IEEE Power & Energy Magazine - May/June 2022 - 55

W
may/june 2022
WITH THE PROGRESSIVE ADVANCEMENT OF THE
smart grid paradigm, electricity systems worldwide are
deploying advanced metering technologies in residential
sectors. Smart meters can monitor and process the electricity
use of households at very high sampling rates. The availability
of hundreds of thousands of time-series load profiles
opens the possibility of applying clustering methods for
grouping consumers with similar consumption patterns.
In this article, we describe the main characteristics of
load profiles and explore several clustering methods for customer
classification and how they can improve the operation
of electricity systems. We also review the most widely used
clustering techniques for smart meter data and present some
recent advances in this field. Specifically, we illustrate how
smart meter clustering techniques need to be fast and efficient,
robust against outliers and missing data, and scalable
to hundreds of thousands of time series.
Smart Grids Revolutionize the
Energy Analytics Paradigm
Power systems are being upgraded worldwide as part of
a transition toward climate-neutral systems. The trend
is to depart from traditional centralized and carbon-based
energy systems to affordable, renewable, and reliable
frameworks. In this regard, several initiatives are being
considered that affect both the regulatory and technical
operations of these systems.
One of the main drivers of this transition is the need for
a full digitalization of the electricity supply chain. In particular,
the integration of modern information and telecommunication
technologies plays a key role in securing the grid
operation while fostering an active role of electricity consumers.
This is aligned with the concept of a " smart grid, "
an electrical network that can integrate the actions of all of
its users to achieve a sustainable, secure, safe, and economically
efficient power system.
These technologies open new business and technical possibilities
(and big challenges, too) in the operation of power
systems. For instance, we can highlight new concepts like distributed
energy resources, prosumers, demand-side management,
peer-to-peer energy trading, virtual power plants, electric
vehicles, and home batteries-all of which need to be deeply
studied and capitalized by energy utilities and system operators.
More specifically, the locally managed electricity systems
known as microgrids are probably undertaking the major
technical transformations in power systems. The installation
of new power electronics that allow bidirectional power flows,
together with advanced control and protection technologies,
is greatly improving the microgrid operation and reliability.
One of the main game changers in the efficient operation
of microgrids is the deployment of advanced metering infrastructures.
For example, around 107 million smart meters
are estimated to be currently installed in the United States
(more than 60% penetration rate) and 123 million in Europe
(a 43% penetration rate), which plans to increase this number
to 266 million (a 92% penetration rate) by 2030. In particular,
Spain, one of the European leaders, achieved a penetration
rate above 99% in 2020.
In the past, aggregated household consumption data
were collected every one or two months. The installation of
smart meters allows utilities to monitor, record, and transmit
household consumption data at a high resolution (typically
from every minute to every hour), leading to a significant
increase in data volume. This technology shift brings
new and important technical and analytical challenges to
process, store, and interpret massive electricity time series.
Thanks to other advanced metering infrastructure technologies,
the smart meter data can now be combined with the
other relevant variables (e.g., weather, mobility, presence,
demographics, economics, and so on) to enrich applications
of advanced data analytics in power grid operation.
Indeed, there is a growing interest in how these new data
sets can be exploited from an analytical point of view-to
gain knowledge by extracting valuable information. On one
hand, market/system operators seek to improve system operation
reliability and efficiency to reduce CO2 emissions and
lower market prices. On the other hand, retailing/generation
utilities are interested in staying competitive while increasing
their profits and customer satisfaction. The electricity industry
now has the opportunity to update and improve its traditional
analytical techniques, which until recently were mostly
based on small data sets and basic descriptive and predictive
approaches. In this regard, prescriptive analytics methods
may need to be implemented to aid in the decision-making
process and cope with complex real-time and massive data
inputs from the smart grid.
This new opportunity is being tackled from the perspective
of big data analytics. Not only is data volume an issue, but
also its variety (structured and unstructured data, e.g., energy
use and weather time series, digital images, sensor data, social
media, voice recordings, and so on) and the need for processing
velocity present challenges. Recent developments in machine
learning and artificial intelligence may foster data mining and
feature selection processes for designing accurate and efficient
prescriptive decision-making tools.
Main Characteristics of Individual
Household Time Series
In this section, we illustrate the main characteristics of a realworld
smart meter data set. Endesa, one of the main electric
utilities in Spain, has provided a sample of 10,000 household
consumptions on an hourly basis for one year. For each household,
a time series containing 8,760 electricity energy consumptions
intervals is provided. Specifically, these households
have power consumptions below 10 kW, and they are distributed
throughout the geography of Spain. The data were anonymized
in compliance with all data protection regulations.
Figure 1 shows the consumption of three clients over
the year. They have been selected to illustrate their different
consumption behavior. Figure 1(a) represents the average
ieee power & energy magazine
55
IEEE Power & Energy Magazine - May/June 2022

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - May/June 2022
