¡solu4on ¡from ¡the ¡sub-­‑problem ¡is ¡a ¡valid ¡cut ¡for ¡the ¡

Master, ¡while ¡the ¡op4mal ¡solu4on ¡is ¡the ¡4ghter ¡cut. ¡

  • 3. The ¡uncertainty ¡of ¡the ¡wind ¡power ¡and ¡of ¡the ¡electricity ¡prices ¡is ¡

described ¡by ¡polyhedral ¡sets ¡

  • 4. Master ¡problem ¡should ¡generate ¡first ¡stage ¡decisions ¡that ¡ensure ¡

that ¡the ¡problem ¡(RP1) ¡is ¡bounded. ¡

  • 5. The ¡objec4ve ¡func4on ¡of ¡the ¡sub-­‑problem ¡involves ¡bilinear ¡terms ¡

44 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-45
SLIDE 45

Subproblem

… ¡ Lineariza4on ¡ Budget ¡constraints ¡

45 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-46
SLIDE 46

Risk management results: budget parameter

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Case ¡1 ¡ ¡ Case ¡2 ¡ ¡

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡total ¡energy ¡

46 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-47
SLIDE 47

Computational details

Model ¡name ¡ Iter ¡ Objec4ve ¡ Nodes ¡ Gap(%) ¡ CPU(s) ¡ EQ ¡ VAR ¡ 0-­‑1Var ¡ Model ¡Status ¡ SolveStatus ¡ etSolver ¡ SUB-­‑PROBLEM ¡ 0 ¡ 628,568 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter1 ¡ 782,316 ¡ 0 ¡ 0 ¡ 0.2 ¡ 3,200 ¡ 2,371 ¡ 675 ¡ 1 ¡ 1 ¡ 0.3 ¡ SUB-­‑PROBLEM ¡ iter1 ¡ 54,970 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter2 ¡ 764,002 ¡ 247 ¡ 0 ¡ 1.1 ¡ 5,218 ¡ 3,883 ¡ 843 ¡ 1 ¡ 1 ¡ 1.2 ¡ SUB-­‑PROBLEM ¡ iter2 ¡ 93,302 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter3 ¡ 751,683 ¡ 33 ¡ 0 ¡ 1.4 ¡ 7,236 ¡ 5,395 ¡ 1,011 ¡ 1 ¡ 1 ¡ 1.5 ¡ SUB-­‑PROBLEM ¡ iter3 ¡ 64,876 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter4 ¡ 750,877 ¡ 44 ¡ 0 ¡ 1.9 ¡ 9,254 ¡ 6,907 ¡ 1,179 ¡ 1 ¡ 1 ¡ 2.0 ¡ SUB-­‑PROBLEM ¡ iter4 ¡ 81,251 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter5 ¡ 749,908 ¡ 70 ¡ 0 ¡ 2.7 ¡ 11,272 ¡ 8,419 ¡ 1,347 ¡ 1 ¡ 1 ¡ 2.8 ¡ SUB-­‑PROBLEM ¡ iter5 ¡ 77,079 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡ MASTER ¡ iter6 ¡ 746,487 ¡ 141 ¡ 0 ¡ 4.2 ¡ 13,290 ¡ 9,931 ¡ 1,515 ¡ 1 ¡ 1 ¡ 4.3 ¡ SUB-­‑PROBLEM ¡ iter6 ¡ 71,364 ¡ 0 ¡ 0 ¡ 0.1 ¡ 2,019 ¡ 3,025 ¡ 672 ¡ 1 ¡ 1 ¡ 0.2 ¡

Case ¡1, ¡Γ ¡=168, ¡EP3 ¡ Primal ¡constraint ¡genera)on ¡algorithm ¡

47 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-48
SLIDE 48

Problem data: electricity prices

20 40 60 80 100 120 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast for the pool Lower and upper bounds 20 40 60 80 100 120 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast Lower and upper bounds 20 40 60 80 100 120 24 48 72 96 120 144 168 Electricity price (Eur/MWh) Time (h) Forecast Lower and upper bounds

Data ¡collected ¡from ¡the ¡Iberian ¡Electricity ¡Market ¡ ¡

4th ¡Week ¡02/2012 ¡ 1st ¡Week ¡07/2012 ¡ 3rd ¡Week ¡04/2013 ¡

10 20 30 40 50 60 70 80 90 100

  • 200
  • 150
  • 100
  • 50

50 100 150 200 Price ($/MWh) Power (MW) EP3 EP2 EP1

48 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-49
SLIDE 49

Case 2 - Budget parameter vs contract selection

Γ ¡= ¡10, ¡Risk ¡prone ¡approach ¡ Γ ¡= ¡168, ¡Conserva)ve ¡approach ¡

50 100 150 200 250 300 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power (MW) Time (h) PUMP HYDRO WIND THERMAL CONTRACT SELL

Minimum power generation

50 100 150 200 250 300 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power (MW) Time (h) PUMP HYDRO WIND THERMAL CONTRACT SELL

Need to buy from the pool

50 100 150 200 250 300 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power (MW) Time (h) PUMP HYDRO WIND THERMAL POOL BUY CONTRACT SELL

49 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-50
SLIDE 50

Case 2, EP1 – Perfect information for Wind

Γ ¡= ¡10, ¡Risk ¡prone ¡approach ¡ Γ ¡= ¡168, ¡Conserva)ve ¡approach ¡

00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time PUMP HYDRO WIND THERMAL 00 50 100 150 200 250 300 350 400 450 500 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 Power [MW] Time CONTRACT SELL CONTRACT BUY POOL SELL POOL BUY 00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time PUMP HYDRO WIND THERMAL 00 50 100 150 200 250 300 350 400 450 500 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 103 109 115 121 127 133 139 145 151 157 163 Power [MW] Time

A ¡more ¡conserva)ve ¡approach: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Different ¡profile ¡for ¡ ¡hydro ¡genera4on ¡

50 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-51
SLIDE 51

Risk management results: budget parameter

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

Case ¡1 ¡ ¡ Case ¡2 ¡ ¡

More ¡conserva)ve ¡approaches: ¡ Decreases ¡the ¡power ¡sold ¡in ¡the ¡pool ¡ Increases ¡the ¡power ¡sold ¡by ¡contract ¡

51 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡

slide-52
SLIDE 52

Conclusions and final remarks (cont.)

  • Future ¡work ¡
  • 4. Build ¡a ¡Stochas4c ¡Programming ¡to ¡model ¡the ¡last ¡problem ¡
  • 5. Compare ¡Robust ¡Solu4ons ¡with ¡Expected ¡Solu4ons ¡

I. Value ¡and ¡price ¡of ¡Robust ¡Solu4on ¡

  • II. Value ¡of ¡Stochas4c ¡Solu4on, ¡Value ¡of ¡Perfect ¡Informa4on ¡
  • III. Build ¡robust ¡sets ¡
  • 6. Op4mize ¡daily ¡decisions ¡ayer ¡fixing ¡the ¡1st ¡stage ¡decisions ¡

I. Day-­‑ahead ¡market ¡

  • II. Large ¡producer ¡with ¡the ¡capacity ¡to ¡influence ¡the ¡market ¡price ¡

– Bi-­‑level ¡op4miza4on ¡problem ¡(stochas4c ¡programming ¡

  • r ¡robust ¡op4miza4on) ¡

52 ¡ SRI ¡UQ ¡Annual ¡Mee4ng ¡2015 ¡