zero-carbon analytics · zero-carbon analytics andy philpott electric power optimization centre...

86
Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre www.epoc.org.nz University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward) Supported by Marsden grant UOA1520 INFORMS, Seattle, October 21, 2019 Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 1 / 55

Upload: others

Post on 17-Mar-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Zero-Carbon Analytics

Andy Philpott

Electric Power Optimization Centrewww.epoc.org.nz

University of AucklandNew Zealand

(Joint work with Michael Ferris and Anthony Downward)

Supported by Marsden grant UOA1520

INFORMS, Seattle, October 21, 2019

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 1 / 55

Page 2: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Extinction Rebellion October 2019(http://rebellion.earth/international-rebellion.)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 2 / 55

Page 3: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Wellington Extinction Rebellion protest, October 8(Photo from Getty Images)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 3 / 55

Page 4: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Wellington student protest, September 27, 2019(Photo by Kevin Stent, Stuff.)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 4 / 55

Page 5: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Wellington student protest, September 27, 2019(Photo by Joe Lloyd, Stuff.)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 5 / 55

Page 6: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Jacinda’s 2017 election deal

Introduce a Zero Carbon Act and establish anindependent Climate Commission.Request the Climate Commission to plan the transitionto 100% renewable electricity by 2035 (which includesgeothermal) in a normal hydrological year.

Stimulate up to $1 billion of new investment in lowcarbon industries by 2020, kick-started by aGovernment-backed Green Investment fund of $100M.

(Confidence and Supply Agreement between the New Zealand LabourParty and the Green Party of Aoteoroa)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 6 / 55

Page 7: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Jacinda’s 2017 election deal

Introduce a Zero Carbon Act and establish anindependent Climate Commission.Request the Climate Commission to plan the transitionto 100% renewable electricity by 2035 (which includesgeothermal) in a normal hydrological year.

Stimulate up to $1 billion of new investment in lowcarbon industries by 2020, kick-started by aGovernment-backed Green Investment fund of $100M.

(Confidence and Supply Agreement between the New Zealand LabourParty and the Green Party of Aoteoroa)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 6 / 55

Page 8: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Jacinda’s 2017 election deal

Introduce a Zero Carbon Act and establish anindependent Climate Commission.Request the Climate Commission to plan the transitionto 100% renewable electricity by 2035 (which includesgeothermal) in a normal hydrological year.

Stimulate up to $1 billion of new investment in lowcarbon industries by 2020, kick-started by aGovernment-backed Green Investment fund of $100M.

(Confidence and Supply Agreement between the New Zealand LabourParty and the Green Party of Aoteoroa)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 6 / 55

Page 9: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

This talk is about

OR: helping government policy . . .

I to distinguish between objectives and actions;

I to understand effects of uncertainty;

I to understand effects of incentives.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 7 / 55

Page 10: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

This talk is about

OR: helping government policy . . .

I to distinguish between objectives and actions;

I to understand effects of uncertainty;

I to understand effects of incentives.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 7 / 55

Page 11: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

This talk is about

OR: helping government policy . . .

I to distinguish between objectives and actions;

I to understand effects of uncertainty;

I to understand effects of incentives.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 7 / 55

Page 12: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

This talk is about

OR: helping government policy . . .

I to distinguish between objectives and actions;

I to understand effects of uncertainty;

I to understand effects of incentives.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 7 / 55

Page 13: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Summary

1 Introduction

2 Understanding uncertainty

3 Getting to 100 percent renewable electricity

4 Results

5 Understanding incentives

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 8 / 55

Page 14: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Summary

1 Introduction

2 Understanding uncertainty

3 Getting to 100 percent renewable electricity

4 Results

5 Understanding incentives

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 9 / 55

Page 15: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Renewable electricity is intermittent and random

[Source: http://www.caiso.com]

Duck curve shows increasing need for ramping plant in evening.

Electricity systems also need backup capacity to ensure supplywith random renewable generation (e.g. wind).

Very large literature on these topics (see e.g. ENRE sessions)using stochastic optimization models of various sorts.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 10 / 55

Page 16: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Renewable electricity is intermittent and random

[Source: http://www.caiso.com]

Duck curve shows increasing need for ramping plant in evening.

Electricity systems also need backup capacity to ensure supplywith random renewable generation (e.g. wind).

Very large literature on these topics (see e.g. ENRE sessions)using stochastic optimization models of various sorts.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 10 / 55

Page 17: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Renewable electricity is intermittent and random

[Source: http://www.caiso.com]

Duck curve shows increasing need for ramping plant in evening.

Electricity systems also need backup capacity to ensure supplywith random renewable generation (e.g. wind).

Very large literature on these topics (see e.g. ENRE sessions)using stochastic optimization models of various sorts.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 10 / 55

Page 18: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Hydro reservoirs have uncertain inflows

Ohau A power station in New Zealand’s South Island(Photo by By Ulrich Lange, Bochum, Germany - Own work)

If inflows are too low in winter then there is an energy shortage.

Need sufficient backup energy to ensure supply.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 11 / 55

Page 19: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Hydro reservoirs have uncertain inflows

Ohau A power station in New Zealand’s South Island(Photo by By Ulrich Lange, Bochum, Germany - Own work)

If inflows are too low in winter then there is an energy shortage.

Need sufficient backup energy to ensure supply.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 11 / 55

Page 20: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Hydroelectric reservoir optimization

Stochastic dynamic programming applied to

P: min limT→∞ E [ 1T

∑Tt=1 C (Xt ,Ut)]

s.t. Xt+1 = ft(Xt ,Ut , ωt),Ut ∈ U ,Xt ∈ X .

Popular finite-horizon approach is Stochastic Dual DynamicProgramming (SDDP) [Pereira & Pinto, 1991].

Use infinite horizon SDDP with discounting [Shapiro & Ding,2019], or average cost per year (e.g. formulation P above)[Downward & P., 2019]. We implement P in sddp.jl package[Dowson & Kapelevich, 2016].

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 12 / 55

Page 21: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Hydroelectric reservoir optimization

Stochastic dynamic programming applied to

P: min limT→∞ E [ 1T

∑Tt=1 C (Xt ,Ut)]

s.t. Xt+1 = ft(Xt ,Ut , ωt),Ut ∈ U ,Xt ∈ X .

Popular finite-horizon approach is Stochastic Dual DynamicProgramming (SDDP) [Pereira & Pinto, 1991].

Use infinite horizon SDDP with discounting [Shapiro & Ding,2019], or average cost per year (e.g. formulation P above)[Downward & P., 2019]. We implement P in sddp.jl package[Dowson & Kapelevich, 2016].

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 12 / 55

Page 22: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Hydroelectric reservoir optimization

Stochastic dynamic programming applied to

P: min limT→∞ E [ 1T

∑Tt=1 C (Xt ,Ut)]

s.t. Xt+1 = ft(Xt ,Ut , ωt),Ut ∈ U ,Xt ∈ X .

Popular finite-horizon approach is Stochastic Dual DynamicProgramming (SDDP) [Pereira & Pinto, 1991].

Use infinite horizon SDDP with discounting [Shapiro & Ding,2019], or average cost per year (e.g. formulation P above)[Downward & P., 2019]. We implement P in sddp.jl package[Dowson & Kapelevich, 2016].

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 12 / 55

Page 23: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Jacinda’s 2017 election deal

Introduce a Zero Carbon Act and establish anindependent Climate Commission.Request the Climate Commission to plan the transitionto 100% renewable electricity by 2035 (which includesgeothermal) in a normal hydrological year.

Stimulate up to $1 billion of new investment in lowcarbon industries by 2020, kick-started by aGovernment-backed Green Investment fund of $100M.

(Confidence and Supply Agreement between the New Zealand LabourParty and the Green Party of Aoteoroa)

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 13 / 55

Page 24: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Optimized hydro with 500 MW coal capacity

Source: [Fulton, 2018]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 14 / 55

Page 25: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Optimized hydro with 0 MW coal capacity

Source: [Fulton, 2018]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 15 / 55

Page 26: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Optimized hydro with 0 MW coal capacity

Source: [Fulton, 2018]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 16 / 55

Page 27: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

CO2 emissions comparison

Source: [Fulton, 2018]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 17 / 55

Page 28: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Summary

1 Introduction

2 Understanding uncertainty

3 Getting to 100 percent renewable electricity

4 Results

5 Understanding incentives

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 18 / 55

Page 29: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

100 percent renewable electricity for New Zealand

New Zealand GHG emissions (80 MT CO2-e) by sector in 2017

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 19 / 55

Page 30: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Integrated assessment modelsClimate economy models and energy economy models [Anadon et al, 2017]

DICE [Nordhaus, 1992-2017]

MERGE [Manne et al, 1995]

MARKAL/TIMES [Loulou et al, 2008]

GCAM [Calvin et al, 2011]

PAGE09 [Hope, 2011]

AIM/GCE [Fujimori et al, 2012]

FUND [Anthoff & Tol, 2013]

IMAGE [Stehfest et al, 2014]

REMIND [Luderer, 2015]

WITCH [Emmerling et al, 2016]

EMPIRE [Skar et al, 2016]

MESSAGE [Huppman et al, 2019]

ReEDS [Cohen et al, 2019]

Texas Case Study [Boffino et al, 2019]

Gemstone [Ferris & P., 2019]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 20 / 55

Page 31: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Simplified two-stage stochastic model

Capacity decisions are x at cost K (x).Operating decisions are: generation y at cost C (y)

loadshedding q at cost Vq.Random demand is d(ω).

Minimize annual capital cost plus expected operating cost.

P: min K (x) + Eω[C (y(ω))− V (d(ω)− q(ω))]s.t. y(ω) ≤ x ,

y(ω) + q(ω) ≥ d(ω),x , y(ω), q(ω) ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 21 / 55

Page 32: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Simplified two-stage risk-averse model

Capacity decisions are x at cost K (x).Operating decisions are: generation y at cost C (y)

loadshedding q at cost Vq.Random demand is d(ω), and F a coherent risk measure.

Minimize annual capital cost plus risk-adjusted operating cost.

P: min K (x) + Fω[C (y(ω))− V (d(ω)− q(ω))]s.t. y(ω) ≤ x ,

y(ω) + q(ω) ≥ d(ω),x , y(ω), q(ω) ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 22 / 55

Page 33: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Penalize load shedding at V (value of lost load)

Capacity decisions are x at cost K (x).Operating decisions are: generation y at cost C (y)

loadshedding q at cost Vq.Random demand is d(ω).

Minimize annual capital cost plus expected operating cost.

P: min K (x) + Eω[C (y(ω))− V (d(ω)− q(ω))]s.t. y(ω) ≤ x ,

y(ω) + q(ω) ≥ d(ω),x , y(ω), q(ω) ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 23 / 55

Page 34: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Gemstone: New Zealand capacity expansion model

[described in full in Ferris & P., 2019]b = load block, H(b) = hours in load block b.

Winter load duration curve for New Zealand

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 24 / 55

Page 35: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Load duration curve

[Ferris & P., 2019]b = load block, H(b) = hours in load block b.

Load duration curve as load blocks

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 25 / 55

Page 36: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Gemstone model detail

k = technology

Minimize fixed and expected variable costs.

P: min∑

k(Kkxk + Lkzk) +∑

t Eω[Z (t, ω)]s.t. Z (t, ω) =

∑b∈t H(b) (

∑k Ckyk(t, ω, b) + Vq(t, ω, b)) ,

xk ≤ uk ,zk ≤ xk + Uk ,

yk(t, ω, b) ≤ µk(t, ω, b)zk ,∑b∈t H(b)yk(t, ω, b) ≤ νk(t, ω)

∑b∈t H(b)zk + s(t − 1)− s(t),

q(t, ω, b) ≤ d(t, ω, b),d(t, ω, b) ≤

∑k yk(t, ω, b) + q(t, ω, b),

s(t) ≤ Szk ,x , z , y , q, s ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 26 / 55

Page 37: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Operating costs are random

ω = scenario, t = season, b = load block containing H(b) hours.

Minimize fixed and expected variable costs.

P: min∑

k(Kkxk + Lkzk) +∑

t Eω[Z (t, ω)]s.t. Z (t, ω) =

∑b∈t H(b) (

∑k Ckyk(t, ω, b) + Vq(t, ω, b)),

xk ≤ uk ,zk ≤ xk + Uk ,

yk(t, ω, b) ≤ µk(t, ω, b)zk ,∑b∈t H(b)yk(t, ω, b) ≤ νk(t, ω)

∑b∈t H(b)zk + s(t − 1)− s(t),

q(t, ω, b) ≤ d(t, ω, b),d(t, ω, b) ≤

∑k yk(t, ω, b) + q(t, ω, b),

s(t) ≤ Szk ,x , z , y , q, s ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 27 / 55

Page 38: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Capacity of wind and run-of-river is random in a season

ω = scenario, t = season, b = load block containing H(b) hours.

Minimize fixed and expected variable costs.

P: min∑

k(Kkxk + Lkzk) +∑

t Eω[Z (t, ω)]s.t. Z (t, ω) =

∑b∈t H(b) (

∑k Ckyk(t, ω, b) + Vq(t, ω, b)),

xk ≤ uk ,zk ≤ xk + Uk ,

yk(t, ω, b) ≤ µk(t, ω, b)zk ,∑b∈t H(b)yk(t, ω, b) ≤ νk(t, ω)

∑b∈t H(b)zk + s(t − 1)− s(t),

q(t, ω, b) ≤ d(t, ω, b),d(t, ω, b) ≤

∑k yk(t, ω, b) + q(t, ω, b),

s(t) ≤ Szk ,x , z , y , q, s ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 28 / 55

Page 39: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Modeling stored hydro

s(t) = stored water at end of season t.

Minimize fixed and expected variable costs.

P: min∑

k(Kkxk + Lkzk) +∑

t Eω[Z (t, ω)]s.t. Z (t, ω) =

∑b∈t H(b) (

∑k Ckyk(t, ω, b) + Vq(t, ω, b)) ,

xk ≤ uk ,zk ≤ xk + Uk ,

yk(t, ω, b) ≤ µk(t, ω, b)zk ,∑b∈t H(b)yk(t, ω, b) ≤ νk(t, ω)

∑b∈t H(b)zk + s(t − 1)− s(t),

q(t, ω, b) ≤ d(t, ω, b),d(t, ω, b) ≤

∑k yk(t, ω, b) + q(t, ω, b),

s(t) ≤ Szk ,x , z , y , q, s ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 29 / 55

Page 40: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Energy input from reservoir inflows is random in a season

ω = scenario, t = season, b = load block containing H(b) hours.s(t) = stored water at end of season t.

Minimize fixed and expected variable costs.

P: min∑

k(Kkxk + Lkzk) +∑

t Eω[Z (t, ω)]s.t. Z (t, ω) =

∑b∈t H(b) (

∑k Ckyk(t, ω, b) + Vq(t, ω, b)),

xk ≤ uk ,zk ≤ xk + Uk ,

yk(t, ω, b) ≤ µk(t, ω, b)zk ,∑b∈t H(b)yk(t, ω, b) ≤ νk(t, ω)

∑b∈t H(b)zk+s(t − 1)− s(t),

q(t, ω, b) ≤ d(t, ω, b),d(t, ω, b) ≤

∑k yk(t, ω, b) + q(t, ω, b),

s(t) ≤ Szk ,x , z , y , q, s ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 30 / 55

Page 41: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Accounting for CO2 emissions

Some capacity xk , k ∈ N , is deemed non-renewable.

Generation gk(ω) =∑

t

∑b∈t H(b)yk(t, ω, b), emits βkgk(ω) tonnes.

For a choice of θ ∈ [0, 1], constraint is either:∑k∈N xk ≤ (1− θ)

∑k∈N xk(T )

(reduce non-renewable capacity compared with year T ),

[∑k∈N gk(ω)

]≤ (1− θ)

∑k∈N gk(ωT )

(reduce expected non-renewable generation compared with year T ),

Eω [∑

k βkgk(ω)] ≤ (1− θ)∑

k βkgk(ωT )(reduce expected CO2 emissions compared with year T ).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 31 / 55

Page 42: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Emission KPIs

Some capacity xk , k ∈ N , is non-renewable.

Generation gk(ω) =∑

t

∑b∈t H(b)yk(t, ω, b), emits βkgk(ω) tonnes.

For a choice of θ ∈ [0, 1], constraint is either:∑k βkgk(ω) ≤ (1− θ)

∑k βkgk(ωT )

(constrain CO2 emissions almost surely ),

Pω [∑

k βkgk(ω) ≤ (1− θ)∑

k βkgk(ωT )] ≥ 1− α(constrain CO2 emissions with probability 1− α ),

Fω [∑

k βkgk(ω)] ≤ (1− θ)∑

k βkgk(ωT )(constrain risk-adjusted expected CO2 emissions ).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 32 / 55

Page 43: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 44: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 45: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 46: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 47: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 48: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 49: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Assumptions for New Zealand in 2035

Huntly 500MW coal plant decommissioned. No new coal.

0.5% p.a. demand growth in residental and commercial load.

Additional demand (5.7 TWh) from transport electrification(assumes 0.25% EVs in 2019 becomes 50% EVs in 2035).

Additional demand (4.0 TWh) from process heat electrification(33% switch from coal/gas).

Electricity demand 39.2 TWh in 2017 becomes 53.6 TWh in 2035

16 wind scenarios × 13 hydrology scenarios = 208.

Can build run-of-river hydro, onshore wind, solar PV, somegeothermal, batteries, gas plant (CCGT and OCGT), CCGT withcarbon capture and storage (CCS).

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 33 / 55

Page 50: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Summary

1 Introduction

2 Understanding uncertainty

3 Getting to 100 percent renewable electricity

4 Results

5 Understanding incentives

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 34 / 55

Page 51: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

CO2 emissions as constraint binds

Gemstone: increasing θ on CO2 constraints

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 35 / 55

Page 52: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Cost increase (2017 NZD M p.a.) as CO2 constraint binds

Gemstone: increasing θ on CO2 constraints

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 36 / 55

Page 53: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Effect of risk aversion

Risk aversion uses (1− λ)E[Z ] + λAVaR0.90(Z ), λ = 0, 0.5, 0.8.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 37 / 55

Page 54: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Capacity mix: non-renewable capacity constraint

Gemstone: solutions as non-renewable capacity constrained

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 38 / 55

Page 55: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Capacity mix: non-renewable energy constraint

Gemstone: solutions as non-renewable energy constrained

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 39 / 55

Page 56: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Capacity mix: CO2 emission constraint

Gemstone: solutions as CO2 emissions constrained.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 40 / 55

Page 57: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Capacity mix: increasing carbon tax

Gemstone: solutions as carbon tax increases.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 41 / 55

Page 58: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Takeaways

Electricity systems have uncertainties at different timescales that we can approximate in a single model.Outcomes depend on definition of 100% renewableelectricity. Reduce nonrenewable capacity, expectednonrenewable energy, or expected CO2 emissions?Outcomes depend on CO2 emissions KPI. Constrainemissions on average, almost surely, with highprobability, risk-adjusted ?Getting to 100% is much harder than getting to 99%Risk averse social plan produces different technologymix.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 42 / 55

Page 59: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Takeaways

Electricity systems have uncertainties at different timescales that we can approximate in a single model.Outcomes depend on definition of 100% renewableelectricity. Reduce nonrenewable capacity, expectednonrenewable energy, or expected CO2 emissions?Outcomes depend on CO2 emissions KPI. Constrainemissions on average, almost surely, with highprobability, risk-adjusted ?Getting to 100% is much harder than getting to 99%Risk averse social plan produces different technologymix.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 42 / 55

Page 60: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Takeaways

Electricity systems have uncertainties at different timescales that we can approximate in a single model.Outcomes depend on definition of 100% renewableelectricity. Reduce nonrenewable capacity, expectednonrenewable energy, or expected CO2 emissions?Outcomes depend on CO2 emissions KPI. Constrainemissions on average, almost surely, with highprobability, risk-adjusted ?Getting to 100% is much harder than getting to 99%Risk averse social plan produces different technologymix.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 42 / 55

Page 61: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Takeaways

Electricity systems have uncertainties at different timescales that we can approximate in a single model.Outcomes depend on definition of 100% renewableelectricity. Reduce nonrenewable capacity, expectednonrenewable energy, or expected CO2 emissions?Outcomes depend on CO2 emissions KPI. Constrainemissions on average, almost surely, with highprobability, risk-adjusted ?Getting to 100% is much harder than getting to 99%Risk averse social plan produces different technologymix.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 42 / 55

Page 62: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Takeaways

Electricity systems have uncertainties at different timescales that we can approximate in a single model.Outcomes depend on definition of 100% renewableelectricity. Reduce nonrenewable capacity, expectednonrenewable energy, or expected CO2 emissions?Outcomes depend on CO2 emissions KPI. Constrainemissions on average, almost surely, with highprobability, risk-adjusted ?Getting to 100% is much harder than getting to 99%Risk averse social plan produces different technologymix.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 42 / 55

Page 63: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

What’s missing?

What is the optimal path to destination? Largeuncertainty over long time scales (e.g. technologyadvances, future costs of climate change). Multistageproblem gives options to adapt, delay or abandon.Multi-horizon scenario trees [Skar et al, 2016 ] integratetime scales. See Emerald, a multistage stochasticversion of Gemstone solved in JuDGE.jl.

WB83 - Stochastic and Robust Optimization in Energy

Systems

11:40-12:00 Anthony Downward, Applying Multi-stage

Stochastic Programming to Investment Planning for

Energy Systems.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 43 / 55

Page 64: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

What’s missing?

What is the optimal path to destination? Largeuncertainty over long time scales (e.g. technologyadvances, future costs of climate change). Multistageproblem gives options to adapt, delay or abandon.Multi-horizon scenario trees [Skar et al, 2016 ] integratetime scales. See Emerald, a multistage stochasticversion of Gemstone solved in JuDGE.jl.

WB83 - Stochastic and Robust Optimization in Energy

Systems

11:40-12:00 Anthony Downward, Applying Multi-stage

Stochastic Programming to Investment Planning for

Energy Systems.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 43 / 55

Page 65: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Postscript: what happened?

Climate Change Response (Zero Carbon) Amendment Bill, May 2019.[http://www.legislation.govt.nz]

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 44 / 55

Page 66: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Summary

1 Introduction

2 Understanding uncertainty

3 Getting to 100 percent renewable electricity

4 Results

5 Understanding incentives

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 45 / 55

Page 67: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive models with risk averse agents

Previous models assume a social planner. How shouldthis change in competitive markets?Even with assumed perfect competition and convexity,competitive outcomes may differ from socially optimaloutcomes, e.g. when agents are risk averse.Competitive investment with risk trading when agentshave coherent risk measures [Ehrenmann & Smeers, 2011,Ralph & Smeers, 2015, De Maere dAertrycke et al, 2018, Kok et al,2019 ]

TC83 - Policy-Enabling Models for the Power Sector II

12:05 - 12:25 Golbon Zakeri, Investment Under

Uncertainty and Risk in Competitive Electricity

Markets.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 46 / 55

Page 68: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive models with risk averse agents

Previous models assume a social planner. How shouldthis change in competitive markets?Even with assumed perfect competition and convexity,competitive outcomes may differ from socially optimaloutcomes, e.g. when agents are risk averse.Competitive investment with risk trading when agentshave coherent risk measures [Ehrenmann & Smeers, 2011,Ralph & Smeers, 2015, De Maere dAertrycke et al, 2018, Kok et al,2019 ]

TC83 - Policy-Enabling Models for the Power Sector II

12:05 - 12:25 Golbon Zakeri, Investment Under

Uncertainty and Risk in Competitive Electricity

Markets.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 46 / 55

Page 69: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive models with risk averse agents

Previous models assume a social planner. How shouldthis change in competitive markets?Even with assumed perfect competition and convexity,competitive outcomes may differ from socially optimaloutcomes, e.g. when agents are risk averse.Competitive investment with risk trading when agentshave coherent risk measures [Ehrenmann & Smeers, 2011,Ralph & Smeers, 2015, De Maere dAertrycke et al, 2018, Kok et al,2019 ]

TC83 - Policy-Enabling Models for the Power Sector II

12:05 - 12:25 Golbon Zakeri, Investment Under

Uncertainty and Risk in Competitive Electricity

Markets.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 46 / 55

Page 70: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive Risk Averse Model

Competing suppliers a ∈ A sell electricity at π(ω) and pay forexpected CO2 emissions at σ.

Pa(π,σ): min K a(xa) + Fa[Z a(ω)]

s.t. Z a(ω) =∑

k∈a(Ck − π(ω)yak (ω))

+σ∑

k∈a E[βkyk(ω)]

yak (ω) ≤ xak , k ∈ a,

yak (ω) ≥ 0, k ∈ a.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 47 / 55

Page 71: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive Risk Averse Model

Electricity consumer c pays for electricity at π(ω).blank

Pc : min Fc [Z c(ω)]

s.t. Z c(ω) = −(V − π(ω))(d(ω)− qc(ω))

qc(ω) ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 48 / 55

Page 72: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Competitive Risk Averse MOPEC

Find electricity prices π(ω) and a CO2 price σ and actions (xa, y a),a ∈ A and qc so that

(xa, y a) solves Pa(π, σ);

qc solves Pc(π, σ);

0 ≤∑a∈A

∑k∈a

y ak (ω) + qc(ω)− d(ω) ⊥ π(ω) ≥ 0;

0 ≤ (1− θ)∑a∈A

∑k∈a

βkyak (ωT )− Eω

[∑a∈A

∑k∈a

βkyk(ω)

]⊥ σ ≥ 0.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 49 / 55

Page 73: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Equilibrium CO2 prices with increasing risk aversion

θ = 0.7, F(Z ) = (1− λ)E[Z ] + λAVaR0.90(Z ), λ = 0, . . . , 0.7.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 50 / 55

Page 74: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Trading risk recovers social optimum

Can recover risk averse social optimum if risk marketis complete.An Arrow-Debreu security (purchased in stage 1) forpays the holder $1 in scenario ω in stage 2.A contingent carbon credit (purchased in stage 1)entitles the holder to emit 1 tonne of CO2-e inscenario ω in stage 2.Arrow-Debreu securities complete the financial market.Expectation CO2 constraint can be satisfied in riskedequilibrium if agents trade contingent carbon credits.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 51 / 55

Page 75: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Trading risk recovers social optimum

Can recover risk averse social optimum if risk marketis complete.An Arrow-Debreu security (purchased in stage 1) forpays the holder $1 in scenario ω in stage 2.A contingent carbon credit (purchased in stage 1)entitles the holder to emit 1 tonne of CO2-e inscenario ω in stage 2.Arrow-Debreu securities complete the financial market.Expectation CO2 constraint can be satisfied in riskedequilibrium if agents trade contingent carbon credits.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 51 / 55

Page 76: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Trading risk recovers social optimum

Can recover risk averse social optimum if risk marketis complete.An Arrow-Debreu security (purchased in stage 1) forpays the holder $1 in scenario ω in stage 2.A contingent carbon credit (purchased in stage 1)entitles the holder to emit 1 tonne of CO2-e inscenario ω in stage 2.Arrow-Debreu securities complete the financial market.Expectation CO2 constraint can be satisfied in riskedequilibrium if agents trade contingent carbon credits.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 51 / 55

Page 77: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Trading risk recovers social optimum

Can recover risk averse social optimum if risk marketis complete.An Arrow-Debreu security (purchased in stage 1) forpays the holder $1 in scenario ω in stage 2.A contingent carbon credit (purchased in stage 1)entitles the holder to emit 1 tonne of CO2-e inscenario ω in stage 2.Arrow-Debreu securities complete the financial market.Expectation CO2 constraint can be satisfied in riskedequilibrium if agents trade contingent carbon credits.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 51 / 55

Page 78: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Trading risk recovers social optimum

Can recover risk averse social optimum if risk marketis complete.An Arrow-Debreu security (purchased in stage 1) forpays the holder $1 in scenario ω in stage 2.A contingent carbon credit (purchased in stage 1)entitles the holder to emit 1 tonne of CO2-e inscenario ω in stage 2.Arrow-Debreu securities complete the financial market.Expectation CO2 constraint can be satisfied in riskedequilibrium if agents trade contingent carbon credits.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 51 / 55

Page 79: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Challenges

Solving competitive equilibrium problems withincomplete risk markets is generally difficult.Competitive equilibrium problems with incomplete riskmarkets can have multiple equilibia [Gerard et al, 2018].

Risk aversion complicates competitive capacity choices[Mays et al, 2019].

Net-zero CO2 emissions targets typically excludeconsumers, which favours importers ofcarbon-intensive products [Economist, October 19, 2019].

International cooperation: a role for IFORS?

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 52 / 55

Page 80: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Challenges

Solving competitive equilibrium problems withincomplete risk markets is generally difficult.Competitive equilibrium problems with incomplete riskmarkets can have multiple equilibia [Gerard et al, 2018].

Risk aversion complicates competitive capacity choices[Mays et al, 2019].

Net-zero CO2 emissions targets typically excludeconsumers, which favours importers ofcarbon-intensive products [Economist, October 19, 2019].

International cooperation: a role for IFORS?

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 52 / 55

Page 81: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Challenges

Solving competitive equilibrium problems withincomplete risk markets is generally difficult.Competitive equilibrium problems with incomplete riskmarkets can have multiple equilibia [Gerard et al, 2018].

Risk aversion complicates competitive capacity choices[Mays et al, 2019].

Net-zero CO2 emissions targets typically excludeconsumers, which favours importers ofcarbon-intensive products [Economist, October 19, 2019].

International cooperation: a role for IFORS?

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 52 / 55

Page 82: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Challenges

Solving competitive equilibrium problems withincomplete risk markets is generally difficult.Competitive equilibrium problems with incomplete riskmarkets can have multiple equilibia [Gerard et al, 2018].

Risk aversion complicates competitive capacity choices[Mays et al, 2019].

Net-zero CO2 emissions targets typically excludeconsumers, which favours importers ofcarbon-intensive products [Economist, October 19, 2019].

International cooperation: a role for IFORS?

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 52 / 55

Page 83: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

Challenges

Solving competitive equilibrium problems withincomplete risk markets is generally difficult.Competitive equilibrium problems with incomplete riskmarkets can have multiple equilibia [Gerard et al, 2018].

Risk aversion complicates competitive capacity choices[Mays et al, 2019].

Net-zero CO2 emissions targets typically excludeconsumers, which favours importers ofcarbon-intensive products [Economist, October 19, 2019].

International cooperation: a role for IFORS?

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 52 / 55

Page 84: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

THE END

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 53 / 55

Page 85: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

References

De Maere dAertrycke, G., Ehrenmann, A. and Smeers, Y. Investment with

incomplete markets for risk: The need for long-term contracts. Energy

Policy, 105, pp.571-583, 2017.

Dowson, O. and Kapelevich, L., SDDP. jl: a Julia package for Stochastic

Dual Dynamic Programming. Optimization Online, 2017

Downward, A. and Philpott,A.B., Infinite horizon SDDP. EPOC technical

report, 2019.

Ehrenmann, A. and Smeers, Y., Generation capacity expansion in a risky

environment: a stochastic equilibrium analysis. Operations Research, 59(6),

pp.1332-1346, 2011.

Ferris, M.C. and Philpott, A.B., 100% renewable electricity with storage,

EPOC technical report, 2019.

Fulton B. Security of Supply in the New Zealand Electricity Market,

University of Auckland Engineering Honours Thesis, 2018.

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 54 / 55

Page 86: Zero-Carbon Analytics · Zero-Carbon Analytics Andy Philpott Electric Power Optimization Centre University of Auckland New Zealand (Joint work with Michael Ferris and Anthony Downward)

References

Kok, C., Philpott, A.B. and Zakeri, G., Value of transmission capacity in

electricity markets with risk averse agents, EPOC technical report, 2018.

Gerard, H., Leclere, V. and Philpott, A., On risk averse competitive

equilibrium. Operations Research Letters, 46(1), pp.19-26, 2018.

Mays, J., Morton, D. and O’Neill, R.P., Asymmetric Risk and Fuel

Neutrality in Capacity Markets, 2019.

Pereira, M.V. and Pinto, L.M., Multi-stage stochastic optimization applied

to energy planning. Mathematical programming, 52(1-3), pp.359-375, 1991.

Ralph, D. and Smeers, Y., Risk trading and endogenous probabilities in

investment equilibria. SIAM Journal on Optimization, 25(4), pp.2589-2611,

2015.

Skar, C., Doorman, G., Prez-Valds, G.A. and Tomasgard, A., A multi-horizon

stochastic programming model for the European power system, 2016.

All EPOC technical reports and this talk can be downloaded fromwww.epoc.org.nz

Philpott (www.epoc.org.nz) Zero-Carbon Analytics INFORMS October 21, 2019 55 / 55