Summary
Introduction
The fundamental premise, that load profiling should be used for direct access customers, whenever RTP metering is not cost- effective, should not be an issue. As long as not all customers are served with RTP metering, the non-RTP customers' service will be based on load profiling for at least one supplier: the UDC. DRA is aware of persuasive vendor presentations regarding the cost-effectiveness of universal or almost universal RTP metering based on local area networks, however we cannot yet conclude that it would be cost-effective to install RTP meters for all customers. (Also, to our knowledge no California electric utility has yet proposed this metering development to the Commission). Even if California electric utilities were to pursue widespread RTP metering it is DRA's understanding that access to this metering would have to be phased in over several years. Thus, absent the load-profiling solution, the vast majority of customers would be prevented using direct access suppliers for generation services. This would violate the Commission's fundamental policy that competitive options and consumer choice must be available to all customers from the beginning of electric restructuring. Furthermore, use of load profiles is essential because various aggregators will have different mixes of customers from each other (and from the UDC, which for some important functions is essentially an aggregator). Ensuring that different aggregators' load responsibility reflects their customers' actual load, as accurately as possible, requires that identifiable differences in load patterns must be used to compile the aggregators' load forecasts and settlement requirements.
The benefits to be achieved through use of load profiling should not be confused by issues dealing with allocation of forecasting errors. There are two separate types of error that will occur: forecasting error and measurement error. Forecasting error will occur regardless of the accuracy of measurement, i.e., metering -- imbalances, i.e., deviations from load forecasts will occur for reasons like the weather being less than perfectly predictable. DRA has previously presented the results of hourly system load forecasting in GRC testimony on an interruptible rate option (a proposal by ACWA for 6-hour notifications of curtailments, which DRA analyzed by using an econometric analysis of recorded hourly system loads, i.e., negligible measurement error) and found that (1) using 6-hour- ahead forecasts required a significant reduction in non-firm credit, and (2) even 1-hour-ahead forecasts did not perfectly forecast the recorded system load.
On the issue of measurement error, load profiling is not a sole cause of error, because even if everyone had a RTP meter, there is a certain allowable level of mismeasurement in a retail billing meter (e.g., 1%). For a large population of customers served by an aggregator, the measurement error of individual meters will offset each other but will be replaced by statistical sampling error. Statistical sampling error can be reduced to any desired level by using an adequate sample size. It can be reduced even with a smaller sample, for purposes like avoiding cost-shifting between large and small customers, or between aggregators (including the UDC), by setting a known total for the sampled load (by subtracting the load on individual RTP meters from the total system load) and using the statistical sampling to allocate the residual.
Furthermore, a fundamental principle should be recognized that true error is random, and not biased in one direction. If analysis reveals that errors in a proposed methodology is biased in relation to an identified parameter, the methodology should be refined, rather than concluding that its purpose should not be pursued.
Instead, the focus of attention should be identifying what issues need to be resolved in order for load profiling to successfully meet the Commission's restructuring goals, and finding solutions to those issues. Some issues are addressed herein, including an outline discussed on 5/31/96, and others will arise in further discussion.
I. Load Profiling
1. Use of existing schedule-specific load profiles (LP)
Except for some rate schedules serving relatively small numbers of customers, there is a variety of customers on each schedule that is comparable to the mix of customers in broad customer classes. Limiting the use of load profiling to a single profile for each customer class would prevent us (collectively) from achieving a goal of the best possible estimate of the load responsibility of different suppliers (including the UDC), because different suppliers will have different mixes of customers. Instead, a failure to reflect identifiable differences in the load characteristics of various customers, served on the same UDC rate schedule, would result in cost shifting between suppliers. The amount of this cost- shifting would remain unknown without analysis based on load profiles for smaller sets of customers. Therefore, DRA opposes this option.
2. Development of LP based on Sample for Commercial Classifications
The utilities already use stratifictions within customer classes (with rate schedules with large numbers of customers spanning multiple strata), which exist without controversy. For example, PG&E stratifies each customer class with two to four variables per class, reflecting Baseline Territory, average daily kWh use, average monthly kW demand, rate schedule, and SIC code; pages DQS-9 to DQS-56 of PG&E's revenue allocation workpapers, Chapter 1, Appendix A, Part 2, "Customer Load", provide details. DRA's recommended approach to use of load profiling, within each customer class, is similar to today's use of load research data, with the exceptions that (1) load responsibility for each supplier would be determined through a common set of load research data, weighted by its own customer mix, (2) the load research sample may need to be augmented in some market segments, or to achieve greater accuracy overall, (3) there should be provision to identify new load research strata to meet emerging market needs, and (4) data analysis would need to be completed more promptly than for the use in rate cases today.
DRA recommends PG&E's general approach to stratification as a minimum for each utility after restructuring, but each utility's existing stratification could form a starting point for post- restructuring stratification, with possible limitations that would be identified; the purpose of such limitations would be to prevent aggregators from focussing on specific market segments before the statistical sample for these market segments can be augmented. (In an extreme example, PG&E has a stratum for Colleges on Schedule A-1, i.e., small commercial, that has one sample point.) Procedures should be developed to augment the existing load research sample in cases where this is an issue, and establish who should pay the cost of doing so. A related issue is that recorded load research data should be available to non-UDC analysts, with provisions for masking the identity of individual customers (as is done with economic Census data), so that parties who might wish to identify strata to be augmented can have an informed basis for doing so.
DRA supports the notion that a threshold should be set for the largest customers to be served using load profiling. One level that has been discussed is 500 kW, but there have been suggestions that this threshold might be as low as 20 kW. The question of the precise level should be left for future resolution once more information has been determined concerning meter standards and cost-effectiveness.
The principle goals of using load profiling as described above are to: (1) allow competitive service to reach all customers, even if RTP metering is not cost-effective, and (2) encourage accurate cost allocation (i.e., not cost-shifting) between suppliers. The criterion for load research sample size should be to determine the load responsibility of aggregators (and the UDC) to a comparable accuracy as direct access customers. Direct access customers' loads will be subject to measurement (metering) error, and aggregators of small customers would be subject to statistical sampling error instead, but the impact of these two types of errors can be made comparable. This use of load profiling does not require creation of new rate schedules. Aggregators will be free to differentiate between their customers to a great extent, using load profiling and/or other information, and UDCs (and their customers) can but are not required to seek similar options.
3. Development of LP based on Specific Companies
DRA participated in the discussion on 5/31/96 of a multiple- site company-specific stratification, for purposes of illustration. Other issues concerning company-specific strata convince DRA that there could be significant drawbacks to this approach. (In particular, the company could know the sites where a load research meter is present and then influence the sampled data to a greater extent than any actual change in its overall operations.) Therefore, DRA is not promoting this approach.
II. Reconciliation with Upstream Interval Meters
1. Single Supplier
On 5/31/96, UCAN discussed placement of RTP meters at central locations such as substations, for the purpose of calibrating the statistically-sampled population served through that metering point; UCAN did not specify that customers served through such central metering points would need to be served by the same supplier. DRA has not identified any benefits to requiring customers below central metering points to be served by the same supplier (although using central metering points can have benefits that do not require a single supplier), believes such a practice would contradict the intent of allowing customer choice to all customers, and therefore opposes such a requirement.
2. Multiple Suppliers
It would be consistent with DRA's recommendations, stated above, to use UCAN's suggestion of RTP meters at central metering points for the purpose of augmenting and calibrating the load research data used in load profiling. Care should be used in defining central metering "points", however, following discussions in PG&E's TY93 GRC that documented periodic transfers of distribution feeders from one substation to another. A suitable unit of data aggregation could be PG&E's Distribution Planning Areas (DPAs) or Transmission Planning Areas (TPAs); in Edison's TY95 GRC, DRA presented an area load analysis using existing data, which was adopted by the Commission, and the Commission reiterated its direction that Edison should develop area load data if it feels more detail is needed.
If other purposes of central metering points are identified by other parties, without requirements that customers below such points need to be served by a single supplier, DRA remains open to considering them.
Conclusion
Debate should be resolved now on a conclusion that load profiling will be necessary for some customers. Several issues exist for the appropriate procedures for its use, and these issues should be the focus of further discussion. The focus should be on establishing a workable initial set of principles, as well as a means for refinements as experience grows under the restructured markets. The initial set of principles could be based fundamentally on the utilities' existing load research data, although discussion should be open concerning augmenting it by 1998.