Working Group's Report on

The Methodology and Key Input Assumptions

To Review the Current

Planning Reserve Margin Requirement

R.08-04-012

June 16, 2008

(Version 3)

TABLE OF CONTENTS

Page

1. Introduction 1

2. Scope and Status 3

2.1. Scope 3

2.2. Status 4

3. Study Methodology 6

3.1. Overview 6

3.2. Introduction to GE Energy's MARS Model 7

3.3. Description of CAISO Revised Study Scope and Work Plan 9

3.4. Expected Results 13

4. Status of Working Group Discussions About Input Assumptions and Modeling Issues 14

4.1. Working Subgroup 1 - General Issues 14

4.1.1. Issue #1: Should the Commission use the 0.1 days per year LOLE standard or a different metric in determining the final PRM requirement? 14

4.1.2. Issue #2: Should the Commission adopt one CAISO-wide PRM or one PRM for each CAISO sub-area? 15

4.1.3. Issue #3: At what operating reserve level should involuntary curtailment be measured to determine the desired PRM? 17

4.1.4. Issue #4: What is the best mechanism to share the input data within the Working Group and ultimately with the public? 18

4.2. Working Subgroup 2 - Intermittent Resource Modeling 18

4.2.1. Issue #5: How to model intermittent resources in MARS to properly reflect the correlation of wind and load? 18

4.2.2. Issue #6: Modeling of Hydro Generation 20

4.3. Working Subgroup 3 - Load and Demand Response Inputs to MARS Simulation 21

4.3.1. Issue #7: Load Forecast 21

4.3.2. Issue #8: Modeling Uncertainty of Load 22

4.3.3. Issue #9: Adjusting Historic Load Shapes for Demand Response 27

4.3.4. Issue #10: Modeling of Demand Response 28

4.3.5. Summary 28

4.4. Working Subgroup 4 - Modeling of Current Generation 29

4.4.1. Generation Forced Outage Data 29

4.4.2. Preferred Data Source 29

4.4.3. Confidentiality of Individual Plant Data 29

4.5. Working Subgroup 5 - Modeling of Transmission Limitations and Imports 30

4.5.1. Issue # 11: Modeling of Transfers Between Sub-Areas Within the CAISO Region 30

4.5.2. Issue # 12: Modeling of imports 31

5. Next Steps 32

Attachment 1 Proposals for Load Forecast Inputs to GE MARS Model

Attachment 2 General Electric's MARS Program Description

Working Group's Report on
The Methodology and Key Input Assumptions
To Review the Current
Planning Reserve Margin Requirement
(R.08-04-012)

1. INTRODUCTION

On April 10, 2008, the California Public Utilities Commission (Commission or CPUC) opened Rulemaking (R.) 08-04-012 "to review, and modify to the extent found to be appropriate, the Planning Reserve Margin (PRM) and the assumptions, methods, and procedures used for its determination."¹

In parallel, the California Independent System Operator (CAISO) initiated a PRM stakeholder process, and held initial stakeholder meetings in late November 2007 to review a preliminary study scope and proposals by potential vendors to perform a study. After reviewing proposals from four vendors (GE Energy, Associated Power Analysts, Siemens PTI, Ventyx Inc.) and receiving comments from stakeholders, the CAISO announced on March 5, 2008² that it had selected GE Energy (GE Energy or Vendor) to perform the study. The primary reasons for selecting GE Energy to perform the PRM study were: (a) wide acceptance of Vendor software by all other Independent System Operator (ISO) organizations in the country, (b) Vendor experience in performing PRM studies and providing on-going support to other major ISOs (PJM, NENY-ISO, ISO-NE and MISO) and reliability organizations, (c) Vendor ability to meet the CAISO study schedule, and (d) competitive bid pricing. The CAISO and the CPUC intend to merge their PRM stakeholder processes (although the CAISO will remain a party to the proceeding) and going forward will work on an integrated basis in R.08-04-012. In this report, the "PRM Study" refers to this joint stakeholder process unless otherwise noted.

On April 15, 2008, staff from the Commission's Energy Division, the CAISO and the California Energy Commission (CEC) met with representatives from GE Energy and the California investor-owned utilities (IOUs)³ to review the study scope and work schedule for the PRM Study, and to prepare for a future stakeholder workshop planned for June 2008.

Following this meeting, five working subgroups were formed to prepare materials that would be useful for discussion at the June 25 and 26 workshops scheduled in R.08-04-012.⁴ These five working subgroups produced this Working Group Report, and are collectively referred to as the "Working Group" in this report. The Working Group is currently comprised of representatives from the Commission's Energy Division, the CAISO, the CEC, GE Energy, and the IOUs, although other stakeholders are encouraged to join and participate in any of the subgroups.

This report presents the materials prepared by the Working Group for discussion at the June 25 and 26 workshops. Specifically, it:

· Presents an updated PRM Study scope and status;

· Familiarizes stakeholders with the probabilistic approach that will be used for analyzing and determining the PRM in this proceeding;

· Develops data sources that can be used for a preliminary study of the year 2010 only (Preliminary Study) to inform more refined data gathering for the final study currently slated for Phase II of the PRM OIR (Final Study); and

· Presents key inputs and modeling issues that the Working Group identified, and the Working Group's recommendations for addressing these inputs and issues in the June 25 and 26 workshops.

2. SCOPE AND STATUS

2.1. Scope

The Commission's PRM preliminary scoping memo describes joint development of a PRM Study with the CAISO and describes a multi-phased proceeding, where the first phase will adopt the methodology, input assumptions, sources of data, and scenarios, and the second phase will determine the proper PRM for the Resource Adequacy (RA) program's 2010 and 2011 compliance years.⁵

Phases one and two of the PRM Study will determine the PRM that meets specified Loss of Load Expectation (LOLE) levels considering load and resource uncertainties, including the availability and performance of intermittent and energy-limited resources, transmission interface constraints, relationships between transmission and generation facilities, and analysis of various case scenarios that examine impacts of changes due to present and future generation, load growth and potential transmission development. These phases are currently intended to include performance of both the Preliminary Study and the Final Study. The Preliminary Study, which will focus on the year 2010, is intended to highlight the data sources to which the PRM Study is most sensitive and the areas where more work in refining data can yield the greatest impact. The Final Study is intended to analyze the PRM requirements for the years 2010, 2014 and 2018.

A preliminary phase of the PRM Study, which is being conducted by the Working Group, is already underway in advance of formal workshops. The purpose of this preliminary phase is to develop an understanding of the sensitivity of the GE Energy PRM model's results to input data, and to learn the current capabilities of the model itself and the experience of the GE Energy modeling team with the topics considered important in a PRM analysis in California. This effort is intended to allow the formal CPUC workshop process to focus on issues that are most important in the PRM Study.

Future phases of the PRM Study intend to apply a similar probabilistic methodology as applied in phases one and two to evaluate the transmission-constrained local areas' reliability requirements. It should be noted that a probabilistic analysis as well as a seasonal approach to assess local requirements have been previously raised in the RA forum. Coordination with the RA proceeding will therefore be important if and when these issues are addressed in future phases of the PRM Study.

Various parties have suggested modifications to the scope, timing and technical model approach described in the CAISO's PRM Study Scope and Work Plan document and in the Commission's preliminary Scoping Memo. This report does not attempt to resolve these proposed modifications, since the Commission will address such modifications in its final Scoping Memo. Rather, this report attempts to provide a means to identify key issues for discussion in this proceeding, and move the PRM Study forward constructively while incorporating a diversity of stakeholder viewpoints.

2.2. Status

What follows is a tentative schedule for Phase I of R.08-04012. It is slightly different than the schedule distributed to the stakeholders at the Prehearing Conference on June 2, 2008.

TABLE 2.2
TENTATIVE SCHEDULE FOR PRM PHASE I

The Working Group is in the process of assembling input data for the Preliminary Study scheduled to be completed in summer 2008. As noted above, the Preliminary Study focuses on the year 2010. The schedule above is contingent on data delivery being accomplished by early in July, and that the data requires little reformatting or correction. The Working Group plans to later complete the input data and analysis for 2014 and 2018 in the Final Study.

The Working Group has not yet addressed important issues associated with resource plans for 2014 and 2018, such as data inputs and modeling of alternative resource portfolios that can change significantly in future years, or the uncertainty associated with the timely development of new resources that could affect PRM values. The Working Group expects to discuss such matters while the Preliminary Study is being conducted by GE Energy.

3. STUDY METHODOLOGY

3.1. Overview

This section explains the methodology that the Working Group proposes to use for the Preliminary Study to determine the PRM for the entire CAISO Controlled Grid, and for each of the three intra-regional areas.⁶

The required PRM would be calculated based on a specified reliability metric. Some parties have noted that the reliability metric most frequently used in the industry is a 1 day in 10 years, or 0.1 days per year, LOLE.⁷ In order to determine the required PRM, resource capacity is added or removed until the system achieves the desired reliability level or maximum LOLE. It should be noted that the PRM, as applied in the CPUC's current RA Program, is a percent of excess capacity above the monthly peak load, not the annual peak load. For example, assume that an area's annual peak load is 20,000 megawatts (MW) but the monthly peak drops to 15,000 MW for the next month. If the PRM is set at 15% then the area will need 23,000 MW of capacity in the peak month (= 1.15 * 20,000) but only 17,250 MW of capacity in the second month ( = 1.15 * 15,000). This means that the area will have 3,000 MW of reserves in the peak month but only 2,250 MW of reserves in the second month. This may lead to all months contributing to the annual system risk.

This is in contrast to systems like the New York ISO (NYISO) where the annual peak load establishes the amount of capacity that must be obtained year-round. In the example provided above, such systems would need 23,000 MW of capacity in both months resulting in 3,000 MW of reserves in the first month and 8,000 MW of reserves in the second month. As a result, the second month would likely not contribute to the annual risk.

Either method for determining the PRM is perfectly valid, although some members of the Working Group have suggested that the current method used in the CPUC's RA Program may result in a higher PRM requiring more capacity to be purchased at time of annual peak but often less capacity to be needed at other times of the year. The key point is that care needs to be taken when comparing a PRM in California to PRMs calculated in NYISO or MISO.

3.2. Introduction to GE Energy's MARS Model

GE Energy will use its model known as the Multi-Area Reliability Simulation Software (MARS) to calculate the daily LOLE for different PRM levels, thus determining the PRM required to maintain the system at a target reliability level or levels. In addition to the daily LOLE, MARS can also calculate hourly LOLE (hours per year) and expected unserved energy (EUE) in megawatt hours (MWh) per year. The reliability indices can be calculated for each of the sub-areas, and the CAISO control area, on both an isolated (assuming no ties between areas) and interconnected basis, with weekly and monthly indices also available.

Frequency of outages (outages per year) and duration of outages (hours per outage) can also be computed in MARS. However, these indices can only be determined if transition rates are provided for the generating units. These indices are not available when using just the forced outage rates, as is currently planned.

The traditional calculation of daily LOLE considers only the peak hour of the day and whether there is sufficient capacity to serve the load at the time of the daily peak. For a multi-area system in which the areas may peak at different hours of the day, MARS offers several options as to which hour should be considered. The first option allows the user to identify one area or pool to use for determining the one hour of the day for doing the calculations for all of the areas. The second option calculates the LOLE for each area at the hour of its own daily peak.

The third option does the calculations for all of the hours of the day. If the area is deficient for any of the hours of the day, even if not the hour of daily peak, the area will be counted as deficient for the day. This method recognizes the variable nature of some types of resources along with the fact that unit outages, and thus capacity shortage, may occur at times other than the peak hour.

Daily LOLEs calculated at different margin levels are particularly helpful to compare with the CAISO's emergency operating procedures (i.e., Stages 1, 2 and 3).

MARS uses a sequential Monte Carlo simulation to calculate the reliability of a generation system that is made up of a number of interconnected areas. The areas are defined based on the limiting interfaces within the transmission system. Generating units and a chronological hourly load profile are assigned to each area.

Thermal resources are modeled in terms of their capacity and planned and forced outage parameters. Wind and solar are modeled using hourly profiles to reflect their non-dispatchable nature. For each energy-limited unit (e.g., run-of-the-river hydro. or demand response), minimum and maximum ratings are specified along with the monthly available energy. The minimum portion, representing the run-of-river, is scheduled across all hours of the month. The remaining capacity and energy are scheduled during those hours that the thermal resources are not sufficient to meet the load.

MARS performs a chronological hourly simulation of the system, comparing the hourly load in each area to the total available generation in the area, which has been adjusted for planned maintenance and randomly occurring forced outages. If an area's available generation is less than its load, the program will attempt to deliver assistance from areas that have a surplus that hour, subject to the transfer limits between the areas. If the assistance is not available or it cannot be delivered to the deficient area, the area will be considered to be in a loss-of-load state for that hour, and the statistics required to compute the reliability indices will be collected. This process is repeated for all of the hours in the year. The year is then simulated with different random forced outages on the generating units and transmission interfaces until the simulation has converged.

The effects of uncertainty in the peak load forecast can be modeled by specifying up to ten per unit multipliers on the peak load and the associated probabilities of each load level occurring. MARS will compute the reliability indices at each of the load levels and calculate weighted-average values based on the input probabilities.

The reliability calculations in MARS are done at the area level - the generating units are assigned to areas, the hourly load profiles are defined by area, and the interface transfer limits are modeled between areas. From the area results, indices for pools or other area groupings can also be computed. For reporting pool results, if one or more of the areas in a given pool or area grouping are deficient in an hour, then the pool is considered as being deficient. Attachment 2 includes a brief description of MARS.

3.3. Description of CAISO Revised Study Scope and Work Plan

The CAISO has revised the original Study Scope and Work Plan document posted on the CAISO website. This document will be revised further to incorporate the feedback of the Working Group and the current stakeholder process in preparation for the Preliminary Study. Before the Working Group began to meet, the CAISO had proposed a Base Case and set of sensitivities that are described in more detail below.

The CAISO's proposed Base Case is meant to determine the PRM level required to meet a 1 in 10 LOLE. If there are insufficient resources, then generic peaking capacity (i.e., combustion turbine) will be added in 50 MW increments until the 1 day in 10 years (or 0.1 days per year) LOLE is reached. The Study Scope includes a number of proposed sensitivities that are described below:

· Future Generation Capacity. This sensitivity will show the impact on the PRM of installing larger units in the form of 500 MW combined-cycle units to meet 0.1 days per year. The variability of intermittent renewable resources will also be simulated by using other wind and solar load shapes included as another sensitivity to meet the target LOLE.

· Load Shape. The hourly load profile for the Base Case will use a historical load shape to model normal summer heat conditions. The effects of extreme summer heat conditions will be simulated using the historical hourly load data from a year representative of those conditions.

· Drought. This sensitivity will show the impact on the PRM of water shortages throughout the CAISO Controlled Grid. The generating conditions will be adjusted to simulate drought conditions through the study period.

· Variation in Generator Forced Outage Rate. The forced outage rate data of the thermal units will be adjusted to show the impact on the PRM of increases in forced outage rates.

· Maintaining CAISO at 0.1 days per year. This sensitivity will determine the impact on the PRM of maintaining the CAISO, rather than the individual areas, at an LOLE of 0.1 days per year. The reserves in the sub-areas will be adjusted to provide the resultant LOLE of 0.1 days per year for the CAISO.

· Maintaining Areas at Different LOLE. This sensitivity will show the impact on the PRM of varying the LOLE target for the areas from the base value of 0.1 days per year. Specifically, the PRM required to maintain the areas at 0.2 days per year and 0.05 days per year will be determined.

· Imports from Outside Areas. In the Base Case, imports into the CAISO from outside areas are calculated based on firm import assumptions, which may be procured by the IOUs as part of their Long Term Procurement Plans (LTPPs). Sensitivity studies on imports include (1) the transmission path (for flow gate) capabilities; (2) historical intertie flow data; and (3) imports estimated using production cost simulation for future years. The CAISO will provide the data inputs for the sensitivities of imports on transmission path capabilities, historical intertie flow data and imports simulated by production cost runs. The IOUs are responsible to provide assumptions on firm imports procured as part of their LTPPs.

To provide parties with an understanding of the impact of various input elements on the PRM Study results, the Working Group recommends that (1) an initial set of cases be completed for 2010 (Preliminary Phase 1a) for discussion in a stakeholder workshop and (2) after better input data is available (i.e., potential unit specific outage data), a full Preliminary Phase 1b be completed by the end of 2008. Table 3.3 below lists the potential number of runs for the proposed Phases 1a and 1b evaluations of the PRM.

TABLE 3.3
PRELIMINARY STUDY'S BASE CASE AND SENSITIVITY CASES

As the PRM Study progresses, the stakeholder group will review the initial results of the Preliminary Study to see if certain planned sensitivities should be replaced by others not presently under consideration, which may provide additional insights and enhance the overall PRM Study.

3.4. Expected Results

For the Base Case and each of the sensitivity cases considered, MARS will be used to compute the daily LOLE for each sub-area as well as the entire CAISO over a range of installed reserve margins, from which the PRM can be determined. An example of the possible results to be generated is shown in the figure below. The figure shows that in Case 1 a 14 percent PRM is sufficient to achieve a 0.1 days per year daily LOLE, while in Case 2, which contains a different mix of generation, a 15.5 percent PRM is needed to achieve that LOLE.

FIGURE 3.4

The sensitivity results will show the impact on the PRM of changes to various key input assumptions. Using the base and sensitivity case results as a starting point, combined with other factors discussed in this proceeding, the CPUC will determine the final desired PRM requirement.

4. STATUS OF WORKING GROUP DISCUSSIONS ABOUT INPUT ASSUMPTIONS AND MODELING ISSUES

This section describes the issues and recommendations of the Working Group to date. It is divided into five sections: (1) general issues, (2) intermittent resource modeling, (3) load and demand response inputs to MARS simulation, (4) modeling of current generation, and (5) modeling of transmission limitations and imports. These sections are meant to spur discussion at the PRM workshops on June 25 and 26.

4.1. Working Subgroup 1 - General Issues

The Working Group has identified four general issues: (1) whether the Commission should use a 0.1 days per year LOLE reliability standard, or other metrics in determining the final recommended PRM, (2) whether the Commission should adopt one CAISO-wide PRM for each forecast year, or multiple PRMs (one for each CAISO sub-area), (3) at what operating reserve level should involuntary curtailments be measured to compare against the maximum desired LOLE, and (4) how best to share the input data within the Working Group and ultimately with the public.

4.1.1. Issue #1: Should the Commission use the 0.1 days per year LOLE standard or a different metric in determining the final PRM requirement?

It is generally regarded that the 1 day in 10 year daily LOLE is the reliability standard most frequently used in the industry. Parties have also suggested considering other reliability metrics including LOLEs of 1 day in 5 years (or 0.2 days per year) and 1 day in 20 years (or 0.05 days per year), EUE, and metrics based on Value of Loss Load (VOLL).

Recommendation: The Working Group recommends that the PRMs needed for each sub-area and the entire CAISO control area to meet LOLEs of 1 day in 5 years, 1 day in 10 years, and 1 day in 20 years be provided as part of the Preliminary Phase 1b Study results. The Working Group also recommends that estimated EUEs for each of these PRM levels also be included in the results. These statistics should also be reported on a month by month basis. Parties will then be able to do their own benefit-cost analysis of multiple reliability metrics, and brief the Commission about their positions on this issue. The Commission will make the final decision on which metrics to use to determine the PRM requirement.

4.1.2. Issue #2: Should the Commission adopt one CAISO-wide PRM or one PRM for each CAISO sub-area?

Because of transmission constraints, and differences in load and resource characteristics between sub-areas within the CAISO, it is possible that the sub-areas may have different PRMs if the same LOLE level is maintained for each area.⁸ Vice versa, it is possible that each sub-area, if holding the same PRM, may have different corresponding reliability levels or LOLEs.

An issue the Commission needs to address is whether there should be: (1) a single PRM for the entire CAISO control area, or (2) different PRMs for each CAISO sub-area. This is a policy issue of course, but one that can best be addressed by reliability modeling that takes inter-area transmission constraints into consideration.

The Working Group supports approaching this issue by modeling two alternatives. For the first alternative, to determine the single PRM for the entire CAISO area, capacity will be added to or removed from the sub-areas so that they all have the same level of installed reserves and the CAISO reliability target of an LOLE of 0.1 days per year has been met. With the CAISO at the target LOLE, each of the sub-areas may be at different levels of reliability, but they will all be at or below the target LOLE. For example, even if all areas have the same PRM and the CAISO has a 0.1 days per year LOLE, Area A could be at 0.08 days per year, Area B at 0.03 days per year, and Area C at 0.02 days per year. Because of the overlap in outage days between the sub-areas, the CAISO would have an LOLE of 0.1 days per year, but Area A would be the major contributor to the overall CAISO risk. In general, the CAISO LOLE will not be lower than that of its least reliable sub-area (which indicates complete overlap of the sub-area outages), and will not be greater than the sum of the sub-area LOLEs (which indicates no overlap between the sub-area outages.)

In the second alternative, to determine the individual CAISO sub-area's PRMs, resources are added to or removed from each sub-area until each sub-area achieves the target LOLE. In this alternative, with all service CAISO sub-areas meeting the specified LOLE of 0.1 days per year, the CAISO will have a lower reliability (or higher LOLE) than in the first alternative. For example, if each sub-area is at 0.1 days/year, then the CAISO could be at 0.25 days/year.

Recommendation: The Working Group recommends that the PRM Study determine: (1) the PRM that each CAISO sub-area needs to meet the 0.1 days per year LOLE, and (2) the PRM that the entire CAISO-wide area needs to have the same the 0.1 days per year LOLE. An additional comparison between a single CAISO-wide PRM versus sub-area specific PRMs can be done using other LOLE metrics, which can be calculated by MARS. The Commission will then need to decide whether a single PRM or multiple PRMs should be adopted.

4.1.3. Issue #3: At what operating reserve level should involuntary curtailment be measured to determine the desired PRM?

For each PRM level, MARS will determine the LOLE, or probability of involuntary curtailments, considering load and resource uncertainties and interface transfer limits. Currently, the CAISO initiates a Stage 3 event, and begins to curtail firm load, when its spinning reserves are anticipated to drop below the WECC minimum operating reliability criteria (MORC) requirement.⁹ This typically occurs when CAISO's operating reserves are between 1.5 and 3 percent.

Because loss of load starts before the CAISO runs out of resources, the Working Group discussed different operating reserve levels at which to measure the LOLE. Four alternatives were identified:

(1) At zero operating reserves (i.e., when load equals resources),

(2) At 1.5 percent operating reserves,

(3) At 3 percent operating reserves, and

(4) At 6 percent operating reserves.¹⁰

In each case, MARS will determine the margin (defined as capacity minus load) for each of the sub-areas each hour and will count in the daily LOLE calculation those days in which the margin, after accounting for assistance from others sub-areas and Demand Response (DR), is negative at some time during the day. The only difference for the four alternatives is the amount of operating reserves the CAISO has when starting firm load curtailments.

Recommendation: The Working Group recommends that GE Energy measure the daily LOLE at: (1) 0 percent operating reserves, (2) 1.5 percent operating reserves, and (3) 3 percent operating reserves. Parties can recommend to the Commission at what point (or operating reserve level) involuntary curtailments should be measured to compare against the desired maximum LOLE and why.

4.1.4. Issue #4: What is the best mechanism to share the input data within the Working Group and ultimately with the public?

Some data inputs are market sensitive and considered confidential, such as generation maintenance schedule, specific generator unit's forced outages, contract imports and specific transmission facility outage information. Multiple agencies will provide this data, including the CPUC, the CAISO, NERC, and the IOUs/LSEs. Each agency operates under a different confidentiality mechanism, and in some cases, an agency's confidentiality responsibilities may preclude data sharing with others.

4.2. Working Subgroup 2 - Intermittent Resource Modeling

4.2.1. Issue #5: How to model intermittent resources in MARS to properly reflect the correlation of wind and load?

Because of the large role that intermittent resources are likely to play in meeting the State's Renewable Portfolio Standard (RPS) goals, it is important to represent these resources accurately in the PRM Study. As explained above, MARS is a Monte Carlo based program that randomly removes thermal generation to simulate outages based on its forced outage rates and mean time to repair. Each study year is simulated multiple times until convergence is reached on the index of interest. Hourly load profiles are specified for the year using a deterministic hourly profile, typically using a historical hourly load. To maintain the relationship between the weather and load and intermittent resource output, intermittent generation is modeled using a deterministic hourly profile corresponding to the resource's historical generation profile for the same years used to model load profile. That is, if the historic 2006 hourly load is used as the profile for a future year's load, MARS will use the historic 2006 wind and solar generation profiles. ¹¹

Since both the Preliminary and Final Studies are focused on future years, GE Energy asks for wind data that reflects "what would have been" produced in the relevant future years given expected increases in wind capacity. Such estimates have been made in the past for weather data from 2002-2004 as part of a CEC study (Intermittency Analysis Project) that also included GE and AWS TrueWind as consultants. However, no estimates of this type have been produced for more recent years. Because of the "staleness" of the load data in these years, the Working Group considers these 2002-2004 estimates to be inappropriate for use in this study.

The Working Group considered using estimated (rather than actual) hourly intermittent resource profiles based on historic weather information to project the expected increase in wind generation in future years. However, because of concerns about not capturing the actual correlation with the weather pattern that contributes to load, the Working Group prefers using actual generation data only. As is done today to calculate intermittent resources' net qualifying capacity (NQC), historical data for existing wind or solar generation would initially be used, and over time as new units are added, their actual performance would be used to update the future PRM requirement. For 2010, the historical wind profile can be scaled up to account for new capacity. However, for future years with large additions of wind capacity, simply scaling up the historical profile may not be appropriate as new wind resources will be located in different areas and new technology is expected to have improved performance.

Recommendation: For the Preliminary Study, the Working Group proposes to represent future load using the 2006 and 2007 hourly load profiles scaled up to the forecasted peaks for 2010. To ensure that load and intermittent generation are appropriately correlated, the Working Group proposes to use actual 2006 and 2007 wind and solar generation from the various wind and solar regions within the CAISO. The scale up factor will be calculated for each wind zone as a ratio of installed capacity (MW nameplate) in 2006 and 2007 compared to the expected installed capacity in 2010 based on the current CAISO interconnection queue. The wind zone scale up factors will be aggregated to match the modeling zones as needed.

For the later analyses covering future periods (i.e., 2014 and 2018), the Working Group prefers a more sophisticated approach, but has not yet reached a recommendation on this issue.

4.2.2. Issue #6: Modeling of Hydro Generation

Hydro resources play an important role in maintaining grid reliability in California. The Working Group considers hydro generation in two categories: (1) dispatchable, energy-limited resources, and (2) non-dispatchable, run of the river resources.

Recommendation: The Working Group proposes to model dispatchable hydro resources as energy-limited resources with monthly maximum and minimum capacities and an amount of energy available to the model to meet load. The minimum and maximum capacities for the Preliminary Study will be based on a 1 in 2 dry hydro year using values from the CAISO's generator database. The Working Group notes that there is very little difference in minimum and maximum capacities between a 1 in 2 year and a 1 in 5 or drier year. The monthly energy available will be the average monthly output using CEC/EIA 906 data from 1994-2005.

For run-of-the-river resources, the Working Group proposes to use a historical deterministic profile corresponding to the same years used to profile load and intermittent resources. SCE and PG&E are willing to provide the hourly generation for those years to the CAISO by the end of June for these units.

The PRM Study will also perform a sensitivity analysis using drought conditions for monthly available energy. The Working Group proposes to illustrate the drought values with the CEC/EIA 906 data and will employ 1992 values which are about 53 percent of average and are the lowest values between 1982 and 2007.

4.3. Working Subgroup 3 - Load and Demand Response Inputs to MARS Simulation

As noted earlier in this report, in order to better understand the sensitivity of the MARS model to various input assumptions and data, a Preliminary Study will be undertaken to understand where resources might be most productively deployed for the main body of analysis. Various elements of the load forecast, hourly shapes, and characterization of uncertainty about load forecasts will follow this general pattern. Table 4.3-1 (Attachment 1) summarizes the current Working Group's proposal for each phase of the analysis. Individual issues are also discussed below.

4.3.1. Issue #7: Load Forecast

The Working Group proposes to use the most recent load forecast from the CEC.¹² Load forecasts for each of the three utilities' transmission areas (IOU distribution service area plus associated municipal utility service areas embedded in the IOU distribution service area) are available. Since MARS uses 8760 hourly loads, the load forecast itself is used to scale a historic load shape to a future year's forecasted demand to reflect economic and demographic expansion, changes in electricity equipment holdings, customer response to price changes (if any), benchmark temperature conditions, and other factors normally included in load forecasts.

Different factors are needed to scale the CEC's 1 in 2 average weather load forecast to more extreme weather scenarios. The CEC's 2007 IEPR developed scaling factors for different weather conditions for the annual peak demand. Further analysis will be required to devise comparable scaling factors for the other months of the year.¹³

4.3.2. Issue #8: Modeling Uncertainty of Load

MARS requires hourly loads for each geographic area and year to be studied. For the Preliminary Study, this means hourly loads for each of the three CAISO sub-areas previously identified for the year 2010. A more refined Final Study for year 2010 and for years 2014 and 2018 will follow in subsequent phases.

GE reports that MARS typically uses a historical hourly load profile for an entire year scaled up to forecast levels. MARS users typically select load profiles from particular years to represent typical and extreme (hot) weather conditions. The Working Group members have reviewed the load patterns over the past five years and do not believe that a single year can be used to represent load. The extreme peaks in July 2006 certainly are representative of annual extremes for all IOU transmission areas as well as the CAISO as a whole. However, 2006 was not actually the extreme in other months. Further, while 2007 seems to be a valid choice for typical summer peak loads for PG&E and SDG&E, for SCE the summer peaks were not high enough to be considered typical.¹⁴

In order to fully capture the uncertainty in loads, the load data must address both load uncertainty due to weather, as well as load uncertainty due to non-weather factors such as underlying economic and demographic conditions, and forecast model specification error. For the Preliminary Study, the Working Group recommends that only weather uncertainty be considered. For the revised 2010 analysis, and certainly for the 2014 and 2018 analyses, additional sources of uncertainty will be considered.

There are two alternatives to model load forecast uncertainty using MARS. For the first alternative, load is specified using a single 8760 hourly load profile/shape which is scaled to up to ten annual peak load variations and the corresponding probability of occurrence for each peak load level. Under this approach, the entire 8760 hourly load is scaled by the same factor used for the peak hour (either monthly or annually). For example, in a simple annual peak approach, if the 1 in 20 year temperature peak is 10% higher than the 1 in 2 peak, all 8760 hourly loads are increased by 10% to represent a 1 in 20 load variation. The reliability calculations are done at each of the specified load levels, and the weighted-average values of the reliability indices are then computed from the indices by load level and their probabilities. To implement this alternative for the Preliminary Study, MARS could use the 2007 historical hourly load shape scaled to the latest CEC 1 in 2 load forecast for 2010. A sensitivity analysis will be run to determine the sensitivity of the results to the load shape by using a 2006 historic hourly as the basis to scale up to the 2010 CEC 1 in 2 load forecast.

For the second alternative, load uncertainty is represented by a number of 8760 hourly load scenarios (say five) and the corresponding probability of occurrence for each scenario. Under this approach, MARS is run multiple times, once for each scenario. Each scenario is characterized by a separate 8760 hourly load representing, for example, a 1 in 2 load, a 1 in 5 load, a 1 in 10 load, a 1 in 20 load, and a 1 in 100 scenario. To implement this alternative, rather than scaling all 8760 hourly load points by a factor that adjusts from observed weather to forecasted hot weather, the adjustment reflecting typical annual load forecast and the shape itself already includes weather extremes. Reliability indices are calculated for each load scenario, and a final set of weighted-average reliability indices is then computed after the fact from the individual load scenario indices and their corresponding probabilities.

To build 8760 hourly load scenarios for the second alternative, segments from different years are spliced together to create synthetic 8760 shapes. For example, load scenarios can be constructed by combining historic hourly loads from 2007 with portions of 2006 heat storm days or weeks to create a complete 8760 set that matches the CEC's load forecast for 2010 for a 1 in 2 load, or 1 in 5 load, etc.

The example below illustrates the first alternative. The example uses a seven point normal distribution to describe the peak load forecast uncertainty. The input annual peak of 10,000 MW has a 38.30 percent probability of occurring. A peak 300 MW above or below the forecast value occurs with a probability of 24.17 percent. Moving to the tails of the curve, a load that differs from the forecast by 1,000 MW has a probability of 0.62 percent.

FIGURE 4.3-1

MARS calculates the LOLE at each of the specified load levels, with the results shown on the red curve. At a peak of as high as 9,700 MW, the LOLE would be 0 days/year, while at a peak of 11,000 MW, it would be 5 days/year. At the forecast peak of 10,000 MW, the LOLE would be only 0.02 days/year. The expected value as calculated from the seven points would be 0.2 days/year. This demonstrates that it is often the higher load levels with their relatively low probabilities of occurrence but large LOLEs that contribute the most to the expected LOLE.

The load forecast uncertainty case can also be viewed in terms of different load forecasts, such as a 1 in 2 annual recurrence interval or 1 in 5 annual recurrence interval, etc. MARS models the load forecast uncertainty at discrete load levels rather than as a continuous function. An approach for defining the discrete load levels would be to treat all loads below a certain level as being equal to that level. So, all loads less than the 1-in-2 forecast level would be modeled at the 1-in-2 level with a 50 percent probability of occurring. Likewise, all loads between the 1-in-2 and 1-in-5 forecast levels would be modeled at the 1-in-5 level with a 30 percent probability. The figure below shows the probabilities that would correspond to different load forecasts using this approach.

FIGURE 4.3-2

Importantly, this discussion of the treatment of load uncertainty highlights issues with the distribution of load uncertainty based on the highest and lowest specified load scenarios. Figure 4.3-1 indicates that modeling load in the higher end of the distribution is more important than at the low end of the distribution because loads at or below the expected value have very little impact on the LOLE estimates while load in the higher end of the distribution have significant impact.

The ramification of this modeling feature is that the results of the model simulations of LOLE, especially at the aggregated level, need to be carefully characterized and understood within the context they were developed. The Working Group recommends that the study result statistics (monthly LOLP, LOLE, EUE, etc.) be shown not only at the aggregated level of expected loads, but that the results also are shown at each discretely specified level of load. This will allow for greater understanding of the results and enable users of the study results to better characterize the nature of the results and their applicability to policy choices.

Recommendation: For the Preliminary Study, the Working Group believes that the 8760 hourly load from 2007 should be chosen to scale up to the 2010 applicable 1 in 2 forecast load. Monthly peak demand variations can be provided for each month to represent load uncertainty at the 1 in 5, 1 in 10 and 1 in 20 recurrence interval levels, and ideally extended to the 1 in 50 and 1 in 100 levels. Also, the 8760 hourly load from 2006 can be used as a sensitivity case to address the very extreme peak loads observed in July 2006.

For the more refined 2010 analysis, or the 2014 and 2018 analyses, the Working Group has not reached consensus about how to represent load uncertainty. The Working Group proposes to discuss the two alternatives identified above with all stakeholders at the June 25 and 26 workshops.

4.3.3. Issue #9: Adjusting Historic Load Shapes for Demand Response

Accurately representing customer demand and its hourly shape are important inputs to MARS calculation of LOLE. When DR programs are triggered, and load is reduced, the system is using load reduction as a resource. Any such load reductions need to be "added back" to the recorded loads to fully characterize customer load. Similarly, if customers have involuntarily lost service, due to for example distribution outages, then recorded demand is lower than it actually is absent involuntary outages or DR. While these two phenomena could happen in any year, they were especially important in summer 2006. The extreme loads recorded in July 2006 actually under count customer demand, since some DR programs were triggered and some distribution outages took place. The IOUs will estimate the aggregate hourly adjustments from these two sources and modify recorded hourly loads to increase them to fully represent customer demand.

Recommendation: The IOUs will estimate the aggregate hourly adjustments from these two sources and modify recorded hourly loads to increase them to fully represent customer demand.

4.3.4. Issue #10: Modeling of Demand Response

MARS models DR resources as energy limited resources. Each individual program is modeled with a minimum output (which can be zero), a maximum output, and a certain amount of available energy per month. The total annual energy needs to be allocated to months when preparing the inputs to MARS because the model does not optimize the hours of operation.

Recommendation: For the Preliminary Study, the IOUs' DR projections for 2009 through 2011 can be utilized as they were recently filed in their respective Applications with the CPUC. These will be translated into the stylized format for the MARS program.

4.3.5. Summary

Table 4.3-1 summarizes the overall pattern of successive development of preliminary 2010, refined 2010, and 2014-2018 demand forecasts and related input assumptions to MARS. The purpose of the Preliminary Study is to understand the sensitivity of the model to alternative inputs. Presumably, this will foster discussion as to where greater resources should be devoted for improved analysis and greater data precision. Because the modeling of extreme load events is of such critical importance to the accuracy of the LOLE, the Working Group looks to GE Energy to provide study results in a format which highlight not only the aggregated probability weighted results but also the results for each level of specified load.

It is also important that more sophisticated approaches to creating typical load shapes be coordinated with wind resource characterization, since wind and load are tightly coupled in the MARS studies conducted to date.

4.4. Working Subgroup 4 - Modeling of Current Generation

4.4.1. Generation Forced Outage Data

MARS generation inputs include forced outages, scheduled or planned outages, unit rating, and capacity states. GE Energy will use five historical years of generation forced outage data as inputs to MARS.

4.4.2. Preferred Data Source

The PRM Study will use the North-American Electric Reliability Corporation's (NERC) Generating Asset Database System (GADS). GADS is a national database of availability and outage data for generators. NERC's proprietary software PC-GAR calculates performance indices from this data. To measure forced outages, Equivalent Forced Outage Rate of Demand (EFORd) calculates the percentage of time that a unit is out of service when there is demand for it to produce power.

NERC provides a comprehensive source of complex and standardized availability and reliability measures, which the New England Power Pool (NEPOOL) and the PJM Interconnection have used in their probabilistic resource planning models. The CAISO is an alternative source of raw outage and availability data from its SLIC system (Scheduling and Logging for the ISO), but does not possess EFORd data.

4.4.3. Confidentiality of Individual Plant Data

The CPUC requires generators to submit GADs data to NERC under General Order (G.O.) 167, and to authorize NERC to release this data to the CPUC. NERC's policy prohibits the release of confidential plant data without a generator's consent. In order for the CPUC to obtain data from individual generators, each participating California generator provided written permission allowing NERC to release GADs data to the CPUC. Therefore, the CPUC's access to the GADS data was made subject to written release of the individual Generating Asset Owners (referred to as GAOs in G.O. 167). GAOs then requested confidential treatment by the CPUC, pursuant to provisions in G.O. 167.

Recommendation: Given the confidential status of individual generators' data, the CAISO legal department should work with NERC and/or the CPUC to discuss a workable approach to access specific GADs data and/or alternative data sources, and follow any applicable confidentiality restrictions.

4.5. Working Subgroup 5 - Modeling of Transmission Limitations and Imports

4.5.1. Issue # 11: Modeling of Transfers Between Sub-Areas Within the CAISO Region

The Working Group looked at ways to model transfers between CAISO sub-areas that correspond roughly to NP 26 and SP 26. At a minimum, annual transfer capabilities as identified above need to be provided. This can be further decomposed, if necessary, by month or peak/non-peak periods. In discussions with the Working Group, GE Energy indicated that the transfer limits can be varied as a function of area load and availability/unavailability of generation. Furthermore, the IOUs have remote generation and/or contracts (out-of-state) that are located outside their respective territories.

Recommendation: The Working Group recommends that the following transfer capabilities should be modeled:

· PG&E to SCE and SCE to PG&E: The transfer limit will be based on the transfer capability across Path 26. One possible source for the Path 26 transfer limit is the RA counting constraint. Specifically, a transfer limit of 3,430 MW north-to-south and 2,583 MW south-to-north was established for purposes of RA counting, after considering Existing Transmission Contracts (ETCs) and loop flow.¹⁵

· SCE to SDG&E and SDG&E to SCE: The transfer limit will be based on the transfer capability taking into account the SDG&E Simultaneous Import Limit (SIL).

· Remotely owned or contracted resources will be modeled as being within the IOUs' respective areas for the Preliminary Study.

4.5.2. Issue # 12: Modeling of imports

The scope established by the CAISO for the PRM study to be prepared by GE Energy is to model transmission paths between the three IOUs, but not between the CAISO and the WECC. Therefore, the impact of imports will have to be done post GE-MARS simulation by assuming an additional amount of resources are available when calculating EUE. The alternative is to establish additional transmission pathways, but that would require modeling loads and resources outside the CAISO, which adds additional modeling complexity. A hybrid approach is to include an amount of imports that are available as a resource for each intertie with a forced outrage rate to represent the probability of a transmission line outage.

The Working Group discussed different approaches to represent the amount of imports in MARS LOLE calculations. The alternatives included: (1) assuming imports are available up to the import capability; (2) using historical import availability; (3) estimating import availability based on WECC production simulations; and (4) running the GE-MARS model without imports, and later estimating how much imports are required to meet a target of the load and PRM level.

The impact of imports is important because they are resources that can contribute to meeting load and each of the approaches listed above have trade-offs. Using an import assumption of transmission capability may overestimate the amount of resources that could be made available to the CAISO during a Stage 3 event. The use of historical import availability may be appropriate for 2010, but it may not forecast the appropriate amount in 2014 or 2018. Using a WECC production simulation would require the scope of the CAISO PRM project to be revised, along with the approved budget. The last option to add imports until the desired PRM is met could result in an amount of imports that may exceed what is actually available or capable due to transmission congestion during a Stage 3 event.

Recommendation: For the Preliminary Study, the Working Group recommends that the PRM be determined without imports into the CAISO, except for the remote generation described in Section 4.5.1, because imports can be used to meet load plus the required PRM.¹⁶

5. NEXT STEPS

The Working Group looks forward to receiving feedback and input from parties at the June 25 and 26, 2008 workshops scheduled by the Commission. The Working Group will discuss next steps at those workshops, and will continue to meet and work with GE Energy to conduct further work in some of the areas described above.

ATTACHMENT 1 - TABLE 4.3-1
OVERVIEW OF WORKING GROUP # 3 PROPOSALS FOR LOAD FORECAST INPUTS TO GE MARS MODEL

ATTACHMENT 2

GE's MARS PROGRAM DESCRIPTION

¹ Order Instituting Rulemaking (OIR), issued April 16, 2008 in R 08-04-012 at 17-18.

² CAISO Market Notice, March 5, 2008, < http://www.caiso.com/1f81/1f81c3b9128d0.html>.

³ The IOUs are Pacific Gas and Electric Company (PG&E), Southern California Edison Company (SCE), and San Diego Gas & Electric Company (SDG&E).

⁴ Notice of Prehearing Conference, issued May 14, 2008 in R.08-04-012.

⁵ OIR at 10-11.

⁶ The CAISO Controlled Grid represents the complete CAISO control Area. The CAISO's PRM Study Scope and Work Plan identified three CAISO sub-areas: (1) Area 1: Northern and Central California area north of Path 26 (NP 26) including the PG&E service area and participating publicly owned utilities in the NP 26 region served by the CAISO, (2) Area 2: Southern California south of Path 26 (SP 26) including the SCE service area and publicly owned utilities in the SP 26 region served by the CAISO, and (3) Area 3: Southern California - SDG&E's service area (primarily southern Orange County and San Diego area).

⁷ See summary of resource adequacy criteria used by reliability councils in the country prepared by the Western Electricity Coordinating Council (WECC) at http://www.wecc.biz/documents/library/RAGTF/RelCouncilRACriteria.pdf.

⁸ These areas are typically defined based on transmission constraints either by limits on major WECC paths or by other planning studies identifying further constraints on the local areas.

⁹ CAISO's Operating Procedure E-508, Version 4.7, at 16.

¹⁰ The CAISO does not curtail firm load at 6 percent operating reserves. Measuring LOLE at this level is useful to estimate the frequency of Stage 2 events, or when interruptible load is curtailed. It is also useful in determining how often the CAISO would violate NERC/WECC operating reserve standards and be liable for financial penalties.

¹¹ Analyses presented in the spring 2008 workshops of CPUC RA rulemaking R.08-01-025 support the contention that there is considerable negative correlation between load and wind generation in California, especially in the periods around summer monthly peak demand. This correlation is not pronounced in spring and winter months. Thus strict chronology between load and wind generation is important for a few system peak conditions, but not in most hours of the year.

¹² Service area load forecasts out to 2018 (annual energy and peak demand) were adopted by the CEC in November 2007 as part of the 2007 Integrated Energy Policy Report (IEPR) proceeding.

¹³ PG&E reports it has such factors for each month.

¹⁴ SCE notes that while using the 2007 load shape for the CAISO wide PRM and other utilities is reasonable, 2007 may not represent a typical year for SCE when calculating a SCE area PRM; instead, SCE proposes a 2004 load shape to represent its typical year shape.

¹⁵ D.07-06-029 (June 21, 2007), section 3.2.1, Path 26 Counting Constraint (Joint Parties).

¹⁶ This establishes an amount of planning reserves (regardless of location) to reach a desired reliability level. A review would be necessary determine the validity of amount of imports that would be available during a Stage 3 event.