Impact of Forecast Errors on Expansion Planning

Impact of Forecast Errors on Expansion Planning of

Power Systems with a Renewables Target
DTU Summer School 2016

Salvador Pineda 1 Juan M. Morales 2 Trine K. Boomsma 1

1 University of Copenhagen, funded by FEMs project (www.futureelmarket.dk)

2 Technical University of Denmark

July, 7, 2016

DTU Summer School 2016 July, 7, 2016 1 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Long-term planning problem

We want to determine the optimal capacity of generating units and

transmission lines to
satisfy future electricity demand
ensure a given share of renewable generation
at the minimum cost.

Generation and transmission expansion planning models can be

classified into
single-year (static) or multi-year (dynamic) models
deterministic or stochastic models

DTU Summer School 2016 July, 7, 2016 2 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Singe-year deterministic G-TEP

p G-T capacities
ΦD
s Operation decisions
s Index of system states
τs Weight of each state
ÿ
τs C D
s ` C ppq
ΦD I
` ˘
Min
p,ΦD
s s CD Operating cost
s.t. ψ ě ψtarget CI Investment cost
f ppq ď 0 ψ Demand covered by renewables
hD ΦD
` ˘
s ; ls “ 0, @s ls Electricity demand
g D p, ΦD
` ˘
s ; ρs ď 0, @s. ρs Capacity factor of renewables
f Expansion constraints
hD Balance equation
gD Operating constraints

DTU Summer School 2016 July, 7, 2016 3 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Time variability in G-TEP

Use of duration curves to characterize the variability throughout the
planning horizon
Duration curves are approximated using blocks
1

0.9

0.8

0.7
Wind power (p.u.)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Hours of the year

DTU Summer School 2016 July, 7, 2016 4 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Uncertainty in G-TEP

G-T planning is made several years ahead and therefore, they face
long-term uncertainty in:
Electricity demand growth
Fuel costs
Renewable electricity generation
This uncertainty is usually modeled using scenarios (w) with different
probabilities (πw ).
Planing decisions are then determined by solving a stochastic G-TEP

DTU Summer School 2016 July, 7, 2016 5 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Singe-year stochastic G-TEP

ÿ
τs π w C D Φ D
sw ` C ppq
I
` ˘
Min
p,ΦD
sw sw
s.t. ψw ě ψtarget , @w
f ppq ď 0
hD ΦD
` ˘
sw ; lsw “ 0, @s, w
g D p, ΦD
` ˘
sw ; ρsw ď 0, @s, w.

w Scenario index
πw Scenario probability
lsw Uncertain electricity demand
ρsw Uncertain renewable capacity factor

DTU Summer School 2016 July, 7, 2016 6 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Long-term uncertainty in G-TEP

0.9

0.8

0.7
Wind power (p.u.)

0.6

0.5

0.4

0.3

0.2

0.1

0
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Hours of the year

DTU Summer School 2016 July, 7, 2016 7 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Time variability, short-term and long-term uncertainty

The previous stochastic G-TEP takes into account the variability of

demand and renewable production throughout the planning horizon
and the long-term uncertainty of demand and renewable production

However, the stochastic G-TEP disregards the short-term uncertainty

of electricity demand and renewable capacity factor. That is, it is
assumed that day-ahead dispatch decisions are made perfectly
knowing the realization of these two parameters.

DTU Summer School 2016 July, 7, 2016 8 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

Short-term uncertainty

Wind power production varies through time

Wind power production is hard to predict in advance

Demand DK2 (12/08/14) Wind DK2 (12/08/14)

2,000 800
Forecast
Wind
1,800
600

1,600

400
1,400

1,200
Forecast
200
Demand
1,000
0 5 10 15 20 25 0 5 10 15 20 25

DTU Summer School 2016 July, 7, 2016 9 / 28

 Impact of Forecast Errors on Expansion Planning Introduction

We were wondering...

How can we account for forecast errors within current generation and
transmission capacity expansion models?

What is the impact of these forecast errors on generation and

transmission capacity expansion planning?

What is the impact of the market design on generation and

transmission capacity expansion planning?

DTU Summer School 2016 July, 7, 2016 10 / 28

 Impact of Forecast Errors on Expansion Planning Modelling

Some assumptions

Single-year model (multi-year model in the paper)

Focus on short-term system operation (no long-term uncertainty)

Unit commitment costs are internalized through balancing offers

No inter-temporal constraints (stationary process)

Energy-only market with marginal pricing (no capacity payments)

Perfect competitive market (central planner approach)

Inelastic demand (high cost for load shedding)

DC representation of the network

DTU Summer School 2016 July, 7, 2016 11 / 28

 Impact of Forecast Errors on Expansion Planning Modelling

Two markets floors

Day-ahead market Balancing market

(24 hours before operation) (30 minutes before operation)

0.9
s1
0.8

s1 r2
Forecast wind power (p.u.)

0.7 s1 r3
s1 r1
0.6
0 0.2 0.4 0.6 0.8 1

0.5
s2 Wind power (p.u.)
s2 r2
0.4

s2 r1 s2 r3
0.3 s3 r1
0 0.2 0.4 0.6 0.8 1
0.2
Wind power (p.u.)
0.1
s3
s3 r2
0
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Hours of the year s3 r3
0 0.2 0.4 0.6 0.8 1
Wind power (p.u.)

DTU Summer School 2016 July, 7, 2016 12 / 28

 Impact of Forecast Errors on Expansion Planning Modelling

Coordination between the two market floors

Inefficient market Efficient market

Day-ahead: min C D pΦD q
Day-ahead + balancing
ΦD
min C D pΦD q ` r πr C B pΦD , ΦB q
ř
Balancing: min C B pΦD , ΦB q

Cheapest day-ahead More expensive day-ahead

Expensive balancing Cheaper balancing
High total cost Minimum total cost
Reserves after energy Simultaneous reserve and energy

DTU Summer School 2016 July, 7, 2016 13 / 28

 Impact of Forecast Errors on Expansion Planning Modelling

We have compared three models

G-T expansion problem without forecast errors

G-T expansion problem with forecast errors under efficient market

G-T expansion problem with forecast errors under inefficient market

DTU Summer School 2016 July, 7, 2016 14 / 28

 Impact of Forecast Errors on Expansion Planning Formulations

G-TEP without forecast errors

ÿ
τs C D ΦD
s ` C ppq
I
` ˘
Min
p,ΦD
s s
s.t. ψ ě ψtarget
f ppq ď 0
h D ΦD
` ˘
s ; ls “ 0, @s
g D p, ΦD
` ˘
s ; ρs ď 0, @s.

The electricity market modeled here can be interpreted as:

A real-time market with completely flexible generating units able to
instantaneously adapt their output to the status of the system
A day-ahead market cleared with perfect forecasts of demand and
renewable generation

DTU Summer School 2016 July, 7, 2016 15 / 28

 Impact of Forecast Errors on Expansion Planning Formulations

G-TEP with forecast errors under efficient market

˜ ¸
ÿ ÿ
C D
ΦD πsr C B
ΦB ` C I ppq
` ˘ ` ˘
Min τs s ` sr
p,ΦD B
s ,Φsr s r
s.t. ψ ě ψtarget
f ppq ď 0
hD ΦD
` ˘
s ; ls “ 0, @s
D D
` ˘
g p, Φs ; ρs ď 0, @s
hB ΦB
` ˘
sr “ 0, @s, @r
B D B
` ˘
g p, Φs , Φsr ; ρs , ∆ρsr ď 0, @s, @r.

r Balancing scenario index πsr Balancing scenario probability

ΦB
sr re-dispatch decisions CB Balancing costs
∆ρsr variation of capacity factor hB , g B Balancing constraints
DTU Summer School 2016 July, 7, 2016 16 / 28
 Impact of Forecast Errors on Expansion Planning Formulations

G-TEP with forecast errors under inefficient market

˜ ¸
ÿ ÿ
C D
ΦD πsr C B
ΦB ` C I ppq
` ˘ ` ˘
Min τs s ` sr
p,ΦD B
s ,Φsr s r
s.t. ψ ě ψtarget
f ppq ď 0
hB ΦB
` ˘
sr “ 0, @s, @r
B D B
` ˘
g p, Φs , Φsr ; ρs , ∆ρsr ď 0, @s, @r
$ ,
C D
` D˘
’Min
’ ΦD
’ Φ s /
/
/
& s .
D
Φs P arg s.t. h Φ ; l “ 0
D
` D
˘ @s.
’
’ s s /
/
g D p, ΦD
’ ` ˘ /
s ; ρs ď 0
% -

Impose cost merit-order at the day-ahead

DTU Summer School 2016 July, 7, 2016 17 / 28

 Impact of Forecast Errors on Expansion Planning Illustrative example

Data of illustrative example

ĝ1 g1 ĝ2

n1
n3 n4
fˆ2 fˆ3 ŵ1
fˆ1
l1 n2

Inflexible and cheap generating unit (g1 )

Inelastic load (l1 )
Expansion projects:
1 wind farm (w
p1 )
2 flexible but more expensive units (p g1 , gp2 ). These units have the same
fuel cost but gp2 provides cheaper downward regulation
3 transmission lines (fp1 , fp2 , fp3 )

DTU Summer School 2016 July, 7, 2016 18 / 28

 Impact of Forecast Errors on Expansion Planning Illustrative example

Data of illustrative example

1
Wind (p.u.)
0.9
Demand (p.u.)
0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Time (h)

20 day-ahead blocks
30 balancing scenarios

DTU Summer School 2016 July, 7, 2016 19 / 28

 Impact of Forecast Errors on Expansion Planning Illustrative example

Results of illustrative example (renewable target 40%)

No forecast errors Efficient market Inefficient market

Cap. ŵ1 392 413 441
Cap. fˆ1 392 369 377
Cap. ĝ1 {fˆ2 0 0 195
Cap. ĝ2 {fˆ3 0 202 0
Inv. cost 20.69 27.15 28.52

Forecast errors involve higher stochastic power capacity

Without forecast errors, no investments in flexible generation
With forecast errors, some investment in flexible generation
Forecast errors increased the investment costs
Investment costs are reduced under efficient market
DTU Summer School 2016 July, 7, 2016 20 / 28
 Impact of Forecast Errors on Expansion Planning Case study

Data of case study

g7 g8 g9

n18 n21 n22

wf4

n17 g6
g10 24-bus system (IEEE)
n23
n16 n19 n20
10 existing conventional units
g5

n15
g4 3 flexible generating units
n14
n13 wf3
Variable demand
n24 n11 n12

4 projects of stochastic units

n3 n9 n10
wf2
3 new flexible generating units
7 new transmission lines
n6
wf1
n4 n5
Renewable target of 20/30/40%
n8
n1 n2 n7

g1 g2 g3

DTU Summer School 2016 July, 7, 2016 21 / 28

 Impact of Forecast Errors on Expansion Planning Case study

Expansion planning (renewable target 20%)

No errors Efficient Inefficient
n6 - - 50
Wind capacity n8 950 1000 900
n13 - - -
n23 400 350 450
n18 - - -
Flexible Generation n21 - 80 -
n22 - - 160
n6 n10 - - -
n8 n9 - - -
n11 n13 175 175 175
Line capacity n11 n14 - - 175
n12 n21 - 350 -
n12 n23 - - -
n14 n16 - - 175
Investment cost 162.6 165.1 172.8
DTU Summer School 2016 July, 7, 2016 22 / 28
 Impact of Forecast Errors on Expansion Planning Case study

Consequences of disregarding forecast errors

Renewables target Renewables target

No forecast errors Forecast errors
One-stage market Eff./Inef. market

Expansion planning (1) Expansion planning (2)

Forecast errors
Eff./Inef. market

Total cost (1) Total cost (2)

Renewables penetration (1) Renewables penetration (2)

DTU Summer School 2016 July, 7, 2016 23 / 28

 Impact of Forecast Errors on Expansion Planning Case study

Consequences of disregarding forecast errors

Market Desired target Cost increase Achieved target

20% 0.1% 19.2%
Efficient 30% 0.5% 26.5%
40% -0.3% 31.5%
20% 8.7% 19.3%
Inefficient 30% 16.2% 28.7%
40% 19.8% 36.1%

Efficient market design

Similar total costs
Renewables penetration below the target
Inefficient market design
Significant increase of total costs
Penetration levels closer to the desired target
DTU Summer School 2016 July, 7, 2016 24 / 28
 Impact of Forecast Errors on Expansion Planning Summary

Summary
We have explained how time variability, long-term and short-term
uncertainty can be modeled in G-TEP

We have presented a set of G-TEP that account for the forecast

errors of stochastic production and two different market designs

These models can be reformulated as single-level mixed-integer linear

programming problems

Considering production forecast errors impacts the generation and

transmission expansion planning of a power system

An efficient market design softens the negative effects of forecast

errors and leads to cheaper expansion plans for a given target

The consequences of an expansion plan determined disregarding

forecast errors highly depend on the market design
DTU Summer School 2016 July, 7, 2016 25 / 28
 Impact of Forecast Errors on Expansion Planning Summary

Future research

Incorporate strategic behaviour of market players

Modify the models to account for flexibility using unit commitment

constraints (ramp rates, minimum times, etc)

Model intermediate market designs between the paradigmatic efficient

and inefficient

Apply dedicated computational methods to improve tractability of the

multi-year case

DTU Summer School 2016 July, 7, 2016 26 / 28

 Impact of Forecast Errors on Expansion Planning Summary

All details can be found in

DTU Summer School 2016 July, 7, 2016 27 / 28

 Impact of Forecast Errors on Expansion Planning Summary

Thanks for the attention!

Questions?

Website: https://sites.google.com/site/slv2pm/

DTU Summer School 2016 July, 7, 2016 28 / 28