IEEE Electrification - December 2022 - 52

Utilities now use an abundance of devices, from infrared
cameras to wind velocity sensors and lidar equipment.
Some have also installed hundreds of weather stations on
their facilities to have better situational awareness of their
systems and improve their forecasting capabilities.
BTM data are now a source of big data. As customers
become more conscious of their energy footprint, they
want to track and store the data from their onsite energy
resources. Also, an abundance of sensors enable customers
to track lighting levels, temperatures, air duct flow
rates, and room occupancies. This provides better visibility
of onsite energy usage. Nonintrusive systems that segregate
the patterns of different appliances and devices
are allowing consumers to track and understand their
electricity use.
BTM DERs and Their Impacts on the Grid
Distributed generation provides a way of supporting the
grid and operating in islanded mode during outages. DERs
have proven their value in severe weather events like Hurricanes
Sandy and Maria. With the cost of renewables consistently
decreasing, solar and wind are more attractive as
resources in microgrids and as supplementary energy
supplies for business campuses and for households.
As promising as they are, these technologies present
some problems. Solar farms do not have the black-start
capability needed when the grid goes down; they need
support from fossil generation (e.g., in California) to take
on the load when the sun sets and the panels stop production.
The distribution protection system is also
designed for unidirectional power flow. Another looming
problem is electric vehicle (EV) ownership and charging,
which will disrupt and tax the distribution system, potentially
doubling the capacity requirements.
TABLE 1. The monthly data volumes for
different devices.
Meter Traditional
Reads/
month
1
AMI
Meter
SCADA
PMU
2,880 1,296,000 77,760,000
At reads every 15 min  4 × 24 × 30 = 2,880.
At reads every 2 s 30 × 60 × 24 × 30 = 1,296,000.
At reads every 2 cycles  30 × 60 × 60 × 24 × 30 = 77,760,000.
This does not include the fact that AMI meters and PMUs read more data types.
Planning for DERs is also a challenge, especially with
BTM resources. Since they are BTM, utilities do not have
direct visibility into their output and only see the netmetered
consumption levels. They present an unknown
quantity of generation that is correlated with weather conditions.
At times when solar irradiance or wind speed is
low, consumers increase their consumption. Utilities and
transmission organizations need to provide backup
resources to meet demand and provide reliable power.
At the same time, solutions to these issues can be
achieved with better monitoring and management of BTM
resources. For instance, converting volumes of data into
information that enables better decision making will
improve utilities' customer care. The ability to separate load
and renewable generation will inform the distributor about
real customer requirements (i.e., in energy and quality),
which helps build a more reliable system. This type of progress
allows the utility to improve its operations and reduce
its costs, enabling it to make dispatch improvements for
peak reduction and loss minimization. With these opportunities
in mind, utilities can enhance the quality of the electricity
they deliver via proper device switching.
Analytics Can Help
Fortunately, advancements in analytic methodologies and
access to real-time data have grown in parallel with
developments in the electricity industry. Analytics uses
data and statistical/machine learning models to discover
relationships, predict unknown outcomes, and automate
decisions. A typical analytical process consists of data
extraction, data exploration and cleanup, data enrichment
(by merging relevant data sources together or adding
new variables derived from those existing), selecting
the best variables for modeling, model training, and
model assessment/comparison. Figure 2 shows an outline
of this process. Advancements in the field have
resulted in deep learning models that can provide further
accuracy in forecasting and more customization to the
problem at hand.
In this section, we provide an overview of current anaData
Preparation
Variable
Selection
Model
Training
Model
Assessment
Figure
2. A typical analytical modeling process.
52
IEEE Electrification Magazine / DECEMBER 2022
lytical techniques. We discuss three main categories of
machine learning: supervised, unsupervised, and reinforcement
learning.
Supervised models have a labeled dataset, meaning
that the outcome is known throughout most of the historical
data. Examples of supervised models include linear
regression, general linear models, decision trees, gradient
boosting, forest, neural networks, and time series forecasting
models. Recent developments of supervised models
are the development of deep learning models like recurrent
neural networks (RNNs) and long short-term memory
(LSTM) models.
Supervised models have many uses for utilities and are
widely used in energy forecasting to predict electricity
load, peak demand, and solar/wind production. Each method
has strengths and weaknesses, and an ensemble of
IEEE Electrification - December 2022

Table of Contents for the Digital Edition of IEEE Electrification - December 2022
