BOSTON DENVER LONDON LOS ANGELES MENLO PARK MONTREAL NEW YORK SAN FRANCISCO WASHINGTON Effective Load Carrying Capability of Wind Generation: Initial Results with Public Data Presented by Ed Kahn at UCEI POWER Conference March 18,2005

2 Background: What are wind integration costs? Its all about reserves Wind power has an intermittent output profile Some kind of back-up requirement is necessary to accommodate fluctuations Extreme cases German grid rules require that wind power look exactly like thermal (next slide) Wind Collaborative study of regulation requirements: there are no back-up costs What is effective load carrying capability ? A measure of the incremental impact of new resources on planning reserves; based on traditional reliability methods, i.e. loss of load probability (LOLP) Related to resource adequacy issues  Stoft LICAP testimony for the NE ISO proposes a transformation of LOLP for making capacity payments Distinct from operating reserve issues Transmission costs are yet another integration issue

3 German grid rules imposes large integration costs (from Sacharowitz)

4 Overview of ELCC study Background California Wind Collaborative RPS Integration Study finds the capacity credit for wind generation is roughly equal to its capacity factor (about 25%) SCE asked Analysis Group to review study methods and make an independent assessment Methodology LOLP methods used in the RPS Study are generally reasonable, but their implementation is unclear The use of confidential ISO data defeats the purpose of transparency Data issues Public data on loads, imports and hydro dispatch must be used in the absence of the ISO data available to the RPS study Initial results Base case results do not replicate RPS study results ELCC for 2002 is 13% using SCE wind data, not the % found in the RPS study ELCC for 2003 is about 20% and for 1999 about 10%

5 Methodology  Formally, we define the probability that in a given hour available capacity is less than load. We call this the LOLP for hour i, or LOLP i LOLP i = Pr (∑ C j < L i ),(1) where C j is the random variable representing the capacity of generator j in hour i and L i is the load in hour i.  The annual LOLE index is defined over all hours of the year i as LOLE = ∑ LOLP i.(2)  Effective load carrying capacity (ELCC) is the amount of new load, call it ∆L, that can be added to a system at the initial LOLE, which we call LOLE I, after a new unit with capacity ∆C max is added. If we denote the random variable representing the available capacity of ∆C max by ∆C, then solving (3) for ∆L gives an implicit definition of ELCC. LOLE I = ∑ Pr (∑ C j + ∆C < L i + ∆L)(3)  In normalized form ELCC = ∆L/∆C max (4)

6 Data issues RPS study relied on ISO data for 2002 that is not publicly available Hourly hydro generation data Proprietary database for outages This analysis relies upon public data from ISO released by FERC in connection with the Western Energy Markets Investigation Hourly hydro from 2000 used as a proxy for 2002 Adjustments to load required to account for SMUD withdrawal from ISO control area Forced outage rates from Henwood database

7 Initial results ELCC depends upon coincidence of wind output with high LOLP hours LOLP is highly correlated with load  The coincidence of wind generation and summer peak demand is low (see next slide) There is year to year variation in this correlation Correlation was comparatively low in 2002; higher in 2003 ELCC is also sensitive to the concentration of LOLE in time; i.e. is 95% of LOLE in the top 20 hours or the top 50 hours We use “perfect load shaving” hydro dispatch in base case; this tends to concentrate the LOLE to the top load hours The LOLE can be spread out more by tightening the supply/demand balance and raising the total LOLE RPS results appear to have a bigger LOLE spread than Analysis Group base case (see Figures 3.1 and 3.2 and compare with following chart) It is unclear why the RPS study finds 50 high LOLP hours instead of 20

8 The coincidence of wind generation and demand is low

Base Case: Top LOLP Hours vs. Wind Generation

10 Sensitivity cases  Disaggregated wind data  Representation of hydro dispatch  Sensitivity to forced outage rates  2003 data  1999 data including adjustments for high LOLE levels

11 Sensitivity Cases: Hydro Dispatch, Low Forced Outage Rates and 2003 Data YearOutage RateHydro DispatchLOLEELCC 2002basePerfect % 2002baseImperfect % 20025%Perfect % 20025%Imperfect % 2003basePerfect4.68E % 2003baseImperfect % 20035%Perfect8.91E % 20035%Imperfect9.58E %

12 Sensitivity Tests: Wind Regions and 2002 and 1999 Data YearAggregationHydro DispatchLOLEELCC 2002Total QFPerfect % 2002Antelope Perfect % 2002Devers Perfect % 2002Vincent Perfect % 1999Total QFPerfect % 1999Antelope Perfect % 1999Devers Perfect % 1999Vincent Perfect %

Data with 2691 MW of Additional Perfect Capacity YearAggregationHydro DispatchLOLEELCC 1999Total QFPerfect % 1999 Antelope Perfect % 1999 Devers Perfect % 1999 Vincent Perfect % ELCC does not depend on LOLE, when LOLE is low, but the 1999 Results show LOLE is high (about 23 hours/year)  We add enough capacity to reduce LOLE to the “1 day in ten years” level (about 2.4 hours/year)  This reduces ELCC by 40% (from 16.5% to 10%)

14 Conclusions ELCC methods are a reasonable approach to determining the capacity value of wind generation, but  Implementing these methods raises issues  The RPS Integration relies on confidential data used in a non-transparent way  Year to year variation is substantial  This raises policy questions about how to compensate wind generators for capacity Other estimates of wind integration costs in the RPS study need further examination