IEEE Power & Energy Magazine - May/June 2020 - 62

Sample stratification seeks to divide a population into groups that
are similar within the individual segments exhibiting characteristics
related to the measure being estimated in the study.
The Load Research Lifecycle
Figure 1 shows the load research lifecycle. We start in the upper
left corner with a "need for information." Load researchers
should work with internal clients to develop a preliminary plan
that specifies the goals, objectives, and analytical approach for
fulfilling clients' needs.
Next, load researchers are responsible for securing or
creating an appropriate population frame. For classical load
research, this comes in the form of the population billing
data known for all customers on the system. The customers
are typically segmented by rate class or other domains of
interest. For some domains, such as customers with rooftop
solar or electric vehicles (EVs), the construct of the population can be more complex and may require combining
multiple internal and/or external sources of data to create an
appropriate framework for analysis.
After identifying the measures of customer usage to be
used as the focus of the research study and the populations
for which the measures need to be reported, load researchers develop an appropriate sample design, selecting the
minimum number of customers needed for estimating a
population-level measure constrained by a desired level of
precision of the estimate. Load researchers are responsible
for making decisions with respect to the stratification strategy for the sample. Sample stratification seeks to divide a
population into groups that are similar within the individual
segments exhibiting characteristics related to the measure
being estimated in the study. The greater the similarity of
the customers within the strata, the smaller the number of
customers needed to be sampled. The particular features of

Need for
Information

Model

Results

Preliminary
Plan

Load Research
Lifecycle

Expand

figure 1. The load research lifecycle.
62

ieee power & energy magazine

Population
Data

Sample
Data

stratification include identifying the variables to use, such as
seasonal and annual use, peak demand, and load factor, and
the key sample design planning variables of interest, such as
system and class peaks.
Next, decisions are made with respect to the confidence
level, desired relative precision, and anticipated variability
of the target variables. These decisions dictate the sample
size requirements. This often becomes an iterative process,
with tradeoffs necessitated by the available budget. Finally,
having settled on an appropriate sample design, the load
researchers guide the selection of the sample.
For some projects, such as an analysis of demand
response, technology assessments, and pricing experiments,
the sampling plans will need to be developed inside an
experimental design paradigm. The load researchers provide
guidance and governance on the appropriate experimental
design to use, potentially randomized control trial, withinsample comparison, test versus comparison, preparticipation
versus postparticipation, and so forth.
Historically, the next step has been to work with the
metering department to procure and install pulse-initiating meters with load recorders. This includes determining
an appropriate data collection strategy, such as handheld
devices, modems, or a fixed network. Due to the high cost
of equipment, equipment installation, and data collection,
load researchers had to rely on interval data collected from
a limited sample of customers in the population of interest.
However, in today's fast-approaching advanced metering
infrastructure (AMI) world, these constraints are fading
away, as we are seeing utilities deploying "super samples" or
even "census" representations.
As we transition to full AMI, sample data present several
unique challenges, such as storage and how to conduct validation, editing, and estimation (VEE). Added to the challenge of storage is the communication bandwidth needed
to transmit interval data from the field. Utilities often find
the need to compromise load interval granularity (5 min,
15 min, hourly, and daily) to ensure that they have the capacity to reliably transmit data while maintaining bandwidth
for other operational priorities. Data quality issues arise
from any number of sources, including communication
problems, database malfunctions, and human errors. These
issues can affect the data capture and its eventual analysis,
if not handled properly.
Some simple sanity checks such as ensuring consistency
with the correct time stamp, interval orientation (beginning or end of period), units of measure, and frequency
may/june 2020

IEEE Power & Energy Magazine - May/June 2020

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - May/June 2020