lign: center;">Curt D. Puckett

RLW Analytics

 

Table of Contents

 

List of Figures

 

Measuring the Impact of a Load Response Program

Written by: Dr. Roger L. Wright

President RLW Analytics

Presented by: Curt D. Puckett

Senior Vice President RLW Analytics

Summary

Measuring the load reduction
of a participant in an interruptible-load program is fundamentally different than
measuring the load generated by a participant in a supply-side program.  The primary challenge is to assess
the load that would have occurred in the absence of the call for curtailment
and compare this to the load that was actually observed.  This
is unavoidably a statistical modeling problem, not just a load measurement
exercise. 

There are three methods
of addressing the problem:

The participant-specific
approach Use the participant-level load data and a customer-specific
statistical model to assess the load response of each individual participant
on a particular curtailment day, and then aggregate the results.
The aggregate-load
approach Aggregate the participant-level load data for all
participants, and then use a single statistical model to assess the
aggregate load response of all participants taken together on a particular
curtailment day.
The continuous-forecasting
approach Use either of these approaches on a continuous basis
to forecast the expected aggregate load of all participants on each
day during the program period on non-curtailment days as well as curtailment
days.  Then estimate the impact of the program as the difference
between forecasted load and actual load on the curtailment day. 

Each approach has its advantages and disadvantages. The primary advantage
of the participant-specific
approach is that it provides results for each participating customer. 
These results may be needed to compensate the participants for their
curtailment.  The fact that these results are available for scrutiny
by the participants may encourage the Integrated System Operator (i.e.,
ISO) to accept them as a valid basis for assessing the impact of the
program.  However, if the ISO feels that it needs to review these
results, it may have to review many different customer-specific statistical
models.  Moreover it is difficult to assess the statistical accuracy
of the aggregate impact.

The aggregate-load
approach utilizes a single statistical model that may be easier
for the ISO to review.  Moreover, the model will usually be more
reliable than the customer-specific models and there are fairly straightforward
ways to assess the statistical accuracy of the aggregate results. 
However this approach does not provide customer-specific results. 
Moreover, the results are dependent on the validity of the statistical
model, and the validity of any model may be subject to debate.

The great advantage of the continuous forecasting approach is that the ISO or any interested
party can quite easily validate the accuracy of the forecasts by comparing
the forecasts of load to the actual load on the non-curtailment days. 
It is not necessary to examine the underlying statistical models or
to debate their validity.  The only additional check is to ensure
that the same forecasting methods are being used on the non-curtailment
and curtailment days.  We believe that this can be confirmed fairly
easily.  The primary disadvantage is that the continuous forecasting
approach might seem to add a burden since it requires each entity to
prepare load forecasts for every day rather than just the curtailment
days.  But in practice, the entity may already have the infrastructure
for doing this. 

The validity of each of these methods can be affected by the placement
of the interval recorder.  In order to obtain the best results
for each individual customer, it is important to carefully consider
the location of the interval recorder. Under the continuous forecasting
approach, the interval meter can be placed on the whole-customer load. 
This eliminates another area of subjectivity and potential gaming.

Considering all of these issues, it appears that the continuous-forecasting
approach provides the best, practical way of objectively assessing the
hourly load impact of the program. 

 

Assessing the Load Response of Individual Participants

Measuring the load reduction
of a participant in an interruptible-load program is fundamentally different than
measuring the load generated by a participant in a supply-side program. With the supply-side program, the question
is what actually happened, specifically, how much power was generated. 
In this case, it is generally straightforward to measure the hourly
energy that was generated by the participant. 

With
a demand-side program like load management, it is necessary to determine
the hourly energy that was not used
by the participant due to the program. In this case, the question is
what is the difference between what actually happened and what would
have happened in the absence of the call for curtailment.  Measuring
what actually happened is easy assuming that interval metering is in
place to measure the hourly load of each customer.  The task that
needs to be discussed is determining what would have happened on a given
curtailment day.

This
is essentially a statistical modeling problem.  We use the customers
observed load during days with no curtailment (non-curtailment days)
to predict the customers load during each curtailment period. 
If the customers load is very stable, there is little problem. 
For example, if the customer has the same pattern of use over all weekdays,
we can calculate the customers average weekday hourly load during the
non-curtailment days to predict the customers load during the curtailment
period. 

Applications

In 1999, Utility A ran
a load management program for large industrial customers. These customers
were asked to reduce their load on several days in 1999.  Table
1 lists a few of the dates and hours of interruption.  Interruption
by the participant was strictly voluntary.

Date

From

To

Day

June 7

10:00 AM

10:30 PM

Mon

June 8

8:06 AM

9:00 PM

Tue

June 28

10:55 AM

8:00 PM

Mon

June 29

10:02 AM

4:30 PM

Tue

Table 1: Sample Interruption Periods

Utility A collected fifteen-minute interval load data
for each of the participants.  To illustrate the issues discussed
in this paper, we will draw on a small sample of the participants. 
To preserve the confidentialit