1

R.12-03-014: Energy Division Workshop – Operating Flexibility Modeling

Nathaniel Skinner Senior Analyst, Generation & Transmission Planning

California Public Utilities Commission

September 19, 2012

Remote Access

2

Call in #: Passcode:866-758-1675 3481442Note: *6 to mute/unmute

Upon entry to the call, please place yourself on mute, and remain on mute unless you are asking a question

WebExMeeting Number: 741 769 312Meeting Password: LTPP

https://van.webex.com/van/j.php?ED=189577152&UID=491292852&PW=NNGQ4MGM0MTBk&RT=MiM0

Agenda

3

Time Item

10:00 - 10:10 Introduction

10:10 – 10:45 Revisiting the “Step 0” and “Step 1” methodologies

10:45 - 11:15 Review of activities between 2010 LTPP settlement and today

11:15 – 12:15 Review probabilistic methodology for evaluating flexibility

12:15 – 1:10 Lunch 

1:10 – 2:00 Should and how can operating flexibility criteria be understood within the context of NERC/WECC standards

2:00 – 2:30 Considering & modeling resources to meet operating flexibility needs

2:30 – 2:45 Break 

2:45 – 3:00 Scenarios and Assumptions and informing other processes

3:00 – 3:15 Thoughts on the Flexibility Procurement Modeling Challenge

3:15 – 3:45 Q&A Session

3:45 – 4:00 Wrap-up / Next steps

Workshop Purpose

• Review past (2010 LTPP) Operating Flexibility modeling approaches

• Examine proposed modeling approaches for the 2012 LTPP

• Begin framing discussion for meeting any needs identified for end of 2013 decision

4

Roadmap

5

PLANNING STANDARDS (6/21)

Straw Proposal

Workshop

Comments

Proposed Scenarios (8/2)

Workshop

Comments

SCENARIOS

2012 – System 2013 - System

Modeling

Need Determination

Meeting Needs

AUTHORIZATION System Needs

Local Area Needs Determination / Authorization Based on 2011/12 CAISO TPP

SDG&E Application & Track 1 LTPP (LA Basin + BC/Ventura)

Bundled Rules / Plans

CAISO 2013/14 TPP

Op Flex Meeting 1

Op Flex Meeting 2

Op Flex Meeting #

Slide 6

Operating Flexibility Analysis for R.12-03-014

Mark Rothleder, Executive Director, Market Analysis and Development

Shucheng Liu, Principal Market Developer

CPUC, Workshop September 19, 2012

Slide 7

Description of Past Method and Model

Slide 8

Study process quantifies operational requirements and evaluates fleets ability to meet operating requirements.

RenewablePortfolios

Variable Resource Wind / Solarand Load

Profiles

FlexibilityRequirements

(Regulation, Balancing)

DevelopProfiles

Shortages

Infrastructure Needs

Costs, EmissionsImport/Export

Capacity Factor

Statistical Analysis/model

Productionsimulation

Slide 9

Study Methodology

• Variability and uncertainty of renewable resources and load largely determine system regulation and load following requirements

• Currently, load is the dominant uncontrollable variable • 33% RPS introduces two additional uncontrollable variables that

impact load-following and regulation requirements – Wind, and

– Solar

• System regulation and load following requirements depend on three factors: – Forecast quality: Load, Wind and Solar forecast errors

– Interaction between load, wind and solar: net variability

– Market timeline: how fast the market re-commit and re-dispatch the controllable resources

Slide 10

CAISO Scheduling Process

MW

tOperating Hour

Hour Ahead Schedule

Day Ahead Schedule

Hour AheadAdjustment

Load Following

Generation Requirement

Regulation

Hour Ahead ScheduleAnd Load Following

Slide 11

Calculating hourly load-following requirement

• Load Following is defined as the difference between the 5-min forecasted net load and the hour ahead forecasted net load

• Determine the 95th percentile of maximum load-following requirement

swhrhahrha

swrtfrtf

lf GLGLG ,1,1,

,min5,min5,

Slide 12

Calculating hourly regulation requirement

• Regulation is defined as the difference between the net load (load– wind– solar) and the 5 minute forecasted net load (load–wind–solar)

• Determine the 95th percentile maximum regulation requirement

srtf

wrtfrtf

sa

waa

r GGLGGLG min5,min5,min5,

Slide 13

Flow chart for calculating load-following requirements

Slide 14

Flow chart for calculating regulation requirements

Slide 15

Review of activities between 2010 LTPP settlement and today

Slide 16

Additional sensitivity and analysis performed since July 2011

• PRM Analysis Deep Dive analysis of PRM– All-Gas reserve margin overstated PRM– All-Gas case PRM~21% instead of 41%

• Step 1 Sensitivity– Assessed and bounded impacts of forecast errors– Assessed drivers of flexible ramp

• 5 minute simulation– Similar but slightly reduced violations observed

• Regional modeling and coordination– Improved modeling of the GHG improves the regional flows– If not import constrained regional coordination can improve

access to flexible resources

Slide 17

Operational criteria within the context of NERC/WECC standards

Slide 18

Balancing Authority ACE limit (BAAL) – System control opposes frequency deviations

-500

-300

-100

100

300

500

59.80 59.85 59.90 59.95 60.00 60.05 60.10 60.15 60.20

Frequency

AC

E (

MW

)

Balancing Authority ACE Limit RADAR

CurrentLast 5 minsLast 6-15 minsLast 16-30 mins

-485.00 MW / 0.1 Hz

CISO 4/23/12 15:03

000000000014:55 15:0215:0115:0014:5914:5814:5714:56Time

Consecutive Minutes Exceeding Limit(s)

00000000014:44 14:45 14:5214:5114:5014:4914:4814:4714:46

000000000014:5414:35 14:4314:4214:4114:4014:3914:3814:3714:36

014:53

60.00Hz Scheduled Frequency

15:03:4415:03

BIAS

ACE

Frequency

33.08

ACPS1

60.022

191

BAAL shall not be exceeded for more than 30 consecutive clock-minutesNumber of BAAL Exceedances

in last 30 minutes =

BAAL shall not be exceeded for more than 30 consecutive clock-minutes

CTRL-T to start timerCTRL-S to stop timer

0

• BAAL is designed to replace CPS2

• BAAL relaxes area regulation needs

• ACE is allowed to be outside BAAL for up to 30 minutes

Slide 19

Control Performance Standard Scores (CPS1) Scores January 2009 through April 2012

0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

150

160

170

180

190

200

Per

cen

t (%

)

CPS1

CPS 1 Scores – January 2009 through April 2012

Began operating to BAAL

Slide 20

The assessment of a balancing authority control performance is based on three components

• Control Performance Standard (CPS1) - measures the control performance of a BA's by comparing how well its ACE performs in conjunction with the frequency error of the Interconnection

• Balancing authority Ace Limit (BAAL) - is a real-time measure of Area Control Area and system frequency which cannot exceed predefined limits for more than 30-minutes

• Disturbance Control Standard (DCS) - is the responsibility of the BA following a disturbance to recover its ACE to zero if its ACE just prior to the disturbance was greater than zero or to its pre-disturbance level if ACE was less than zero - within 15 minutes

• Control Performance Rating

Pass is when CPS1 ≥ 100%; BAALLimit ≤ 30 minutes & DCS = 100%

Slide 21

A Stochastic Model for Analyzing Ramping Capacity Sufficiency

Slide 22

A stochastic model is needed to assess the probability of upward ramping capacity sufficiency.

• A deterministic production simulation case adopts only one of the many possible combinations of input assumptions

• A stochastic model can evaluate various input combinations based on probability distributions of the stochastic input variables

• Monte Carlo simulation determines insufficient ramping capacity probability

• It complements the deterministic production simulation

Slide 23

Available ramping capacity depends on the balance of supply and demand.

Supply curve is constructed based on variable cost of each generation unit

Slide 24

Uncertainties in supply and demand affect availability of ramping capacity.

Slide 25

Available ramping capacity of each generation unit is determined based on the following factors:

• Maximum and minimum capacity

• Unit availability (due to forced and maintenance outages)

• Dispatch level

• Ramp rate

• Ramp time allowed (10 or 20 minutes)

Slide 26

Ramping capacity shortage may occur due to variations in both availability and requirement.

Slide 27

This stochastic model considers uncertainties in some of the key inputs, including:

• Load forecast

• Inter-hour load ramp

• Requirements for regulation-up and load following-up

• Generation by wind, solar, and hydro resources

• Availability of generation units (due to forced and maintenance outages)

Slide 28

A model is developed for a time period in which all hours have similar conditions.

• No unit commitment• No chronologic constraint (such as min run time and min

down time, etc.)• Independent with identical probability distribution

functions for each hour in the period• Insufficient ramping capacity probability for each hour

determined through Monte Carlo simulations• Insufficient ramping capacity probability for the whole

year calculated based on Binomial distribution

Slide 29

Probability distributions are developed based on data from the Plexos production simulation model.

• Hourly load forecast

• Hourly regulation and load following-up requirement

• Hourly wind, solar, and hydro generation

• Uniform distribution functions based on forced and maintenance outage rates of each generation unit

Slide 30

Inter-hour load ramp is calculated based on hourly load forecast.

Upward direction only

A new stochastic variable

Met by 60-min ramping capability

A part of load

)0max( 1 ttt LoadLoad,RampInter-Hour

Slide 31

These are examples of probability distribution functions of stochastic variables.

5.0% 95.0% 0.0%3.8% 96.2% 0.0%

38,935 69,949

35

,00

04

0,0

00

45

,00

05

0,0

00

55

,00

06

0,0

00

65

,00

07

0,0

00

75

,00

08

0,0

00

0

1

2

3

4

5

6

7

Val

ue

s x

10

^-5

Fit Comparison for Total CA LoadRiskBetaGeneral(1.5122,1.5704,36266,70059)

Input

Minimum36,429.66Maximum69,948.41Mean 52,846.29Std Dev 8,409.00Values 630

BetaGeneral

Minimum36,266.00Maximum70,059.00Mean 52,843.49Std Dev 8,360.86

1-in-2 forecast

5.0% 94.1% 0.9%3.1% 94.2% 2.6%

38,935 69,949

35

,00

04

0,0

00

45

,00

05

0,0

00

55

,00

06

0,0

00

65

,00

07

0,0

00

75

,00

08

0,0

00

0

1

2

3

4

5

6

7

Val

ue

s x

10

^-5

Fit Comparison for CA Load ExtendedRiskBetaGeneral(1.9596,2.7721,35699,77352)

Input

Minimum36,429.66Maximum76,569.35Mean 53,035.74Std Dev 8,593.85Values 636

BetaGeneral

Minimum35,699.00Maximum77,352.00Mean 52,949.29Std Dev 8,569.90

1-in-10 forecast

1-in-2 forecast

Exploring the probability to have load higher than 1-in-2 forecast

Slide 32

Examples of probability distribution functions of stochastic variables. (cont.)

50. % 90.0% 5.0%6.0% 90.0% 4.1%

645 880

40

05

00

60

07

00

80

09

00

10

00

11

00

12

00

0.000

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

Fit Comparison for Total RegURiskLogLogistic(354.15,383.65,9.9920)

Input

Minimum 506.1000Maximum1172.3400Mean 746.1442Std Dev 77.5920Values 630

LogLogistic

Minimum 354.1500Maximum +∞Mean 744.1946Std Dev 72.2464

5.0% 90.0% 5.0%2.0% 82.9% 15.1%

3169 6993

10

00

20

00

30

00

40

00

50

00

60

00

70

00

80

00

0

1

2

3

4

5

6

7

8

Val

ue

s x

10

^-4

Fit Comparison for Hydro GenRiskBetaGeneral(16.221,1.9444,-7509.4,7572.3)

Input

Minimum1275.9663Maximum7500.7600Mean 5975.4646Std Dev 1214.3980Values 630

BetaGeneral

Minimum-7509.4000Maximum7572.3000Mean 5957.9751Std Dev 1065.0696

Slide 33

Examples of probability distribution functions of stochastic variables. (cont.)

5.0% 90.0% 5.0%6.0% 89.8% 4.2%

647 4774

-10

00

01

00

02

00

03

00

04

00

05

00

06

00

07

00

0

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

Val

ue

s x

10

^-4

Fit Comparison for Wind GenRiskInvGauss(2611.5,10804.9,RiskShift(-489.50))

Input

Minimum 197.6900Maximum6509.0300Mean 2121.9544Std Dev 1245.0564Values 630

InvGauss

Minimum-489.5000Maximum +∞Mean 2122.0000Std Dev 1283.8807

5.0% 90.0% 5.0%3.6% 88.8% 7.6%

4236 9454

10

00

20

00

30

00

40

00

50

00

60

00

70

00

80

00

90

00

10

00

0

0

1

2

3

4

5

6

Val

ue

s x

10

^-4

Fit Comparison for Solar GenRiskBetaGeneral(6.3810,1.1541,-2999.5,9688.9)

Input

Minimum2491.6537Maximum9688.6904Mean 7751.9740Std Dev 1620.0039Values 630

BetaGeneral

Minimum-2999.5000Maximum9688.9000Mean 7745.5041Std Dev 1564.1544

Slide 34

Correlations among the stochastic variables are enforced.

LoadLoad Ramp

Wind Gen

Solar Gen

Hydro Gen

RegU LFU

Load 1 0.2884 -0.0947 -0.1997 0.4302 0.3801 0.0722

Load Ramp

0.2884 1 -0.3782 0.6156 0.0779 0.2064 -0.3193

Wind -0.0947 -0.3782 1 -0.1618 0.2855 -0.0108 0.0609

Solar -0.1997 0.6156 -0.1618 1 0.0254 -0.1101 -0.5064

Hydro 0.4302 0.0779 0.2855 0.0254 1 0.3094 -0.1283

RegU 0.3801 0.2064 -0.0108 -0.1101 0.3094 1 0.1415

LFU 0.0722 -0.3193 0.0609 -0.5064 -0.1283 0.1415 1

This is an example of the correlation matrix

Slide 35

Generation units in the stochastic model have the following characteristics from the Plexos model.

• From input data– Maximum and minimum capacity– Ramp rate– Forced outage and maintenance outage rates

• From simulation results– Average generation cost (to determine an initial dispatch order)

Slide 36

Generation unit availability is stochastically determined in each iteration of the Monte Carlo simulations.

• Forced and maintenance outages are determined independently for each generation unit

• Each of the outages is determined based on the unit’s outage rate and a draw using a uniform distribution function

• A maintenance outage allocation factor is used to represent seasonal pattern of maintenance schedules

• A unit is unavailable when any one of the outages occurs

Slide 37

Contributions of a generation unit to energy and ramping capacity are subject to:

• 10-min upward ramping capacity constraint

• 20-min upward ramping capacity constraint

• 60-min upward ramping capacity constraint

• Maximum capacity constraint

),10min( iiii MinCapMaxCapRampRateAS

),20min( iiiii MinCapMaxCapRampRateLFUAS

iiiii MaxCapLdRampLFUASE

energy dispatch total upward ancillary service contribution

load following up contribution inter-hour load rampi i

i i

E AS

LFU LdRamp

),60min( iiiiii MinCapMaxCapRampRateLdRampLFUAS

Slide 38

Total contributions by all generation units should meet energy and ramping capacity requirements.

• 10-min upward ancillary service requirement

• 20-min upward ramping capacity requirement

• 60-min upward ramping

• Energy balance

i ASi

AS Req

i i AS LFUi

AS LFU Req Req

i ii

E LdRamp Load

upward AS requirement load following-up requirement

inter-hour load ramp total loadAS LFU

Load Ramp

Req Req

Req Load

i i i AS LFU Load Rampi

AS LFU LdRamp Req Req Req

Slide 39

The model seeks a least-cost solution to meet energy and all ramping capacity requirements.

• Generation units are dispatched economically to meet load first

• Remaining qualified ramping capacity is used to meet upward ancillary service, load following, and inter-hour load ramp requirements

• Energy dispatch and ramping capacity contributions are co-optimized when there is insufficient ramping capacity initially

Slide 40

Monte Carlo simulation determines insufficient ramping capacity probability.

• Monte Carlo simulation is conducted using this stochastic model

• Insufficient ramping capacity results are presented in a probability distribution format

• The key results are the probability to have ramping capacity shortage each hour and the probability distribution of the volume of the shortages

Slide 41

This example has a 0.8% probability to have 20-min ramping capacity shortage each hour.

0.0% 0.8% 99.2%

-4,661 -1

-6,0

00

-4,0

00

-2,0

00

02

,00

04

,00

06

,00

08

,00

0

0.00

0.05

0.10

0.15

0.20

0.25

20- min Ramping Capacity Sufficiency

20-min Ramping Capacity Sufficiency

Minimum -4,660.87Maximum 7,267.15Mean 1,400.35Std Dev 1,158.94Values 5000

Results for the Super-Peak period.

Slide 42

The highest 20-min ramping capacity shortage is 4,661 MW in this example.

100.0%105 4,661

05

001

,00

01

,50

02,0

002

,50

03,0

003

,50

04

,00

04,5

005

,00

00.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

20- min Ramping Capacity Shortage

20-min Ramping Capacity Shortage

Minimum 105.39Maximum 4,660.87Mean 1,331.02Std Dev 1,142.42Values 39 / 5000Filtered 4961

Results for the Super-Peak period.

Slide 43

The probability to have 10-min ramping capacity shortage each hour is 0.1%.

0.0% 0.1% 99.9%

-2,180 -1

-3,0

00-

2,0

00-

1,0

00

01

,00

02

,00

03

,00

04

,00

05

,00

06

,00

0

0.00

0.02

0.04

0.06

0.08

0.10

0.12

10- min Ramping Capacity Sufficiency

10-min Ramping Capacity Sufficiency

Minimum -2,179.62Maximum 5,262.96Mean 1,618.69Std Dev 649.86Values 5000

100.0%219 2,180

20

04

00

60

08

001

,00

01,2

001

,40

01,6

001

,80

02,0

002

,20

0

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

10- min Ramping Capacity Shortage

10-min Ramping Capacity Shortage

Minimum 219.17Maximum 2,179.62Mean 1,334.97Std Dev 762.50Values 6 / 5000Filtered 4994

Results for the Super-Peak period.

Slide 44

Monte Carlo simulation results for all periods are summarized as follows:

10-min 20-min 10-min 20-min# of Hours in the Period 630 630 2298 2298Probability of Shortage 0.12% 0.78% 0.04% 0.16%

Max Shortage (MW) 2,180 4,661 1,420 3,855

Super-Peak Summer Off-Peak

Example Case

Slide 45

Cumulative probabilities of ramping capacity shortage are calculated using Binomial distribution.

i 10-min 20-min

1 81.3% 100.0%2 49.9% 99.8%3 23.6% 99.1%4 8.9% 97.2%5 2.8% 93.0%6 0.7% 85.8%7 0.2% 75.4%8 0.0% 62.7%9 0.0% 49.0%

10 0.0% 35.9%11 0.0% 24.6%12 0.0% 15.9%13 0.0% 9.6%14 0.0% 5.5%15 0.0% 2.9%16 0.0% 1.5%17 0.0% 0.7%18 0.0% 0.3%19 0.0% 0.1%20 0.0% 0.1%21 0.0% 0.0%22 0.0% 0.0%

Example Case

It is the probability to have at least i hours with ramping capacity shortage in year 2020.

Slide 46

Expected number of hours with ramping capacity shortage in 2020 are calculated based on the probabilities.

10-min 20-min

1.68 8.59

Example Case

It is insufficient ramping capacity expectation.

Slide 47

What we learned from this approach:

• It does– Use probability distributions to capture uncertainties in key input

factors– Implement ramping constraints and flexibility requirements– Present insufficient ramping capacity events in probabilistic

format

• It does not– Decide unit commitment– Impose chronological constraints

• It can be improved to– Use multi-year synchronized historical data to capture more

variations of input stochastic variables

Slide 48

A Study to Support Meeting Assembly Bill (AB) 1318

Slide 49

Plexos simulations are conducted about performance of local capacity requirement resources in 2020.

• Base model: High-Load scenario in 2011 LTPP study– Reduced min run time of some demand response resources

from 4 to 1 hour

• LCR resources– Added 3,173 MW local capacity requirement (LCR) resources

• A sensitivity case– Reduced capacity of event-based demand response resources

Slide 50

LCR resources are added to SCE and SDG&E zones.

• 3,173 MW LCR resources based on the ISO OTC study– Los Angeles Basin: 2,370MW– Big Creek Ventura areas: 430MW– San Diego: 373MW*

• LCR resources added as a combination of CCGT and GT units– SCE: 2 x 500 MW CCGT units– SCE: 18 x 100 MW LMS100 GT units– SDG&E: 1 x 373 MW CCGT unit

* It assumes that San Diego proposed generation is included already. So the total need in San Diego should be 373 MW plus

Pio Pico = 300MW Quail Brush = 100MW Escondido Energy Center = 45 MW

Slide 51

LCR resources have generic operating characteristics comparable to newer exiting resources.*

Resource Max/Min Capacity

(MW)

Full-Load Heat Rate (Btu/kWh)

Ramp Rate

(MW/min)

Forced Outage Rate (%)

Maintenance Rate

(%)

Start-up Time (hour)

Start-up Cost ($)

SCE NEW GT 100/40 9,191 12.0 7.24 10.0 1,200

SCE NEW CCGT 500/200 7,000 7.5 4.96 10.0 2 44,520

SDGE NEW CCGT 373/200 7,000 7.5 4.96 10.0 2 44,520

Gateway (CCGT) 530/265 7,000 10.0 10.00 10.0 2 24,411

Sentinel (GT) 106/43 9,191 12.0 10.00 10.0 1,000

* Forced outage rates of the new resources are based on NERC GADS 2006-2010 average EFORd, CCGT for all MW sizes and GT for 50 plus MW

0

2

4

6

8

10

12

14

16

18

Freq

uenc

y

Ramp Rate (MW/min)

Histogram of ISO CCGT Unit Ramp Rates

New CCGT

05

101520253035404550

Freq

uenc

y

Ramp Rate (MW/min)

Histogram of ISO GT Unit Ramp Rates

New GT

Slide 52

LCR resources have high capacity factors and contributions to ancillary services and load following.

Resource 1 2 3 4 5 6 7 8 9 10 11 12 Annual

SCE NEW GT 9.5 11.2 10.0 9.8 12.0 16.5 20.3 17.9 7.9 10.0 8.0 10.2 11.9

SCE NEW CCGT 53.1 60.0 61.4 64.2 59.4 64.1 73.7 83.4 80.9 66.9 61.1 68.3 66.4

SDGE NEW CCGT 49.2 62.1 55.9 20.4 72.6 76.5 69.0 87.4 83.7 50.9 37.8 20.3 57.1

Gateway (CCGT) 52.0 45.6 55.3 48.7 45.5 56.1 62.8 55.2 60.1 56.2 60.3 60.7 54.9

Sentinel (GT) 22.1 20.3 17.2 18.3 21.1 19.6 20.4 19.1 11.6 16.2 16.0 12.1 17.8

GT Average 10.9 10.7 8.0 10.8 10.9 12.0 11.2 9.5 6.6 8.4 9.3 10.4 9.8

CCGT Average 48.5 45.9 40.6 39.8 36.1 40.2 62.0 65.4 55.1 51.0 49.6 51.9 49.4

Resource LF Down LF Up Non-Spin Reg-D Reg-U Spin

SCE NEW GT 23.9 537.3 1.9 32.1 320.0 914.8

SCE NEW CCGT 1,888.0 849.2 0.5 101.8 11.6 577.2

SDGE NEW CCGT 264.9 217.8 0.0 202.7 78.6 56.4

Monthly Capacity Factors

Ancillary Service and Load Following Contribution (GWh)

Slide 53

LCR resources are highly utilized.

• Example - utilization of SCE NEW CCGT resource– Max capacity: 1,000 MW (2 x 500 MW)– Total possible generation in 2020: 8,784 x 1,000 = 8,784 GWh– Annual capacity factor: 66.40% (5,833 GWh generation)– Upward AS and load following contribution: 1,439 GWh– Generation plus AS and LF contribution: 7,271 GWh– Total utilization rate: 82.78%– Forced and maintenance outage rate: 14.96%– Unutilized capacity: 2.26%

Slide 54

Additional flexible capacity is needed in addition to the LCR resources.

• Up to 1,251 MW 20-min ramping capacity shortage identified

• Consistent with the findings in previous study– Need for 4,600 MW LMS100 GT capacity with zero outage

identified in previous study– 3,173 MW LCR capacity added– 1,251 MW residual shortage– Additional demand response usage from reduced minimum run

time

Slide 55

In the sensitivity case event-based demand response capacity is reduced by 40.7%.*

Region Original DR Capacity

(MW) Reduced DR Capacity

(MW) Reduction

(MW)

PG&E 1,687 732 955

SCE 2,827 1,977 850

SDGE 302 146 156

TOTAL 4,816 2,855 1,961

* Energy usage limits for the demand response resources are not reduced so some of the resources may be deployed for more hours.

Slide 56

Demand response resources are used frequently in summer months.

Case Jul Aug Sep Oct Sum

Original DR Capacity 44 22 3 2 71

Reduced DR Capacity 47 23 2 2 74

Number of Hours DR Resources Deployed

Slide 57

Ramping capacity shortage increases with reduction of demand response capacity.

• Utilization of LCR resources shows small changes

• 20-min ramping capacity shortage is increased from 1,251 MW to 3,212 MW– 1,961 MW increase in shortage with 1,961 MW reduction in

demand response capacity

• Demand response capacity is “flexible”– No ramp rate constraint– Not dispatchable (min capacity = max capacity)– Freeing up flexible dispatchable capacity once deployed

Slide 58

Methodologies for Next Phase of Study

Slide 59

Three methodologies were used in the previous phase of study.

• Production cost simulation– Detail, chronological, unit commitment, flexibility constraints– Slow, deterministic

• Loss of Load Probability (LOLP) calculation– Probabilistic, fast, multi-year historical data, probabilistic outputs– No flexibility constraint, non-chronological

• Stochastic simulation modeling– Stochastic, fast, flexibility constraints, probabilistic outputs– Non-chronological, no unit commitment, single-year data

Slide 60

The methodologies for the next phase of study should be able to:

• Use stochastic modeling approach to capture uncertainties in key input variables– Load, renewable and hydro generation, flexibility requirements,

resource availability, etc.

• Enforce operational constraints– Chronological linkages between intervals– Ramping constraints and flexibility requirements

Slide 61

The methodologies for the next phase of study should be able to: (cont.)

• Identify capacity and flexibility shortages– Differentiating capacity shortages from flexibility shortages– interpreting in terms of operational performance standards

(BAAL, CPS1, etc.)

• Propose solutions to meet the performance standards– Quantifying in term of capabilities not technologies

• Have manageable calculation/simulation time

62

Scenarios and Assumptions

Nathaniel Skinner Senior Analyst, Generation & Transmission Planning

California Public Utilities Commission

September 19, 2012

Scenarios Purpose• Inform policy-makers by providing

information on a broad range of plausible future scenarios

• Inform bundled procurement plans and positions

• Inform the transmission planning process and analysis of operating flexibility

• Limit the range of analysis to conform with resource constraints, while meeting policy objectives for the current LTPP63

Problem Statement

Scenarios should be developed to answer the following primary questions:

•What new infrastructure needs to be constructed to ensure adequate reliability?

•What mix of infrastructure minimizes cost to customers over the planning horizon?

64

Planning Area & Period

65

• CAISO controlled transmission grid & distribution systems

• Period 1 – 10 years ahead– Today through 2022

• Period 2 – 20+ years ahead– 2023 through 2034– Simplified analysis to understand impacts of

choices made to meet Period 1 needs

Example Scenario: 1 - Base• Reflect expectations of the future with little change

from existing policies• Key Assumptions:

– Mid load, mid inc. EE, mid small PV, low CHP– Mid DR, high probability additions, commercial RPS– Retirements: Low nuclear, low hydro/wind/solar, mid

other

How to get there: No change to business as usual and programs achieve results consistent with forecast expectations

66

Thoughts on the Flexibility

Procurement Modeling ChallengeCPUC Workshop

9/19/2012

Arne Olson, E3 Partner

Agenda

What Question Are We Trying to Answer?

What Tools Are Available Today?

What Would a Hybrid Model Look Like?

68

What Question Are We Trying to Answer?

Today’s Planning Problem Has Two Related Questions

1. How many MW of dispatchable resources do we need to (a) meet load, and (b) meet ramping requirements on various time scales?

2. What is the optimal mix of new resources, given the characteristics of the existing fleet?

Answer is a Matter of Matching Demand and Supply

Demand is a function of a number of variables:

• Load

• Load forecast error

• Load variability

• Expected renewable production

• Expected renewable production forecast error

• Expected renewable production variability

Supply is the ability of the fleet of dispatchable resources to respond on the appropriate time scale

71

Problem is Stochastic in Nature

Load is stochastic, variableand uncertain

• Often characterized as “1-in-5” or “1-in-10”

• Subject to forecast error

Renewable output is also stochastic, variable and uncertain

Supplies can also be stochastic

• Hydro endowment varies from year to year

• Generator forced outages are random

Need to know size, probability and duration of any shortfalls

72

Stochastic Modeling Must be Robust

Need will occur during “tail” events for both demand and supply

• Need enough iterations to accurately capture low-probability events

Flexibility need shortages will be related to capacity shortages

• Inflexible capacity can be a substitute for flexible capacity under some circumstances

73

Loss of Load Events

Need Must Be Defined on a Number of Time Scales

Peak Load is an annual phenomenon

• Loss of Load can occur in 50-250 hours per year

• Typically measured on an hourly basis

Ramping needs must be defined over much smaller time increments

• 5 hours

• 1 hour

• 20 minutes

• 5 minutes

• 1 minute

74Source: Russ Philbrick, PES General Meeting, Detroit, July 2011

What Tools are Available Today?

Planning Models

Calculate Loss-of-Load Probability (LOLP) and related metrics to determine the probability that resources will be inadequate to meet peak loads

• Conducted to support calculation of a reserve margin or to determine least-cost expansion plan

Advantages

• Stochastic model that considers the full range of load conditions

Limitations

• Ignores operations

• In the past, planning to prevent loss of load has provided enough flexibility

76

Operations Models

Production simulation models minimize operating costs subject to constraints on unit availability, transmission availability, etc.

• Conducted to determine the total cost associated with meeting electric load over 8760 hours

Benefits

• Treats details of generator operating limitations, transmission, and time-sequential behavior

Limitations

• Deterministic

• Significant unnecessary detail

• Lengthy run time

77

Enhanced Planning Models

E3’s LOLP/ELCC Model incorporates reservesinto LOLP framework

Does not treat flexibility characteristics explicitly

• Flexibility need modeled as fixed hourly parameter

• Does not measure ability of fleet to meet flexibility needs

• Requires reference to a portfolio that is assumed to be sufficiently flexible (All-Gas Case)

78

E3 LOLP Model Flow Chart

5

Hourly load

Generator Model

Net Load Model

Ancillary Service Model

Hourly wind

Hourly solar

Thermal fleet

Hourly Reg. Up

Hourly LFU

Outage Probability

Table

Net Load Mean & Variance

Spin, LFU & Reg. Up

Requirement

LOLP Model

MW of Need ELCCTarget PRMLOLP/ LOLE

Spinning Reserve

Hydro NQC

Import limits

What Would a “Hybrid” Model Look Like?

Planning for Reliable Operations

A model is needed that plans for reliable system operations, satisfying:

• Capacity Requirement – according to traditional metrics for capacity planning

• Flexibility Requirement – accounting for the limitations of the fleet in time sequential operations

Measures taken to meet need in one category may satisfy need in the other

• The model should select the least-cost array of portfolio/operational changes to meet flexibility

• Differentiated value of resource types over different time scales should be captured

• Storage provides fast response ramping over short time periods, CCGTs provide capacity and ramp, etc.

80

Flexibility Requirement – Defining new set of metrics

Attempts to quantify unit flexibility (NERC’s IVGTF Task 1.4)

Still unanswered: How could these metrics be used in a procurement process?• How would standards be determined and adopted?

• Is there a way to compare flexibility with adequacy using a common currency?

81

LOLP

FOR

ELCC

Inadequate Ramp Resource Probability (IRRP)

Ramp Availability Rate (RAR)

Effective Ramping Capability (ERC)

Power Reliability Metrics: Ramp Reliability Metrics:*

*Proposed by Lannoye et al.

Best solution to satisfy need

Two options to mitigate flexibility violations:

A useful model will be able to quantify the trade-offs between these options

• What measure or combination of measures satisfies need?82

Flip a switch

Operational changes can mitigate flexibility shortfalls

• Reserve scheduling, “pre-curtailment” of renewables

Grab a shovel

Steel in the ground can help to meet both capacity and flexibility requirements

• Fast, expensive resources vs. cheaper, slower ones

The “Sledgehammer” Approach: Stochastic Production Simulation

Minutely time step resolution

Monte Carlo for forecast errors

Requires large datasets• Detailed load, wind, solar datasets

• Individual unit specifications

• Scheduled and forced outages

• Hydro and import conditions

83

Run time: full stochastic simulation may be impractical

Year-long simulation does not capture long-term uncertainties, important for planning analysis

Flexibility of system depends on chosen reserve requirements – possibility of “false violation”

Difficult to incorporate expansion decision

Challenges

One Path Forward: Reduced-Form Production Simulation Modeling

Three key modifications to production simulation modeling framework:

1. Stochastic operations: Run thousands of draws of a single day per month to accurately characterize long-term uncertainty

• Preserve time-sequential unit commitment and operations over 24 hours

2. Endogenous reserves: Include endogenous, minutely specification of reserve flexibility requirements to avoid false violations and accurately characterize fast-ramping resource

3. Expansion decisions: Incorporate operational and expansion decisions (with fixed costs) to find optimal solutions

Requires elimination of all detail that doesn’t help answer question at hand in order to minimize run time (e.g., transmission)

84

+

Thank You!Energy and Environmental Economics, Inc. (E3)

101 Montgomery Street, Suite 1600

San Francisco, CA 94104

Tel 415-391-5100

Web http://www.ethree.com

Arne Olson, Partner (arne@ethree.com)

86

Wrap Up & Next Steps

Nathaniel Skinner & Noushin KetabiSenior Analysts, Generation & Transmission Planning

California Public Utilities Commission

September 19, 2012

Operating Flexibility #3

• Planning for an early November third workshop– Would explore the path forward for operating

flexibility analyses and interpreting these needs into the LTPP

– In person or fully over WebEx?

87

CalendarSeptember

19: Track II Final Scenarios to be sent to service list

24: Track I Briefs Due

October

5: Track II Final Scenarios Comments Due LTPP/Energy Storage Workshop Comments Due

12: Track I Reply Briefs Due

19: Track II Final Scenarios Reply Comments Due LTPP/Energy Storage Workshop Reply Comments Due

88

CalendarNovember

2: Track III Rules Comments Due [PENDING ALJ RULING]

20: Track II Scenarios PD

December

12: Track I PD

20: Track II Scenarios Decision on Agenda

January

TBD Commission Meeting: Track I Decision on Agenda

89

90

Thank you!For Additional Information:

www.cpuc.ca.govwww.GoSolarCalifornia.ca.gov

www.CalPhoneInfo.com