2015 11 03 soutenance these v4 -...

Méthodes pour la planification pluriannuelle des réseaux de distribution. Application à l’analyse technico-économique des solutions d’intégration

des énergies renouvelables intermittentes.

Héloïse DUTRIEUX

PhD defense

03/11/15

• Doctoral advisors• Bruno François (L2EP EC Lille)

• Gauthier Delille (EDF R&D)

• Julien Bect (L2S CentraleSupélec)

• Project ANR APOTEOSE (Analyse Probabiliste et nouveaux Optimums Technico-EcOnomiques des SystèmesElectriques en présence de taux de pénétration élevés d’énergies intermittentes)

2��

1. Scope and motivation

2. Novel framework for the study of RES-integration solutions in multi-year distribution planning

3. Approximation methods for computing the multi-year electrical network state

4. Case studies

5. Conclusion and further work

1. Scope 2. Framework 3. Approximation 4. Case studies 5. Conclusion

3��




4. Case studies



4��

HTB2 HTB3

HTB1

HTA (MV)

BT(LV)

HV/MV substation

MV/LV substation

Transmission network

Distribution network

Scope of the work

(HV)


5��

Historically, electricity was mainly produced by central plants connected to the transmission network.

Consumption

Generation


6��

Consumption

Generation

Since the 2000s, more and more Renewable Energy Sources (RES) have been connected to the distribution networks.


7��

Consumption

Generation

RES

RES

RES

RESRES

In the near future, a massive integration of RES is expected in the distribution networks.


8��

Risk of voltage/current constraints


9��

+ !!!

Network investments to connect 25 GW by 2030: 300 k€/MW of photovoltaic generation100 k€/MW of wind generation

Network reinforcement



10

• Advanced reactive power control of generators• Generation curtailment• Real-time voltage control in the substations• Energy storage• …

��

Alternatives to reinforcement

Pg ↓

Qg ~

P, Q~

Uref ~

?Network reinforcement



11

• Real-time voltage control in the HV/MV substations• Advanced reactive power control of generators

• Generation curtailment• Energy storage• …

��

Alternatives to reinforcement

Pg ↓

Qg ~

P, Q~

Uref ~

?Network reinforcement


Which are the best solutions to reduce costs of integrating RES in the medium/long termat an acceptable level of risk and quality?


12

Current planning Requirements with RES

RES-integrationsolutions

Operatinglimits?

No Yes, in energy and/or duration

Operating costs?

Power lossesEnergy not supplied

+ Non-negligible costs depending on the constraints

Network planning methods

Studied scenarios

« Worst case »Multi-year profiles of

generation/consumption

Constraint indicators

Boolean (yes or no)

Statistical(frequency, severity, duration)

ObjectivePrevent 100 % of the networks constraints

Reach a tradeoff between costs and quality of supply

��


13



Operatinglimits?


Operating costs?




Studied scenarios




Boolean (yes or no)




��


14



Operatinglimits?


Operating costs?




Studied scenarios




Boolean (yes or no)




��


15��

Need to modify the current planning methods to assess the techno-economic impacts of the RES-integration solutions

Extensive literature review on the planning approaches taking into account RES-integration solutions

Main limits of the existing planning approaches:

• Inappropriate time sampling: too small time period and/or time step

• Future events sometimes assumed to be perfectly known

• Interactions between MV and LV networks generally neglected

• Economic analysis sometimes incomplete or missing


16��

Need to modify the current planning methods to assess the techno-economic impacts of the RES-integration solutions

Extensive literature review on the planning approaches taking into account RES-integration solutions

Main limits of the existing planning approaches:

• Inappropriate time sampling: too small time period and/or time step

• Future events sometimes assumed to be perfectly known

• Interactions between MV and LV networks generally neglected

• Economic analysis sometimes incomplete or missing


The limits of the existing planning approaches can distort the assessment of the techno-economic

impacts of the RES-integration solutions.

17

• Methods for medium-term distribution network planning (10 years)

• Network constraints: slow voltage/current variations (10 minutes)

• Quality of supply: voltage limits and weakly-supplied LV consumers

• Network and communication facilities assumed reliable

• Provide a suitable framework for the study of RES-integration solutions in the medium/long run

• Address the main limits of the existing planning approaches:‒ Distribution System Operator (DSO)’s behavior‒ Uncertainty on the arrivals of new RES‒ Interactions between MV and LV networks

��

OBJECTIVES

SCOPE


18




4. Case studies



19

Question

How can we address the main limits of the existing approaches?

��

Initial state Final state

?

Interactions between

MV and LV networks

DSO’s behavior

Uncertainty on the arrivals of

new RES


20��

Initial state Final state

?


MV and LV networks

Define multi-variable planning strategies

including RES-integration solutions

Use stochastic scenarios of RES

arrivals

Use a LV network model

DSO’s behavior


new RES

Question

Proposed approach


21��

Proposed tool

Network + DSO simulator

DSO’s decision analysis

MV and LV network state

Initial network

Variables

Scenario

Costs

Scenario parameters

Fixed planning strategy

ConstraintsActions

Technical outputs

Multi-variable planning strategy

Economic analysis

Multi-year scenario builder

Simulate DSO’s analyses / actions+ Compute the network state

Define the DSO’s decision tree / strategy

Compute the total costs of the strategy over the given scenario

Consider stochastic arrivals of RES + load and generation uncertainties


22��


23��

Scenario builder

Generate multi-year scenarios:

• Features of new RES producers

• 10-minute profiles of generation, consumption and voltage reference

∆Uref

Pc

Pw

Ppv

RE

S p

ower

to b

e ac

com

mod

ated

1 2 3 4 5 6 7 8 9 100

48

12

Scé

nario

1P

[M

W/a

]

0

102030

Pto

t [M

W]

1 2 3 4 5 6 7 8 9 100

48

12

Scé

nario

2P

[M

W/a

]

0

102030

Pto

t [M

W]

1 2 3 4 5 6 7 8 9 10048

12

Scé

nario

3P

[M

W/a

]

0102030

Pto

t [M

W]

1 2 3 4 5 6 7 8 9 100

48

12

Scé

nario

4P

[M

W/a

]

0

102030

Pto

t [M

W]

1 2 3 4 5 6 7 8 9 100

48

12

Scé

nario

5P

[M

W/a

]

Année

0

102030

Pto

t [M

W]

PBT

PHTA

Ptot

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/12-0.02

0

0.02

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

11.2

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1


24��

Planning strategy

Model the Distribution System Operator (DSO)’s behavior.

Planning strategy the sequence of the decision analyses made by the DSO in everyday life,

as well as the associated decisions, when the DSO notices an eventthat may require some adaptations of the existing network.

Examples of event connection of a new user, too many weakly-supplied consumers,

constraint limit violations, expected load growth in the coming years, etc.


25��

Planning strategy

Model the Distribution System Operator (DSO)’s behavior. • Each planning strategy is a sequence of rules.

• Each rule describes the RES-integration solutions to remove constraints.

Example of the current French strategy:If the connection of a new MV producer to a feeder supplying loads may cause overvoltages, then it is necessary to:

R1: decrease the fixed tan(φ) reference of the new producer, with tan(φ) ≥ tan(φ)min.

R2: decrease the fixed tan(φ) reference of the existing MV producers, with tan(φ) ≥ tan(φ)min,DSO.

R3: decrease the fixed voltage reference of the on-load tap changer of the HV/MV transformer UOLTC, with UOLTC ≥ UOLTC,min.

R4: reinforce the MV network.

minmin,min ,)tan(,)tan( OLTCDSO U


26��

Planning strategy

Model the Distribution System Operator (DSO)’s behavior. • Each planning strategy is a sequence of rules.

• Each rule describes the RES-integration solutions to remove constraints.

Example of a new strategy including generation curtailment:If the connection of a new MV producer to a feeder supplying loads may cause overvoltages, then it is necessary to:



R3: decrease the fixed voltage reference of the on-load tap changer of the HV/MV transformer UOLTC, with UOLTC ≥ UOLTC,min.

R5: limit the active power of the MV producers considering a maximal curtailment rate τcurt ≤ τcurt,max.

R4: reinforce the MV network. max,minmin,min ,,)tan(,)tan( curtOLTCDSO U


27��


Simulate the network evolutions on a year-by-year basis:

1. Apply the planning strategy at the beginning of the year to remove limit violations and accommodate all LV and MV producers.

2. Estimate the network state over the whole year on the basis of 10-minute generation/consumption profiles.


28��


Simulate the network evolutions on a year-by-year basis:

1. Apply the planning strategy at the beginning of the year to remove limit violations and accommodate all LV and MV producers.

2. Estimate the network state over the whole year on the basis of 10-minute generation/consumption profiles. a. MV network state: voltages, currents, losses and total apparent power

b. LV network state: extreme voltages and number of weakly-supplied LV consumers

Approximation method

n (fast) approximateload-flows

)( mnX )(~

lnY

n* (slow) exact load-flows

Sampling method

)*(* mnX )*(* lnY

n* << n

PART 3


29��

Economic analysis

Compute the investments and operating costs over the studied period.

Stake-holders

Investment and operating costs

DSO

Upgraded and new MV lines Upgraded HV/MV transformersUpgraded MV/LV transformersMV line losses

MV producers

Upgraded and new MV linesUpgraded HV/MV transformersPower Conversion System (PCS) oversizing PCS maintenance

LV producers

Only if connected to dedicated feeders:Upgraded MV/LV transformers New LV dedicated feeders(New MV lines)

T

kT

kk

kk

k

i

V

i

C

i

INPC

111 111

Gross Present Cost

Net Present Cost

T

kk

kk

k

i

C

i

IGPC

111 11


30��




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis



31

S

Example of application: the current French planning strategy

��

Year 10

Load: +2 %/y

Gen.: 20 MW

Year 0

Load: 11 MVA

Generation: 0


32��


Studied scenario

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/12-0.02

0

0.02

Ure

f [pu

]

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

11.2

Pc [

pu]

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1

Pw [

pu]

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1

Pp

v [pu

]

Features of the new producers10-minute profiles of generation,

consumption and voltage reference


33

S

Upgraded lineUpgraded transformer

��

Year 0

1 2 3 4 5 6 7 8 9 100

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30

New lineNew MV producer



34

S

��

Year 1



Example of application: the current French planning strategy1 2 3 4 5 6 7 8 9 10

0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


35

S

��

Year 2




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


36

S

��

Year 3




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


37

S

��

Year 4




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


38

S

��

Year 5




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


39

S

��

Year 6




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


40

S

��

Year 7




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


41

S

��

Year 8




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


42

S

��

Year 9




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


43

S

��

Year 10




0

4

8

12

Scé

nario

1P

[M

W/a

]

0

10

20

30

Pto

t [M

W]

12 30


44


Technical results

1 2 3 4 5 6 7 8 9 10048

12

P [

MW

/a]

PLV

PMV

1 2 3 4 5 6 7 8 9 100

250

500

750

1000

1250

1500

1750

Year

Cos

ts [

k€/a

]

IMVDSO

IMVMP

ILVDSO

ILVLP

ClossDSO

IPCSMP

CPCSMP

��

Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8 Y9 Y10 Total

Length of upgraded/new

MV lines [km]0 1,2 0 0 1,9 1,2 4,3 15 3,3 2 28,9

Number of upgraded/new

MV/LV transformers9 0 2 2 1 9 4 0 0 1 28

Power losses [MWh] 308 331 353 352 340 352 293 596 528 549 4001

Level of weakly-supplied

LV consumers [%]0,73 0,41 1,05 0,63 0,63 1,81 0,03 0,10 0,19 1,01 –


45

1 2 3 4 5 6 7 8 9 10048

12

P [

MW

/a]

PLV

PMV

1 2 3 4 5 6 7 8 9 100

250

500

750

1000

1250

1500

1750

Year

Cos

ts [

k€/a

]

IMVDSO

IMVMP

ILVDSO

ILVLP

ClossDSO

IPCSMP

CPCSMP

��

MV network investments

Power losses

Power conversion supply of MV

producers

LV network investments


Economic results


46




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis


��


New opportunities to study RES-integration solutions

Compare different strategies


Take into account the scenario uncertainty

0 0.5 1 1.5 2 2.50

0.5

1

1.5

2

Net Present Cost NPC [pu]

Em

piric

al p

roba

bilm

ity d

ensi

ty

MeanMedian10 % quantile90 % quantile

Optimize a strategy






Statistical results

47��

In a nutshell




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis





arrivals



new RESInteractions

between MV and LV networks

DSO’s behavior


48




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis



MV and LV networks

��




arrivals


DSO’s behavior


new RES


In a nutshell

?Computation

time

49




4. Case studies



50��

Problem of the computation time in network planning

Which time step size for studying RES-integration solutions?

The smallest as possible because:

• network constraints are defined over 10 minutes

• RES power can vary in a few seconds/minutes

Time step size considered here: ΔT = 10 minutes

Load-flow process f

)( mnX )( lnY

m = number of input variablesl = number of output variablesn = number of time steps = number of load-flows

n = 52560 load-flows per year

tcomput ≈ 3 minutesper year


51��



Study a strategy over a 10-year scenario ≈ 30 minutes

Study a strategy over 200 10-year scenarios ≈ 4 days

Optimize a multi-variable strategy with 50 candidate points ≈ 6 months

Compare 10 optimized strategies ≈ 5 years

Load-flow process f

)( mnX )( lnY



52��



Study a strategy over a 10-year scenario ≈ 30 minutes

Study a strategy over 200 10-year scenarios ≈ 4 days

Optimize a multi-variable strategy with 50 candidate points ≈ 6 months

Compare 10 optimized strategies ≈ 5 years

Load-flow process f

)( mnX )( lnY



Performing exact load-flows is not suitable.It is necessary to reduce computation time while

providing accurate results.

53��

Options to reduce computation time in network planning

Option 1: increase the time step size.

• Commonly used to study RES-integration solutions with ΔT = 30 min or 1 hour.

• Easy to be implemented.

• High loss of accuracy compared with the time saving.

n exactload-flows

)( mnX )( lnY

n* exactload-flows

)*( mnX )*( lnY

n* factor of n

Subsampling


54��


Option 2: simplify the load-flow equations using hypotheses and/or intrusive approximation techniques.

• Often used to study network stability and the statistical impacts of input variables.

• Efficiency depending on:– the hypotheses,

– the intrusive approximation techniques.

n exactload-flows

)( mnX )( lnY

n simplifiedload-flows

)( mnX )(~

lnY n* exactload-flows

)*( mnX )*( lnY

n* factor of n

Intrusive approximationSubsampling


55


Option 3: build a surrogate model of the load-flow process using non-intrusive approximation techniques.

• Often used in application domains when the observed phenomenon is not explicit,but rarely used in network studies.

• Efficiency depending on:‒ the sampling method used to select the points where the exact model has be evaluated,

‒ the approximation method used to build the surrogate model based on the evaluation points.

��


n exactload-flows

)( mnX )( lnY

n approximateload-flows

)( mnX )(~

lnY

n* exact load-flows

Sampling method

)*(* mnX )*(* lnY

n* << n


)( mnX )(~


)*( mnX )*( lnY

n* factor of n

Non-intrusive approximation

Intrusive approximationSubsampling


56

Subsampling Intrusive approximation


Option 3: build a surrogate model of the load-flow process using non-intrusive approximation techniques.

• Often used in application domains when the observed phenomenon is not explicit,but rarely used in network studies.

• Efficiency depending on:‒ the sampling method used to select the points where the exact model has be evaluated,

‒ the approximation method used to build the surrogate model based on the evaluation points.

��


n exactload-flows

)( mnX )( lnY

n approximateload-flows

)( mnX )(~

lnY

n* exact load-flows

Sampling method

)*(* mnX )*(* lnY

n* << n


)( mnX )(~


)*( mnX )*( lnY

n* factor of n

Non-intrusive approximation


57

General procedure to estimate a scalar variable y = f(x)

��

Select a samplingmethod and an approximation

method

STEP 1

) (~

lnY ) ( mnX

?


58


��

) *(* mnX

) (~

lnY ) ( mnX

Build an experimental

design using the sampling method

STEP 2


method

STEP 1

?

with n* << n


59


��

) *(* mnX

) (~

lnY ) ( mnX



STEP 2


method

STEP 1

?

)*(* lnY STEP 3 Perform n* exact

calculations

f


60


��

) *(* mnX

) (~

lnY ) ( mnX



STEP 2


method

STEP 1

?


calculations

f

Estimate the parameters of the surrogate model usingthe approximation method

STEP 4

*f

) *(* mnX ) *(* lnY


61


��

) *(* mnX

) (~

lnY ) ( mnX



STEP 2


method

STEP 1


calculations

f


) *(* mnX ) *(* lnY

STEP 4

) ( mnX Perform napproximatecalculations

STEP 5

*f

*f


62


��

) *(* mnX )*(* lnY

) ( mnX ) (~

lnY ) ( mnX

STEP 3


Perform napproximatecalculations

STEP 4

STEP 5



STEP 2 Perform n* exact calculations

*f


method

STEP 1

f

*f

) *(* mnX ) *(* lnY

Problem: in our work, y is 2(b+1)-

dimensional where bis the number of

buses in the network.


63

General procedure to estimate a vector variable y = f(x)

��

) *(* mnX

) *(* qnZ

) ( mnX ) (~

lnY ) ( mnX

STEP 3

Order

Compute the parameters

W

q

Do PCA

Perform n approximatecalculations

STEP 5

STEP 6





method

STEP 1Find the principal

componentsWYYZ ).*(* 1

f

2

*

2

* %9,99 YZ

) *(* mnX

STEP 4

)*(* lnY


Perform napproximatecalculations

*f

*f

Principal Component Analysis (PCA) to reduce the

number of variables to be estimated:

q << l


64


��

) *(* mnX

) *(* qnZ

) ( mnX ) (~

lnY ) ( mnX

STEP 3

Order


W

q

Do PCA

Do reverse PCA

STEP 5

STEP 6




*...,*,*,...,1 qk fff

STEP 7


method



f

2

*

2

* %9,99 YZ

TWZYY .~

.~ 1

) *(* mnX

Perform n approximatecalculations

*...,*,*,...,1 qk fff

Reverse PCA to compute the final

variable y based on its first principal

components z.

STEP 4

)*(* lnY

Principal Component Analysis (PCA) to reduce the

number of variables to be estimated:

q << l


) (~

qnZ


65


��

) *(* mnX )*(* lnY

) *(* qnZ

) ( mnX ) (~

lnY ) ( mnX

STEP 3

Order


W

q

Do PCA

Do reverse PCA

STEP 4

STEP 5

STEP 6




*...,*,*,...,1 qk fff

STEP 7


method



f

2

*

2

* %9,99 YZ

TWZYY .~

.~ 1

) *(* mnX

Perform n × q approximatecalculations

*...,*,*,...,1 qk fff

Estimate the parameters of the q surrogate models using

the approximation method

) (~

qnZ


66


��


(Fast) approximateload-flow process)( mnX )(

~lnY

(Slow) exact load-flow process

Samplingmethod

)*(* mnX )*(* lnY

n* << n

• Nearest-Neighbor Interpolation (NNI)• Polynomial Regression (PR)• Kriging (K)

• Pruned Factorial Designs (PFD)• Mean factorial-based designs (MFD)• Latin Hypercube Samples (LHS)


67

The most efficient techniques

��

Results of the comparison of the approximation techniques

Nearest-Neighbor Interpolation + Mean factorial-based designPolynomial Regression + Latin Hypercube SampleKriging + Latin Hypercube Sample

Case 1Case 2Case 3

Approximation error

Computation time


68��



Case 1Case 2Case 3

10-3

10-2

10-1

10010

-2

100

102

Tr

err Q

95(U

) [V

]

Voltage

10-3

10-2

10-1

10010

-2

100

102

Tr

err Q

95(I

) [A

]

Current

10-3

10-2

10-1

100

10-2

100

102

104

Tr

err Q

95(S

0) [k

VA

]

Total apparent power

10-3

10-2

10-1

100

10-2

100

102

Tr

err Q

95(P

loss

) [k

W]

Power losses


69��



Case 1Case 2Case 3

10-3

10-2

10-1

10010

-2

100

102

Tr

err Q

95(U

) [V

]

Voltage

10-3

10-2

10-1

10010

-2

100

102

Tr

err Q

95(I

) [A

]

Current

10-3

10-2

10-1

100

10-2

100

102

104

Tr

err Q

95(S

0) [k

VA

]

Total apparent power

10-3

10-2

10-1

100

10-2

100

102

Tr

err Q

95(P

loss

) [k

W]

Power losses

In this case study, very satisfactory results are obtained by Polynomial Regression and Kriging when combined with Latin Hypercube Samples.


70

Final procedure to estimate the network state over one year

��

Surrogate load-flow process

)( mnX )(~

lnY

+ validation on a 200-point sample

= 400 exact load-flows

+ validation on the same sample

= 200 exact load-flows = 52 560 exact load-flows

1st trial 2nd trial Exact calculations


71��

General performances of the proposed procedure

Proposed procedure To be compared with:

Error

Voltage < 150 V Un = 20 kV

Current < 5 A 185 A < In < 615 A

Power losses < 1 % Eloss ≈ 200-600 MWh

Total apparent power < 200 kVA 20 MVA < Sn < 72 MVA

Time saving 8 to 35! 3 or 6 with time subsampling


72




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis



MV and LV networks

��




arrivals


DSO’s behavior


new RES

In a nutshell


Use approximation methods to obtain fast and accurate load-flows

Computation time

73




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis



MV and LV networks

Computation time

��




arrivals


DSO’s behavior


new RES

In a nutshell


Use approximation methods to get fast and

accurate load-flows

For a given computation time, about 20 times more scenarios can be studied

with a very limited loss of accuracy.

74




4. Case studies



75

Considered studies

1. Impact of the minimal tangent phi of the MV producers

2. Impact of the “Last In First Out” generation curtailment

3. Optimization of the current French planning strategy

��

Year 10

Load: +2 %/y

Gen.: 20-22 MW

Simulations over200 scenarios

Use of an optimization

algorithm

Year 0

Load: 11 MVA

Generation: 0

S


76

Study 1: impact of the minimal tangent phi of the MV producers

Variables: θ = tan(φ)min = tan(φ)min,DSO

Possible impacts

Less network reinforcement

More MV producers connected to feeders supplying loads

More or less power losses

Bigger power conversion chain for all the MV producers

��

Fixed strategy

θ



Current planning strategy

tan(φ)min Which impact on the Net Present Cost?

T

kT

kk

kk

k

i

V

i

C

i

INPC

111 111


77

-0.6 -0.5 -0.4 -0.3 -0.25 -0.2 -0.1 00

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

tan(phi)minR

[pu

, ba

se N

PC

m(-

0.25

)]

RRm

Rq25

Rq75

Rm*

Rq25*

Rq75*


��

-0.6 -0.5 -0.4 -0.3 -0.25 -0.2 -0.1 0

0.5

1

1.5

2

2.5

tan(phi)min

NP

C [

pu,

base

NP

Cm

(-0.

25)]

NPCNPCm

NPCq25

NPCq75

NPCm*

),(min),(),( SNPCSNPCSR Θ

High dispersion of the scenario costscompared with the average cost: impossible

to say which value of tan(φ)min is optimal.

The values of tan(φ)min between –0.2 and 0 have a similar average impact on the costs.

),( SNPC


78

-0.6 -0.5 -0.4 -0.3 -0.25 -0.2 -0.1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

tan(phi)min

NP

Cm

[pu

]

ICDSO

ICMP

ILP


��

Average allocation of the costs

-0.6 -0.5 -0.4 -0.3 -0.25 -0.2 -0.1 00

0.2

0.4

0.6

0.8

1

1.2

1.4

tan(phi)min

NP

Cm

[pu

]

IMVDSO

IMVMP

ILVDSO

ILVLP

ClossDSO

IPCSMP

CPCSMP

-0.6 -0.5 -0.4 -0.3 -0.25 -0.2 -0.1 0 0

0.1

0.2

0.3

0.4

0.5

tan(phi)min

NP

Cm

[pu

]

IMV

ILV

Closs

ICPCS

Average cost per category Average cost per stakeholder

Current value


79

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

curt,max

[%]

NP

Cm

[pu

]

ICDSO

ICMP

IPB

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

curt,max

[%]

NP

Cm

[pu

]

IMV

ILV

Closs

ICPCS

Ccurt

0 1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

1.2

curt,max

[%]

NP

Cm

[pu

]

IMVDSO

IMVMP

ILVDSO

ILVLP

ClossDSO

IPCSMP

CPCSMP

CcurtMP

CcurtDSO

Study 2: impact of the “Last In First Out” generation curtailment

��



Current value


80

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

curt,max

[%]

NP

Cm

[pu

]

ICDSO

ICMP

IPB

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

curt,max

[%]

NP

Cm

[pu

]

IMV

ILV

Closs

ICPCS

Ccurt

0 1 2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

1.2

curt,max

[%]

NP

Cm

[pu

]

IMVDSO

IMVMP

ILVDSO

ILVLP

ClossDSO

IPCSMP

CPCSMP

CcurtMP

CcurtDSO

Study 2: impact of the “Last In First Out” generation curtailment

��



Current value

The proposed planning approach makes it possible to assess the techno-economic impacts of the

RES-integration solutions.


81��

Study 3: optimization of the current planning strategy

Why should we use an optimization process?

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.5

1

1.5

2

2.5

tan(phi)min

NP

C [pu

, ba

se N

PC

m(-

0.25

)]

NPCNPC

m

NPCm*

N = 200 scenarios per candidate point

tcomput ≈ 4 hours per candidate point

Accuracy on the average cost: δ = 2 %

Considering 10 candidate values for each decision variable:

• Optimize a 1-variable strategy ≈ 40 hours

• Optimize a 2-variable strategy ≈ 16 days

• Optimize a 3-variable strategy ≈ 5 months


82��


General problem

Studied problem

Objective f• Economic: Net Present Cost, Regret...

• Quality: Number of weakly-supplied consumers...

• Scenario uncertainty: Mean, Quantile…

Constraints g• Finite space of the decision variables,

• Quality on the network…

Decision variables θ• Number of variables

• Continuous or discrete

)(min f

0)(s.t. g


83��


General problem

Studied problem

Objective f• Economic: Net Present Cost, Regret...

• Quality: Number of weakly-supplied consumers...

• Scenario uncertainty: Mean, Quantile…

Constraints g• Finite space of the decision variables,

• Quality on the network…

Decision variables θ• Number of variables: 1

• Continuous or discrete

)(min f

0)(s.t. g

0;6.0)tan( min

),()( SNPCEf S S


84��


Studied problem

Problem particularities?

• No explicit formulation of f

• Noisy evaluation results

• Expensive-to-evaluate model

Optimization algorithm used:

Informational Approach to Global Optimization (IAGO)


),()( SNPCEf S S

min)tan( )(min fΘ

0;6.0Θ

85

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*70

= 0.96 pu

*70

= -0.40

f ( )

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*70

= 0.96 pu

*70

= -0.40

f ( )


After 70 evaluations:

��

Mean of the evaluationsMean of the kriging model95% CI of the kriging model


Estimation of the objective function Allocation of the evaluations

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n ob

s

10 evaluations per point

86

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n obs

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*200

= 0.98 pu

*200

= -0.42

f ( )

��






87

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n ob

s

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*500

= 0.93 pu

*500

= -0.28

f ( )

��






88

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n obs

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*500

= 0.93 pu

*500

= -0.28

f ( )

��


After 1 000 evaluations:

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*1000

= 0.93 pu

*1000

= -0.27

f ( )




89

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n ob

s

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*1000

= 0.93 pu

*1000

= -0.27

f ( )

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*5000

= 0.97 pu

*5000

= -0.25

f ( )

��






83 scenarios per candidate point without optimization

90

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00

200

400

600

800

1000

1200

n obs

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*1000

= 0.93 pu

*1000

= -0.27

f ( )

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*5000

= 0.97 pu

*5000

= -0.25

f ( )

��






83 scenarios per candidate point without optimization

Algorithms such as IAGO are suitable for the optimization of expensive-to-evaluate functions

in presence of noisy evaluations.

91




4. Case studies



92

Need to modify the distribution planning methods to assess the techno-economic impacts

of RES-integration solutions.

��


Network reinforcement = Reinforcement + alternatives = ?

TODAY TOMORROW

93

Points often neglected in the existing planning approaches

��



TODAY TOMORROW


MV and LV networks

DSO’s behavior


new RES

94

Proposed planning approach

��



TODAY TOMORROW


MV and LV networks

Define multi-variable planning

strategies


arrivals


DSO’s behavior


new RES

95


��



TODAY TOMORROW


MV and LV networks


strategies


arrivals


DSO’s behavior


new RES

Computation time

96


��



TODAY TOMORROW


MV and LV networks


strategies


arrivals


DSO’s behavior


new RES

Computation time

Use approximation methods to obtainfast and accurate

load-flows

97


��



TODAY TOMORROW


MV and LV networks


strategies


arrivals


DSO’s behavior


new RES

Computation time

Use approximation methods to obtainfast and accurate

load-flows

98

Proposed simulation tool

��



TODAY TOMORROW




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis


99

Possible case studies

-0.6 -0.5 -0.4 -0.3-0.25 -0.2 -0.1 0

0.5

1

1.5

2

2.5

tan(phi)min

NP

C [

pu,

base

NP

Cm

(-0.

25)]

NPCNPCm

NPCq25

NPCq75

NPCm*

��



TODAY TOMORROW

S

1 2 3 4 5 6 7 8 9 10048

12

P [

MW

/a]

PLV

PMV

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 00.6

0.8

1

1.2

1.4

f*5000 = 0.97 pu

*5000 = -0.25

f ( )

Compare different strategies

Study parameter impacts Optimize a strategyStudy one RES scenario

0 10

0.2

0.4

0.6

0.8

1

NP

Cm

[pu

]




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis


100

Adapt the proposed methods for an industrial application by the DSOs

Further work

Improve the procedure used to create profiles of consumption/generation

��


01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

11.2

Pc [

pu]

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1

Pw [

pu]

01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08 01/09 01/10 01/11 01/12 31/120

0.20.40.60.8

1

Pp

v [pu

]

Take into account novel RES-integration solutions

Tackle complex optimization problems




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis


?

?

?

)(min f

0)(s.t. g

?

? ?




Initial network

Variables

Scenario

Costs

Scenario parameters


ConstraintsActions

Technical outputs


Economic analysis


101��

• H. Dutrieux, G. Delille et B. François, “An innovative method to assess solutions for integrating renewable generation into distribution networks over multi-year horizons”, Proc. 23rd International Conference on Electricity Distribution (CIRED), article 1103, juin2015.

• H. Dutrieux, I. Aleksovska, J. Bect, E. Vazquez, G. Delille et B. François, “The Informational Approach to Global Optimization in presence of very noisy evaluation results. Application to the optimization of renewable energy integration strategies”, Proc. 47èmes Journées de Statistique de la SFdS (JDS), article 199, juin 2015.

• H. Dutrieux, G. Delille, B. François et G. Malarange, “Assessing the Impacts of Distribution Grid Planning Rules on the Integration of Renewable Energy Sources”, Proc. IEEE PowerTech, article 464270, juillet 2015.

�� ! "#$ %#&"#$& �''( ')#

2015 11 03 soutenance these v4 -...

Documents

Transcript of 2015 11 03 soutenance these v4 -...