energies

Article
Model for Identification of Electrical Appliance and
Determination of Patterns Using High-Resolution Wireless
Sensor NETWORK for the Efficient Home Energy Consumption
Based on Deep Learning
Fernando Ulloa-Vásquez 1 , Victor Heredia-Figueroa 2, * , Cristóbal Espinoza-Iriarte 2 , José Tobar-Ríos 2 ,
Fernanda Aguayo-Reyes 2 , Dante Carrizo 3 and Luis García-Santander 4

1 Facultad de Ingeniería, Universidad Tecnológica Metropolitana, Santiago de Chile 7800002, Chile;

fulloa@utem.cl
2 Programa de Investigación en Radiocomunicación Digital, Facultad de Ingeniería, Universidad Tecnológica
Metropolitana, Santiago de Chile 7800022, Chile; cristobal.espinozai@utem.cl (C.E.-I.);
jtobarr@utem.cl (J.T.-R.); faguayo1001@gmail.com (F.A.-R.)
3 Departamento Ing. Informatica y Cs. de la Computación, Facultad de Ingeniería, Universidad de Atacama,
Copiapó 1531772, Chile; dante.carrizo@uda.cl
4 Departamento de Ingeniería Eléctrica, Universidad de Concepción, Concepción 4089100, Chile;
luis.garcia@udec.cl
* Correspondence: victor.herediaf@utem.cl

Abstract: The growing demand for electricity and the constant increase in electricity rates have
intensified the interest of residential and non-residential energy consumers to reduce their energy
consumption. The introduction of non-conventional renewable energies (photovoltaic and wind, in
the residential case) demands new proposals to obtain a home energy management system (HEMS),
Citation: Ulloa-Vásquez, F.;
which allows reducing the use of electrical energy. This article incorporates artificial intelligence
Heredia-Figueroa, V.; Espinoza-Iriarte,
C.; Tobar-Ríos, J.; Aguayo-Reyes, F.;
techniques to demand response, allowing control, switching, turning on and off of appliances,
Carrizo, D.; García-Santander, L. modifying and reducing consumption, and achieving improvements in the quality of life in the home.
Model for Identification of Electrical In addition, an architecture based on a smart socket and an artificial intelligence model that recognizes
Appliance and Determination of the consumption of electrical appliances in high resolution (sampling every 10 s) is proposed. The
Patterns Using High-Resolution system uses the Wi-Fi communication protocol, ensuring that the smart sockets wirelessly provide
Wireless Sensor NETWORK for the the data obtained to the public cloud. The use of Deep Learning allows us to obtain a central control
Efficient Home Energy Consumption model of the home, which, when interconnected to the smart electrical distribution networks of
Based on Deep Learning. Energies companies, could generate a positive impact on the environmental effects and CO2 reduction.
2024, 17, 1452. https://doi.org/
10.3390/en17061452
Keywords: deep learning; AMR; smart-meter; smart-socket; HEMS; smart-cities; ILM
Academic Editors: Krushna
Mahapatra and José Matas

Received: 16 December 2023

1. Introduction
Revised: 12 February 2024
Accepted: 21 February 2024 Home energy management systems (HEMS) play a crucial role in measuring, control-
Published: 18 March 2024 ling and optimizing the energy consumption of appliances and appliances belonging to a
residential user’s home. For the electricity distribution company, knowing the behavior
profile allows the use of the capacity of its infrastructure to be maximized, resulting in a
reduction in emissions and contributing to sustainable development. For its part, for the
Copyright: © 2024 by the authors. residential consumer, it allows the responsible use of energy, resulting in a reduction in
Licensee MDPI, Basel, Switzerland.
their monthly billing. As a whole, the technological advances of recent years in the commu-
This article is an open access article
nication and information sectors are allowing the current electricity supply service to be
distributed under the terms and
more modern and efficient by improving control, monitoring, detection and automation
conditions of the Creative Commons
capabilities, thanks to the use of artificial intelligence.
Attribution (CC BY) license (https://
The proposal to face this new challenge is to achieve the development of a home energy
creativecommons.org/licenses/by/
4.0/).
demand management platform (HAN-HEMS) through the measurement and estimation

Energies 2024, 17, 1452. https://doi.org/10.3390/en17061452 https://www.mdpi.com/journal/energies

Energies 2024, 17, 1452 2 of 19

of the energy consumption of a home, using agent nodes and an intelligent system based
on the availability of non-conventional renewable energy, so that the platform allows
consumption control and remote monitoring through the Internet. In order to identify
the surplus of renewable energy harvested by the end user, it is necessary to obtain a
classification of electrical appliances and a real-time consumption profile which is defined
by their daily habits of using household appliances.
In this article, a model for identifying electrical devices and determining consumption
patterns is proposed through a network of high-resolution sensors for the efficient use
of home energy based on the availability of electrical energy with the use of artificial
intelligence. In this way, and through Smart-Sockets, which allow the measurement of the
real demand, together with the estimation of the user’s home consumption, the model
based on deep learning associated with the classification of time series, seeks greater energy
efficiency by part of the end customer. The experimental results show that the proposed
smart plug can correctly read the power consumption of wirelessly connected devices
without data loss. The central controller can effectively control multiple sockets based on
user program code. With the proposed architecture of smart meters, in conjunction with the
artificial intelligence model, it is possible to control electrical loads, enabling or disabling
electrical appliances so that the user, based on their stored and/or generated energy budget,
can sustain themselves. The rest of the document is organized as follows: section two
presents the works related to this research; Section three describes the methodology used
and the development of the proposed system; Section four presents the experimental
development and the results obtained; Section five shows the discussion of results and
Section six shows the conclusions of the paper and future work.

2. Literature Review
The massification of the meters of accountants, smart devices, low-cost sensors and
smart home appliances over the last decade has encouraged a favorable environment
for the development of new management strategies of demand response, which include
communication, decision-making and the interaction between users, devices and the
electrical grid. Therefore, there is a problem of interest in smart networks or smart grids:
obtaining, storing and analyzing hard data on the electrical consumption of residential
users alone is not sufficient to provide root cause elements of demand response and what it
can do to reduce and/or optimize it [1]. Additionally, the more we want to delve into this
last aspect, the higher the resolution of the data needed for a precise energy analysis, which
implies, as a final consequence, overcoming these technological obstacles by automating
the operations of the electrical network.
An important element that motivates the development of this work is the different
alternatives that exist for measuring and/or estimating the energy consumption of a home,
commonly called the “pattern of consumption of electrical appliances”.
HEMS use monitoring techniques that are generally classified into non-intrusive
load monitoring (NILM) and intrusive load monitoring (ILM) [2]. The characteristics that
differentiate these two techniques are that NILM only uses one monitoring point that
is generally located at the main power input of the home’s electrical system (known as
Smart Meter), while ILM uses sensing devices in each of the sockets (Smart Socket SS;
Smart-Plug SP) of all or main connected loads [3]. On the other hand, NILM systems have
the advantage of not intervening in the home circuit with expensive monitoring devices [4].
On the other hand, ILM systems, by having sensors in different loads of the circuit, provide
more precise details about the consumption and alarms referring to each appliance [5],
making it possible to monitor very low-power devices and differentiate between variable
consumption and consumption between a set of devices [6]. Scientific publications on
NILM-type approaches outnumber those on the ILM approach [7], as the NILM approach
proves to be much older.
The climate change that the planet is experiencing rapidly has led each country to
propose strategies that involve the efficient use of energy resources [8]. Particularly, the
Energies 2024, 17, 1452 3 of 19

Chilean state has established an energy policy for the year 2050, which establishes energy
efficiency as one of its pillars [9]. In this context, the residential consumer has taken an active
role in managing their use of electrical energy and thereby influencing the behavior of their
consumption pattern. These assertions inspire the development of new technologies and
methodologies that allow the participation of consumers in the operation of the electrical
network at the end-user level, an example of this is the smart-socket wireless sensor
networks, which, apart from measuring each appliance, can turn on or off.
A smart-sockets sensor network is designed for use under multiple operating states as
a single device. This feature is particularly useful for work, industrial, building and do-
mestic environments, since by identifying the appliances that are used in that space, based
on the behavior of the current patterns consumed by the electrical appliance, it is possible
to monitor, program and control the energy consumption efficiently [10]. Achieving the
desired degree of coverage in space and time requires the use of appropriate communica-
tion protocols for each use case (Wi-Fi, PLC, or ZigBee, among others). Seamless sensor
network integration is important for success in telemetry and ubiquitous computing.
In [11,12], various hardware and software architectures of systems that analyze test
devices are discussed and the current consumption patterns and profiles of home appliances
are explained. In addition to this, the relevance of these architectures for the development
and implementation of new infrastructures and applications for smart cities is explained.
From this review, relevant background information is presented such as the use of classifier
algorithms that do not need training, but do require a database that is updated periodically
and automatically, thus generating the need for a dynamic connected data storage to each
of the electronic devices, and consequently take advantage of the benefits that the world of
the public cloud offers [13]. From this point, it is relevant to mention that the identification
of the consumption patterns of electrical appliances focuses on energy disaggregation and
device recognition.
The works [14–17] show the development of an intelligent system that analyzes the
periodic use of electrical appliances in a home to extract and determine the behavioral
patterns of residential users in an IoT environment. Particularly, in [15,16] there is an
additional objective within their research which consists of reducing the excessive electricity
consumption of appliances through an alert system. In [18–20] data obtained by Smart-Plug
type smart meters are used to identify appliances, using an approach based on artificial
neural networks. The results of [20] are considered acceptable, given that the score of the
classification capacity estimator obtained with respect to the recognition of F1 appliances
reaches 77.6%. In [21] they also use Smart Plugs for data collection; however, in this
case, the authors intelligently process the collected time series data and classify them
using multi-layer feedback neural networks. The results of their experiment allow us to
obtain the energy consumption pattern of various appliances, based on a daily cycle of
use. A k-active neighbor-based appliance recognition approach is presented in [22] to
learn from unlabeled data collected through Arduino-operated Smart Plugs. In [23] the
authors provide an affinity propagation clustering algorithm with the NILM approach
based on a vector graphical model and the theory of belief propagation. The experiment
carried out demonstrates that it is possible to correctly recognize basic and combined
classes of electrical devices. In [24,25], various unsupervised machine learning techniques
are presented to identify electricity consumption patterns in home environments. The
particularity of these articles is that additional data mining techniques are used to meet the
aforementioned objective. In works [26–28] algorithms based on fuzzy logic are used that,
by activating the calculation of parameters such as maximum power, average power and
cycle duration, obtain a result for the membership function, thus achieving the identification
of electrical devices through degrees of truth. The works [29,30] carry out data sampling
in the order of 10 to 15 min, for the automatic recognition of electrical devices using
fuzzy logic, since their objective is to determine the energy consumption associated with
billing systems.
Energies 2024, 17, x FOR PEER REVIEW 4 of 20

Energies 2024, 17, 1452 4 of 19

fuzzy logic, since their objective is to determine the energy consumption associated with
billing systems.
In
In this
this work,
work,aahomehomeelectrical
electricalenergy
energymanagement
management model
modelis proposed
is proposedbased on the
based on
capture of electrical
the capture parameters
of electrical parametersthrough smart
through sockets,
smart allowing
sockets, residential
allowing usersusers
residential to par-to
ticipate in the
participate control
in the controlof of
demand
demandresponse
responseininreal-time,
real-time,thus
thusadapting
adapting consumption
consumption in in
such
such a a way
way toto allow
allow the
the reduction
reduction in in demand
demand response
response during
during peak
peak hours,
hours, intelligently
intelligently
adapt consumptiontotooff-peak
adapt consumption off-peak hours,
hours, make
make responsible
responsible use use of energy
of energy and asand as a conse-
a consequence,
quence,
obtain a obtain
reductiona reduction in the billing.
in the monthly monthlyThe billing. The proposed
proposed model managesmodel tomanages
achieveto a
achieve a sampling rate of the order of 10 s, which is higher than the works
sampling rate of the order of 10 s, which is higher than the works [31–35]. This resolution [31–35]. This
resolution not only
not only allows allows the differentiation
the differentiation of each
of each electrical electrical
device device
but also but also the
the activation of activa-
alarms
tion
and alerts. Finally, the Deep Learning algorithm used, based on a multi-layera neural
of alarms and alerts. Finally, the Deep Learning algorithm used, based on multi-
layer neural network, manages to discriminate each of the devices
network, manages to discriminate each of the devices with great efficiency (the scorewith great efficiency
(the
of thescore of the regarding
estimator estimator regarding the classification
the classification capacity ofcapacity of an appliance
an appliance with respectwithto re-
its
spect to its identification
identification F1 reaches
F1 reaches 77.6%), 77.6%),
sending sending information
information in the
in real-time to real-time to the user,
user, determining
determining
the electrical the electrical consumption
consumption of the home and of the home
each and
of the each of the appliances.
appliances.

3. Methodology and System Development

Figure 1 shows the schematic diagram that briefly describes each of the stages of the
methodology implemented in this work.

Figure 1. Description of the methodology used.

1. Phase (1).
1. Phase (1).
A current/voltage measurement module with Internet Protocol (IP) wireless commu-
A current/voltage measurement module with Internet Protocol (IP) wireless commu-
nication is designed. The ranges of maximum values are established for the voltage
nication is designed. The ranges of maximum values are established for the voltage
(<230 Volts) and for the current (<16 Amperes) and, finally, a protection system for
(<230 Volts)
overloads and and for the current
automatic (<16isAmperes)
operation configured and, finally,
based a protection
on the system for
electrical parameters
overloads and automatic operation is configured based on the electrical
already defined. The smart-socket type device allows the capture of electrical parame- parameters
already
ters suchdefined. The current,
as voltage, smart-socket
activetype
powerdevice
andallows
reactive the capture
power, as of electrical
well param-
as the sending
eters such as voltage, current, active power and reactive power, as well
of data in real-time, securely (mqtts protocol), to a public cloud repository/service. as the sending
2. of data(2).
Phase in real-time, securely (mqtts protocol), to a public cloud repository/service.
2. Phase (2). measurements are recorded for a set of previously defined electrical appli-
Data and
ances,
Data which
and are part of are
measurements a common
recordedhomefor a environment
set of previouslyfor adefined
residential user appli-
electrical in the
Santiago
ances, Metropolitan
which are part ofRegion.
a common home environment for a residential user in the
Santiago
a. Metropolitan
Phase (2a). Region.
a. Phase (2a).
Measurements of the seven most common household appliances are recorded
and measuredofinthe
Measurements 10 seven
different
mosthomes,
common withhousehold
samplingappliances
times of 10are s and work
recorded
cycles
and that depend
measured on the operation
in 10 different homes, of each
with appliance.
sampling In this
times of 10phase,
s andparticular
work cy-
care
cles thatmust be taken
depend on thewith some elements
operation that correspond
of each appliance. to recurring
In this phase, failures
particular care
in data collection tasks: Failures in data integrity associated
must be taken with some elements that correspond to recurring failures in data with data storage
problems;
collection connection
tasks: Failuresproblems between
in data integrity the device
associated and
with thestorage
data public problems;
cloud ser-
vices used for data analysis; inconsistencies in data structuring
connection problems between the device and the public cloud services used (data typeforor
column type).
data analysis; inconsistencies in data structuring (data type or column type).
b. Phase (2b).
b. Phase (2b).
This exploratory analysis phase seeks to optimally manage the data source
This exploratory analysis phase seeks to optimally manage the data source gen-
generated from the collection phase, with the aim of facilitating the identifi-
erated
cationfrom the collection
of patterns and/or phase, withanomalies.
possible the aim of facilitating
To this end, thethe
identification
review of the of
patterns and/or possible anomalies. To this end, the review of
following elements is considered: Review of missing data and its possible im-the following ele-
ments is considered:
putation; Reviewusing
variable selection of missing
statisticaldata and itsreview
methods; possible imputation;
of distributions
and correlations; variable engineering.
3. Phase (3).
 3. Phase (3).
It contemplates the implementation of a multi-layer artificial neural network, where
the nodes of these layers are linked in a unidirectional manner to identify non-linear
characteristics.
Energies 2024, 17, 1452 4. Phase (4). 5 of 19
The data are sent to the cloud with the objective of generating representative values
of each of the appliances to be compared. The respective fingerprint of electrical con-
It contemplates
sumption the implementation
is identified for each of theofappliances.
a multi-layer artificial neural network, where
the nodes of these layers are linked in a unidirectional manner to identify non-
a. Phase (4a).
linear characteristics.
The implemented neural network learns through training. Examples with
4. Phase (4).
The dataknown are results
sent to and the responses
the cloud to compare
with the objective with the results
of generating of the proposed
representative values
model are continually presented to the neural
of each of the appliances to be compared. The respective fingerprint network. For this phase to deliver
of electrical
the correct
consumption performance,
is identified it is necessary
for each to take into consideration the following:
of the appliances.
Definition of the method of dividing the data sets; definition of the model to be
a. Phase (4a).
used to verify or refute the initially stated hypothesis; definition of model pa-
The implemented neural network learns through training. Examples with
rameterization.
known results and the responses to compare with the results of the proposed
b. Phase model (4b).
are continually presented to the neural network. For this phase to
For
deliverevaluation
the the correctofperformance,
the model, the following
it is necessary elements
to take are
intoconsidered:
consideration Defini-
the
tion of the set
following: of tests toofbe
Definition compared;
the method of definition
dividingof the
the acceptance
data threshold
sets; definition for
of the
performance
model to be metrics, in order
used to verify to evaluate
or refute whether
the initially the performance
stated or accuracy
hypothesis; definition of
ofmodel
the model is satisfactory or not.
parameterization.
b. Phase (4b).
The developed system allows data collection through the use of a network of intelli-
gent wireless Forsensors,
the evaluation of the model, the
called smart-sockets. Thesefollowing elements
single-phase are considered:
devices are located in Defini-
each
outlet in the tion of theallowing
home, set of teststhetomonitoring
be compared; of definition
electrical of the acceptance
parameters threshold
of each appliance,for
performance metrics, in order to evaluate whether the
such as voltage, current and power, at nominal 50 Hz and 220 VAC loads, considering 230 performance or accuracy
Volts and 16 of Amps
the model as theis maximum
satisfactoryoperating
or not. values. Additionally, the proposed model
has the
Theflexibility
developedofsystem the portability
allows data of the devicethrough
collection to different power
the use outlets within
of a network the de-
of intelligent
ployed coverage environment, whose operation can be visible through
wireless sensors, called smart-sockets. These single-phase devices are located in each outlet a web platform or
via a smartphone-type device (Figure 2).
in the home, allowing the monitoring of electrical parameters of each appliance, such as
As Smart
voltage, currentSockets are fundamentally
and power, at nominal 50orientedHz andto monitoring
220 VAC loads, domestic electrical
considering 230 loads,
Volts
they16are
and Ampsexpected
as thetomaximum
operate under the following
operating conditions: the proposed model has the
values. Additionally,
•
flexibility
Loads ofwith
the portability
low presence of the deviceof
or impact totransients
different power outlets
or electrical within the deployed
impulses.
coverage
• Electrical network with good stability, voltage regulation and lowplatform
environment, whose operation can be visible through a web presenceorofviahar- a
smartphone-type
monics. device (Figure 2).

Figure2.2. Smart
Figure SmartSocket
SocketILM
ILMsystem
systemdeveloped.
developed.

As
TheSmart Sockets
proposed are fundamentally
architecture oriented
for this Smart to monitoring
Socket considers thedomestic electrical
use of the loads,
Wi-Fi stand-
they are expected to operate under the following conditions:
ard as the communication protocol. The microprocessing unit considered for this device
•is the ESP8266
Loads chip,
with low which or
presence allows
impactthe processingorof
of transients the measured
electrical impulses.data and their
• Electrical network with good stability, voltage regulation and low presence of harmonics.
The proposed architecture for this Smart Socket considers the use of the Wi-Fi standard
as the communication protocol. The microprocessing unit considered for this device is the
ESP8266 chip, which allows the processing of the measured data and their transmission to
a gateway-type device, which is responsible for transmitting data to public cloud services
over the Internet. Voltage measurement is performed through a small power transformer
and current measurement is performed through a shunt resistor capable of measuring up to
16 Amps, allowing power calculation (Figure 3). With all the parameters already described,
it is possible to calculate the power and energy consumed by a certain electrical appli-
 transmission to a gateway-type device, which is responsible for transmitting data to public
cloud services over the Internet. Voltage measurement is performed through a small
Energies 2024, 17, 1452 power transformer and current measurement is performed through a shunt resistor 6capa- of 19
ble of measuring up to 16 Amps, allowing power calculation (Figure 3). With all the pa-
rameters already described, it is possible to calculate the power and energy consumed by
a certain
ance. electrical
Based appliance. Based
on the architecture on thethe
described, architecture
summary described, the summary
of smart-sockets of smart-
functionalities is
sockets functionalities
as follows: is as follows:
•• Measure
Measurethethevoltage
voltageand
andinstantaneous
instantaneouscurrent
currentthat
thatare
arepresent
presentininthe
theappliance.
appliance.
•• Calculate
Calculatethe
theinstantaneous
instantaneouspower
powerdemanded
demandedby bythe
theconnected
connectedappliance.
appliance.
•• Calculate
Calculatethe
theamount
amountof ofenergy
energyconsumed
consumedby bythe
theconnected
connectedappliance.
appliance.
•• Execute
Execute the
the communication
communication request
request algorithm
algorithm to
to transmit
transmit the
thepower
powerconsumption
consumption
andmake
and makethe thedecision
decisionto
toturn
turnon/off
on/off the smart-socket device by the coordinator.
coordinator.

Figure3.3.Architecture
Figure Architectureof
ofaaSmart
SmartSocket.
Socket.

Given
Giventhetheneed
needtotogenerate
generatea arecognition
recognition model
model forfor
electrical devices
electrical devices that is economical
that is econom-
and easy to implement with a low-power integrated circuit it was, therefore,
ical and easy to implement with a low-power integrated circuit it was, therefore, chosen chosen to use
atoclassification model based on time series. In this work, and since time series
use a classification model based on time series. In this work, and since time series data data cannot
be used be
cannot directly
used in a classification
directly model, itmodel,
in a classification was necessary to generate
it was necessary to agenerate
representation that
a represen-
role models the characteristics of the series, generating a fixed input for
tation that role models the characteristics of the series, generating a fixed input for the the subsequent
classification model. To carry
subsequent classification model.outTo
this task,
carry outthe “Time
this Series
task, the “TimeFeature
Series Extraction on Basis
Feature Extraction
of
on Basis of Scalable Hypothesis test” method was used. After the representation series
Scalable Hypothesis test” method was used. After the representation of the time of the
using the “Red
time series usingFeed Forward”
the “Red Feed model based
Forward” modelon Deep
basedLearning, and considering
on Deep Learning, as an
and consider-
input value a curve vector of the electric charge and as an output the recognition of an
ing as an input value a curve vector of the electric charge and as an output the recognition
appliance, it was possible to determine and classify each electrical appliance connected to
of an appliance, it was possible to determine and classify each electrical appliance con-
the smart-socket device.
nected to the smart-socket device.
4. Experimental Development
4. Experimental Development
The steps carried out for experimental development are highlighted below:
The steps carried out for experimental development are highlighted below:
4.1. Data Set
4.1. Data Set study, a classification model for electrical appliances is designed and im-
For this
For this
plemented study,
based onatime
classification model with
series analysis, for electrical appliances
the objective is designed
of detecting and imple-
and classifying
mented based on time series analysis, with the objective of detecting
appliances connected to the home electrical network. For this experiment, the proposed and classifying ap-
plianceswas
system connected
installedtointhe
10home
homeselectrical network.
in Santiago Forduring
de Chile this experiment,
the months the
ofproposed sys-
July, August,
temSeptember,
and was installedtheinmiddle
10 homesof thein Santiago de Chile
winter. Each during
residence hasthe months of July,
measurement August,with
equipment and
September,
an the middle
ILM (intrusive) of theinstalled
approach, winter. Each residence
in each electricalhas measurement
outlet in the home, equipment
but more with an
precise
ILMreliable
and (intrusive)
for approach, installed in and
on-off measurement eachcontrol.
electricalThe
outlet in the home,
different but more
consumption precise
patterns
and reliable
generated for constructed
were on-off measurement and control.
and analyzed Thethat
from data different
come consumption
from multiplepatterns
brands gen-
and
models of household
erated were appliances
constructed present in
and analyzed the data
from homesthatof the
come collaborators
from multipleof this research.
brands and
To carry
models out the data
of household collection
appliances process
present in associated
the homes withof thethe consumption
collaborators of of household
this research.
appliances,
To carry theout
operation
the data of the electrical
collection processparameters
associated of each
with theofconsumption
them was measured
of house-
through an intrusive
hold appliances, charging of
the operation system using a parameters
the electrical high-resolution
of each AMR (Automatic
of them Meter
was measured
Reading), which recorded the behavior of each electrical device in terms of electrical
current consumed under multiple operating states. Between 360 and 400 samples were
obtained per hour of consumption of each of the variables for each electrical appliance
under study: washing machine, refrigerator, kettle, toaster, heater, router and electric stove.
The intrusive charging system recorded a sample measurement of electrical consumption
 through an intrusive charging system using a high-resolution AMR (Automatic Meter
Reading), which recorded the behavior of each electrical device in terms of electrical cur-
rent consumed under multiple operating states. Between 360 and 400 samples were ob-
Energies 2024, 17, 1452 tained per hour of consumption of each of the variables for each electrical appliance under 7 of 19
study: washing machine, refrigerator, kettle, toaster, heater, router and electric stove. The
intrusive charging system recorded a sample measurement of electrical consumption
every
every 10
10 ss and
and the
the behavior
behaviorwas
was measured
measuredfor forat
atleast
least24
24hhof
ofoperation.
operation. Despite
Despite the
the large
large
volume
volumeof ofdata
dataobtained,
obtained,only
only70%
70%of ofthe
thedata
datawere
wereused
usedfor
fortraining
trainingthe
themodel.
model.
The
Theinitial
initialresults
resultsshown
shownin in this
this article
article were
were obtained
obtained after
after experimentation
experimentationwith with only
only
seven
seven appliances, given
given that
thatthey
theyare
arethetheelectrical
electrical devices
devices of of main
main useuse
forfor a domestic
a domestic hu-
human being (heating, cooking food, washing clothes and teleworking).
man being (heating, cooking food, washing clothes and teleworking). Figure 4 shows
shows thethe
simultaneous
simultaneousbehavior
behaviorofofthe
theseven
seven devices
devices depending
dependingon on the
the electrical
electrical current
current demanded
demanded
during
during 60
60min
minofofoperation.
operation.

Figure4.4.Simultaneous
Figure Simultaneouselectrical current
electrical consumption
current profiles
consumption of appliances
profiles under
of appliances studystudy
under in oneinhour
one
hour of consumption. The average arithmetic consumption is 675
of consumption. The average arithmetic consumption is 675 [watts/h].[watts/h].

Although
Although the the total
total household
household energy
energy consumption
consumption behavior
behavior tends
tends to
to be
be complex
complex to to
visualize, if the curves are separated intuitively, it can be seen that some
visualize, if the curves are separated intuitively, it can be seen that some devices devices have similar
have
behaviors in relation
similar behaviors in to the electrical
relation current demanded
to the electrical (inductive,(inductive,
current demanded resistive elements
resistiveand
ele-
non-linear
ments andloads, for example,
non-linear data
loads, for transmission).
example, It is preciselyItthis
data transmission). differentiation
is precisely that we
this differenti-
seek
ationtothat
replicate
we seekthrough artificial
to replicate intelligence
through techniques,
artificial allowing
intelligence the detection,
techniques, allowingprediction
the de-
and classification of electrical devices according to their electrical behavior
tection, prediction and classification of electrical devices according to their electrical be-depending on
time and conditions of use.
havior depending on time and conditions of use.
4.2. Appliance Classification Model
4.2. Appliance Classification Model
Once the parameters and variables of the electrical energy consumption of each device
Once the parameters and variables of the electrical energy consumption of each de-
under study were captured, a source of data and information was generated as a means of
vice under study were captured, a source of data and information was generated as a
training the model. From this point, a household appliance classification model based on
means of training the model. From this point, a household appliance classification model
the time series is developed.
based on the time series is developed.
Since time series data cannot be used directly in a classification model, it is necessary
Since time series data cannot be used directly in a classification model, it is necessary
to generate a representation that allows modeling the characteristics of the series and
to generate a representation that allows modeling the characteristics of the series and gen-
generates a static input for the subsequent classification model. To carry out this task,
erates a static input for the subsequent classification model. To carry out this task, the
the “Time Series Feature Extraction on Basis of Scalable Hypothesis Tests” method was
“Time Series Feature Extraction on Basis of Scalable Hypothesis Tests” method was used,
used, which allows for generating many time series features, since it contains 63 series
which allows
methods for generating
that contemplate many
a total oftime series features,
794 attributes sincerespective
with their it containssignificance.
63 series methods
After
that contemplate a total of 794 attributes with their respective significance.
the representation of the time series, a Deep Learning model called “Feed Forward” After the was
rep-
resentation of the time series, a Deep Learning model
used, where the parameters used were the following (Table 1): called “Feed Forward” was used,
where the parameters used were the following (Table 1):
Table 1. Parameters defined in the neural network model used.
Table 1. Parameters defined in the neural network model used.
Parameters Values
Parameters Values
Hidden layers 5
Hidden layers 5
Neurons by layers 512, 256, 128, 64, 32
Neurons by layers
Optimizer 512, 256,
SGD128, 64, 32
Optimizer
Learning rate SGD
0.001
Iterations
Learning rate 250
0.001

The graphical detail of the network architecture used can be analyzed in Figure 5
and corresponds to a neural network with five hidden layers where the defined input
is a statistical representation of the behavior of the time series data and the result is the
classification of the respective appliance.
 Iterations
Iterations 250
250

The graphical
The graphical detail
detail of
of the
the network
network architecture
architecture used
used can
can be
be analyzed
analyzed in
in Figure
Figure 55 and
and
corresponds to a neural network with five hidden layers where the defined input
corresponds to a neural network with five hidden layers where the defined input is a sta- is a sta-
Energies 2024, 17, 1452 tistical representation of the behavior of the time series data and the result is the classifi-
tistical representation of the behavior of the time series data and the result is the classifi-
8 of 19
cation of
cation of the
the respective
respective appliance.
appliance.

Figure
Figure 5.
Figure 5. Graphical
5. Graphical representation
Graphical representation of
representation of the
of the neural
the neural network
neural network architecture
network architecture used.
architectureused.
architecture used.
used.

Finally, and
Finally, and for
for model
model validation
validation purposes,
purposes, the
the accuracy
accuracy measure
measure is
is used,
used, given
given that
that
the classes are represented in a balanced
the classes are represented in a balanced way with the number
balanced way with the number of
number of measurements.
of measurements.
measurements.

4.3. Results
4.3. Results
4.3.
First campaign
campaign of
First campaign
First measures:
of measures:
of During
measures: During the
During the months
the months of
months of April, May
of April, May and
and June
June 2020,
2020, autumn
autumn
in Santiago
in Santiago
in Santiago de de Chile,
de Chile, measurements
Chile, measurements
measurements of of electrical
of electrical appliances
electrical appliances were
appliances were carried
were carried out
carried out using
out using Smart
using Smart
Smart
Sockets ILM
ILM in
Sockets ILM
Sockets in
in 2020 different
20 different homes,
different homes,
homes, withwith sampling
with sampling
sampling timestimes
times of of
of 20 20 s
20 ss andand
and workwork
work cycles cycles
thatthat
cycles that de-
de-
depended
pended on on
thethe operation
operation ofof each
each appliance.
appliance. These
These results
results are
are presented
presented below,
below,
pended on the operation of each appliance. These results are presented below, where the where
where the
the
measured
measured electrical
electrical device
device is
is clearly
clearly indicated,
indicated, along
along with
with its
its respective
respective electrical
electrical
measured electrical device is clearly indicated, along with its respective electrical current current
current
consumption
consumption pattern
consumption (Figures
(Figures 6–18):
pattern (Figures
pattern 6–18):
6–18):

Figure
Figure 6.
Figure 6. Electrical
6.Electrical
6. Electrical
Electrical current
current
current
current consumption
consumption
consumption
consumption pattern
pattern
pattern
pattern of ofUrsus
an
of an
of an an Ursus
Ursus
Ursus Trotter
Trotter
Trotter
Trotter 2000 2000
brandbrand
brand
2000 brand
2000 kettle kettle
[watts].[watts].
[watts].
kettle [watts].
kettle Meas-
Meas-
Meas-
urement interval
Measurement
urement interval
urement interval of 60
interval
of 60
of min.
60 of 60 min.
min.
min.

Figure 7. Electrical current consumption pattern of a generic 700 [watts] brand stove; 20 min
measurement interval.
 Energies 2024, 17, x FOR PEER REVIEW 9 of 20
Energies
Energies 2024,
2024, 17,
17, x
x FOR
FOR PEER
PEER REVIEW
REVIEW 99 of
of 20
20

Energies 2024, 17, 1452

Figure 7. Electrical current consumption pattern of a generic 700 [watts] brand stove; 20 min meas-9 of 19
Figure
Figure 7. Electrical
Electrical current
7. interval. current consumption
consumption pattern
pattern of
of aa generic
generic 700
700 [watts]
[watts] brand
brand stove;
stove; 20
20 min
min meas-
meas-
urement
urement interval.
urement interval.

Figure 8. Electrical
Electrical current
currentconsumption
consumptionpattern
patternofofaaSomela
Somela5555[watts]
[watts]fan.
fan.Measurement
Measurementinterval
interval of
Figure
Figure 8.
8. Electrical
Electrical current
current consumption
consumption pattern
pattern of
of aa Somela
Somela 55
55 [watts]
[watts] fan.
fan. Measurement
Measurement interval
interval
of
60 60 min.
min.
of
of 60
60 min.
min.

Figure 9. Electrical current consumption pattern of a Mademsa 200 [watts] fridge. Measurement
Figure
Figure Electricalcurrent
9. Electrical
Figure 9.
9. currentconsumption
current consumption
consumption pattern
pattern of of a Mademsa
aa Mademsa
Mademsa 200200 [watts]
[watts] fridge.
fridge. Measurement
Measurement
interval ofElectrical
60 min. pattern of 200 [watts] fridge. Measurement
interval
interval of 60
of 60
interval of min.
60min.
min.

Figure 10. Electric current consumption pattern of a Nec 2000 brand heater [watts]; 30 min measure-
Figure 10.
Energies 2024, 17, x FOR PEER REVIEW
Figure 10. Electric
10. Electriccurrent
Electric currentconsumption
current consumption
consumption pattern
pattern of
of aaofNec
pattern 2000
a Nec
Nec brand
2000
2000 heater
brand
brand [watts];
heater
heater 30
30 min
[watts];
[watts]; 30 measure-
min 10 of
min 20
measure-
measure-
ment interval.
ment
ment interval.
interval.
ment interval.

Figure 11.
Figure 11. Electrical
Electricalcurrent
currentconsumption
consumptionpattern of an
pattern HPHP
of an 245245
[watts] notebook;
[watts] 40 min
notebook; 40 measure-
min measure-
ment interval.
ment interval.
Energies 2024, 17, 1452 10 of 19
Figure 11. Electrical current consumption pattern of an HP 245 [watts] notebook; 40 min measure-
ment interval.
Figure 11. Electrical current consumption pattern of an HP 245 [watts] notebook; 40 min measure-
Figure 11. Electrical current consumption pattern of an HP 245 [watts] notebook; 40 min measure-
ment interval.
ment interval.

Figure 12. Electriccurrent

12.Electric currentconsumption
consumption pattern
pattern of a Thomas 1200 [watts] oven; 40 min measure-
Figure 12. Electric current consumption pattern of of a Thomas
a Thomas 1200
1200 [watts]
[watts] oven;
oven; 40 min
40 min measurement
measurement
Figure
ment 12. Electric
interval.
interval. current consumption pattern of a Thomas 1200 [watts] oven; 40 min measurement
interval.
interval.

Figure Electricalcurrent
Figure 13. Electrical currentconsumption
consumption pattern
pattern of of a Cisco
a Cisco brand
brand Wi-Fi
Wi-Fi router
router 60 [watts].
60 [watts]. Measure-
Measure-
Figure 13. Electrical
Figure 13. Electricalcurrent
currentconsumption
consumption pattern
pattern of of a Cisco
a Cisco brand
brand Wi-Fi
Wi-Fi router
router 60 [watts].
60 [watts]. Measure-
Measure-
ment interval
ment interval of
of 60
60 min.
min.
ment interval
ment interval of
of60
60min.
min.

Energies 2024, 17, x FOR PEER REVIEW 11 of 20

Figure
Figure 14. Electric current
14. Electric current consumption
consumption pattern
pattern of
of aa Samsung
Samsung brand
brand cell
cell phone
phone charger
charger 55 [watts].
[watts].
Measurement interval of 60 min.

Electricalcurrent
Figure 15. Electrical currentconsumption
consumptionpattern
patternofofananLG
LG7575 [watts]
[watts] television.
television. Measurement
Measurement in-
terval ofof
interval 6060
min.
min.
Energies 2024, 17, 1452 11 ofin-
19
Figure 15.
Figure 15. Electrical
15. Electrical current
Electrical current consumption
current consumption pattern
consumption pattern of
pattern of an
of an LG
an LG 75
LG 75 [watts]
75 [watts] television.
[watts] television. Measurement
television. Measurement
Measurement in-
in-
Figure
terval of
terval of 60
of 60 min.
60 min.
min.
terval

Figure 16.
Figure 16. Electrical current
Electrical
16. Electrical
Electrical current consumption
current consumption pattern
consumption pattern of
pattern of an
of an Electrolux
an Electrolux 200
Electrolux 200 [watts]
[watts]
200 [watts] washing
washing
[watts] washing machine.
machine.
washing machine.
machine.
Figure current consumption pattern of an Electrolux
Measurement interval
Measurement interval of
interval of 60
of 60 min.
60 min.
min.
Measurement

Figure
Figure 17.
Figure 17. Simultaneous
17.Simultaneous electrical
Simultaneouselectrical current
electricalcurrent consumption
consumption
current consumption profiles
profiles of appliances
of appliances
profiles under
under
of appliances
appliances study
study
under in
in one
study one
inhour
one
Figure 17. Simultaneous electrical current consumption profiles of under study in one
hour
hour
of of
of consumption.
consumption.
consumption. The The
The arithmetic
arithmetic
arithmetic average
average
average of of
of consumption
consumption
consumption is 850 is
is 850
850 [watts/h].
[watts/h].
[watts/h].
hour of consumption. The arithmetic average of consumption is 850 [watts/h].

Figure 18. Energy consumed simultaneously by appliances under study in one hour of consumption,
“Energy Budget” in the first campaign. The arithmetic average of consumption is 850 [watts/h].

Second campaign of measures: During the months of July, August and September
2023, the middle of the winter in Santiago de Chile, measurements of electrical appliances
were carried out using Smart Sockets ILM in 10 different homes, with sampling times of
10 s and work cycles that depended on the operation of each appliance. The motivations
for this new measurement campaign were the improvement in the sampling rate of the
Smart Socket used, going from a sampling every 20 s to a sampling of 10 s per piece of data
obtained, and the improvements implemented in the architecture and parameterization
of the measurement model of neural networks used previously. These results are more
precise than those previously obtained, and are presented below, where the electrical device
measured is clearly indicated, along with its respective electrical current consumption
pattern (Figures 19–27):
 Smart Socket used, going from a sampling every 20 s to a sampling of 10 s per piece of
data obtained, and the improvements implemented in the architecture and parameteriza-
tion of the measurement model of neural networks used previously. These results are
more precise than those previously obtained, and are presented below, where the electri-
Energies 2024, 17, 1452 cal device measured is clearly indicated, along with its respective electrical current12con-
of 19
sumption pattern (Figures 19–27):

Figure 19. Electrical

Figure 19. Electrical current
currentconsumption
consumptionpattern
patternofofa aMademsa
Mademsa1400 [watts]
1400 toaster;
[watts] 3 min
toaster; measure-
3 min meas-
ment interval.
urement interval.

Figure 20. Electrical

Figure 20. Electricalcurrent
currentconsumption
consumptionpattern
patternof of
a Somela 1800
a Somela brand
1800 kettle
brand [watts].
kettle Measurement
[watts]. Measure-
ment interval
interval of 60 min.
of 60 min.

Energies 2024, 17, x FOR PEER REVIEW 13 of 20

Figure 21. Electrical current consumption pattern of aa Mademsa

Mademsa 110
110 [watts]
[watts] fridge.
fridge. Measurement
Measurement
interval of 60 min.

Figure 22. Electrical current

Figure current consumption
consumption pattern
pattern of
of aa generic
generic GAMA
GAMA 1600
1600 [watts]
[watts] hair dryer; 10 min
measurement interval.
Energies 2024, 17, 1452
Figure 22. Electrical current consumption pattern of a generic GAMA 1600 [watts] hair dryer; 13
10 of
min19
measurement interval.

Figure
Figure 23. Electrical current
23. Electrical current consumption
consumption pattern
pattern of
of an
an Oster
Oster 700
700 brand
brand juicer
juicer [watts]. Measurement
[watts]. Measurement
interval of 60 min.

Energies
Energies2024,
2024,17,
17,xxFOR
FORPEER
PEERREVIEW
REVIEW 14 of
of 20
14inter-
20
Figure 24. Electrical
Figure 24. Electrical current
currentconsumption
consumptionpattern
patternofofan
anAskey
Askey2020[watts]
[watts]router. Measurement
router. interval
Measurement
val60
of ofmin.
60 min.

Figure 25.Electrical
Figure25.
Figure 25. Electricalcurrent
Electrical currentconsumption
current consumptionpattern
consumption pattern
patternofof
a aMidea
of 600
a Midea
Midea [watts]
600
600 washing
[watts]
[watts] machine.
washing
washing Measure-
machine.
machine. Meas-
Meas-
ment interval
urement
urement of 60
interval
interval ofmin.
of 60
60 min.
min.

Figure 26.
Figure26. Simultaneous
26.Simultaneous electrical
Simultaneouselectrical
electrical current
current consumption
consumption
current consumption profiles
profiles of
of appliances
of appliances
profiles under
under
appliances study
study
under in
in one
study one
inhour
one
hour
hour
of of
of consumption.
consumption.
consumption. The
The arithmetic
arithmetic
The arithmetic average
averageaverage of
of consumption
is 675isis
consumption
of consumption 675
675[watts/h].
[watts/h].
[watts/h].
Energies 2024, 17, 1452
Figure 26. Simultaneous electrical current consumption profiles of appliances under study 14
inofone
19
hour of consumption. The arithmetic average of consumption is 675 [watts/h].

Figure 27. Energy

Figure 27. Energyconsumed
consumedsimultaneously
simultaneouslybybyappliances under
appliances study
under in one
study hour
in one of consumption,
hour of consump-
tion, “Energy
“Energy Budget”Budget” in the campaign.
in the second second campaign. The arithmetic
The arithmetic average ofaverage of consumption
consumption is 675
is 675 [watts/h].
[watts/h].
5. Discussion of the Results
The appliances
5. Discussion of the that we consider as “critical path” for modern human survival are
Results
refrigerator, washingthat
The appliances machine, dishwasher,
we consider television
as “critical path”and
for notebook. The standby
modern human func-
survival are
tion of some appliances is also incorporated, such as cell phone chargers, Wi-Fi routers,
refrigerator, washing machine, dishwasher, television and notebook. The standby func-
printers, scanners and others. The results of both measurement campaigns are presented in
tion of some appliances is also incorporated, such as cell phone chargers, Wi-Fi routers,
Tables 2 and 3.
printers, scanners and others. The results of both measurement campaigns are presented
in Tables
Table 2 andof3.the first measurement campaign.
2. Results
A. Fridge.
Accumulated
Average Consumption Carbon
At the top of the Footprint
ranking Accumulated
of which appliance consumes the most connection time and
Appliance Consumption of Carbon
(watts/h) (kg CO2 )
electricity is the refrigerator Consumption
(it consumes 24 h(watts/h)
a day). TheFootprint
refrigerator represents
(kg of CO2 )
30.6% of the electricity consumption in our homes. It has approximately a consump-
Notebook 44.2 0.0060996 44.2 0.0060996
Router tion of 250–500 [watts/day]
6.37 0.00087906 or 15 [kw/month].
50.57 0.00697866
TV 45.25 0.0062445 95.82 0.01322316
Smartphone charger 3.95 0.0005451 99.77 0.01376826
Fan 42.8 0.0059064 142.57 0.01967466
Fridge 127.7 0.0176226 270.27 0.03729726
Washing machine 55.6 0.0076728 325.87 0.04497006
Oven 216.2 0.0298356 542.07 0.07480566
Brand stove 174.7 0.0241086 716.77 0.09891426
Brand heater 82.4 0.0113712 799.17 0.11028546
Brand kettle 51 0.007038 850.17 0.11732346

Table 3. Results of the second measurement campaign.

Accumulated
Average Consumption Carbon Footprint Accumulated
Appliance Consumption of Carbon
(watts/h) (kg CO2 ) Consumption (watts/h)
Footprint (kg of CO2 )
Toaster 30.3 0.0041814 30.3 0.0041814
Brand kettle 289.1 0.0398958 319.4 0.0440772
Fridge 90.5 0.012489 409.9 0.0565662
Hair dryer 74.2 0.0102396 484.1 0.0668058
Brand juicer 58.3 0.0080454 542.4 0.0748512
Router 5.24 0.00072312 547.64 0.07557432
Washing machine 127.5 0.017595 675.14 0.09316932

A. Fridge.
At the top of the ranking of which appliance consumes the most connection time
and electricity is the refrigerator (it consumes 24 h a day). The refrigerator repre-
sents 30.6% of the electricity consumption in our homes. It has approximately a
consumption of 250–500 [watts/day] or 15 [kw/month].
Energies 2024, 17, 1452 15 of 19

B. TV.
In second place, of the appliances that consume the most electricity is the television
with 10 h of on time. Television consumes 12.2% of total energy expenditure. You
should avoid using stand-by and turn it off completely when not in use. Its con-
sumption will depend on the type of screen, for example, a 32-inch LCD screen is
115 [watts/h], while plasma is around 300 [watts/h].
C. Washing machine.
The washing machine represents 11.8% of the total electricity consumption, it is in
third position in the ranking of which appliance consumes the most energy. If we
use short programs and cold washes, electricity can be saved. Average consumption
is approximately 255 [watts/h].
D. Consumption “vampire” or Stand-by.
The consumption of cell phone chargers, Wi-Fi routers, and music equipment turned
off but connected, among others, is known as Stand-by. The well-known standby
is not recommended as it seems, it consumes a total of 10.7% of electricity, much
more than a stationary or laptop computer. Avoiding this would mean considerable
savings on a residential user’s monthly bill.
E. Electric oven.
The oven accounts for 8.3% of the energy expenditure. If it is kept clean of fat,
avoiding opening it during cooking and turning it off beforehand to take advantage
of the residual heat will help reduce consumption. It consumes approximately
between 1800 and 2200 [watts] per hour depending on the model and conditions of
use and if it is used daily for cooking, it will occupy the first place on the list with an
average consumption of 60 [kw/month].
F. Computer.
The computer occupies 6th place on the list of which appliance consumes the most
energy; 7.7% is what the computer consumes in electricity. Energy consumption can
be reduced if we lower the screen lighting, remove the screensaver, and do not have
the peripherals connected (Wi-Fi speakers, printer, scanner, to name a few).
G. Dishwasher.
It consumes 6.1% of electricity. Using short programs and at a lower temperature
can save energy, as well as charging it to the maximum. Per hour, it consumes about
246 [watts].
H. Dryer.
Finally, we find the dryer, which according to this study on which appliances con-
sume the most electricity, we know consumes 3.3% of electricity, although it could
be an incorrect figure since only 28.3% of the population has a dryer in their home
(Chile, Undersecretary of Energy and Fuel 2022). The approximate consumption of
the dryer during one hour is 2700 [watts].
Although the above were the appliances that are most noticeable in our monthly bill,
we must not forget about the rest of the appliances, which can consume more Watts in an
hour, but of course, if we do the annual calculation, they are not the ones that consume the
most, such as air conditioning or ceramic hob burners.
• Air conditioning: 690 [watts/h].
• Vacuum cleaner: 675 [watts/h].
• Blender: 200 [watts/h].
• Mini music component: 75 [watts/h].
• Juicer: 50 to 200 [watts/h].
• Vitroceramic (one stove): 1200–2000 [watts/h].
• Fryer: 1000 [watts/h].
On the other hand, the results of the validations of the Deep Learning model, used in
the classification of household appliances, are shown in Figure 28. The results indicated in
Energies 2024, 17, 1452 16 of 19

Energies 2024, 17, x FOR PEER REVIEW 17 of 20

this article are better than those reported in [9,18–21], given that an average F1 score of 97%
was achieved.

Figure 28. Results

Figure 28. Results obtained
obtained from
from the
thevalidation
validation of
of the
theDeep
DeepLearning
Learningmodel.
model.

6. Conclusions
6. Conclusions
Based on the results of both campaigns as a whole presented in this article and
Based on the results of both campaigns as a whole presented in this article and con-
considering the current profile or pattern of each device under study, it was possible to
sidering the current profile or pattern of each device under study, it was possible to deter-
determine the average electrical energy consumption during one hour of operation. The
mine the average electrical energy consumption during one hour of operation. The use of
use of several electrical appliances in simultaneous activity was also considered, thus
several electrical appliances in simultaneous activity was also considered, thus determin-
determining the total value of the average energy consumed (area under the curve) and
ing the total value of the average energy consumed (area under the curve) and used by
used by said appliances in that same time range. As a consequence of the above, the “Energy
said appliances in that same time range. As a consequence of the above, the “Energy
Budget” required for the operation of residential demand can be determined; on the one
Budget” required for the operation of residential demand can be determined; on the one
hand, it can determine future billing on the other hand, it will allow knowing how much
hand, itbe
should can determine
generated future
with billing on the other
Non-Conventional hand, itEnergy
Renewable will allow knowing
systems in the how much
Harvest
should be generated with Non-Conventional Renewable Energy
of Photovoltaic-Wind Energy (Winter-Summer) for home supply. Also, it is possible to systems in the Harvest
of Photovoltaic-Wind
establish Energy (Winter-Summer)
the efficient administration for home
of home electrical supply.
energy as aAlso,
resultitof is the
possible
precise to
establish the
knowledge ofefficient
the averageadministration
consumption of per
home electrical
hour, energy
a fact that as athe
allows result of the precise
determination of
knowledge of the average consumption per
the energy budget of a home for a day, a month, or a year. hour, a fact that allows the determination of
the energy budget of a home for a day, a month, or a year.
Considering the two measurement campaigns indicated in this document, together
Considering
with the the two measurement
profile recognition campaigns
current consumption indicated inappliances,
of household this document, whichtogether
allows
with the profile recognition current consumption of household
the separation of electrical appliances into three clusters (cluster 1 of appliances appliances, which allows that
the separation
transform of electrical
electrical energyappliances
into heat, into three
cluster 2 ofclusters (cluster
appliances 1 oftransform
that appliances that trans-
energy into
form electrical
motor movement, energy
and into heat,
cluster cluster
3 of 2 of appliances
appliances that transformthat transform
electrical energy
energy intointo motor
data),
movement, and cluster 3 of appliances that transform electrical energy
it is possible to make decisions in real-time to turn on and off appliances and to correctly into data), it is pos-
sible to make decisions in real-time to turn on and off appliances and
discriminate the appliances that should really be in use from those that should not (switched to correctly discrim-
inate the
energy appliances that should really be in use from those that should not (switched en-
management).
ergyAdditionally,
management). the use of public cloud services for the implementation of the artificial
Additionally,
intelligence models the use of public
described in this cloud
documentservices for the implementation
is a significant improvementof theprevious
over artificial
intelligence
works, givenmodels
that theirdescribed
use allowsin this document
obtaining data is on
a significant improvement
the electrical over previ-
current consumption
ous
of works,
each given
device that theirsecurely
in real-time, use allows andobtaining
with a high data on theofelectrical
degree availability current
(average consump-
SLAs
tion of each device in real-time, securely and with a high degree
of 99.9% availability). These benefits allow decisions to be made in real-time with high of availability (average
SLAs of 99.9%
levels certaintyavailability).
(average F1 These
valuebenefits
of 97%), allow
mainlydecisions
becauseto they
be made helpin toreal-time
have a better with
high levels of
knowledge ofenergy
certainty (average
(watt per hour F1 concept).
value of 97%), mainly because they help to have a
better knowledge of energy (watt per hour concept).
7. Future Work
7. Future
Fault Work
Detection: It is expected that as a result of the improvement in the sampling rate
of theFault
measurements,
Detection: considering
It is expectedthethat
behavior and long-time
as a result variability ofinthe
of the improvement thehousehold
sampling
appliances being measured,considering
rate of the measurements, any difference in the current
the behavior or voltage variability
and long-time pattern, fundamentally
of the house-
of theappliances
hold resistive, inductive or some any
being measured, non-linear
differencebehavior,
in the means
currentanorevident
voltageanomaly which
pattern, funda-
could be detected through a model and treated in predictive maintenance
mentally of the resistive, inductive or some non-linear behavior, means an evident anom- programs. In
other words, any sudden change or change out of the current profile pattern
aly which could be detected through a model and treated in predictive maintenance pro- would be
detected
grams. Inasother
an electrical
words, anyfailure of the
sudden device,
change orand the out
change userofcould be informed
the current profileby their
pattern
mobile phone in real time. Along with this, if the measurement campaign
would be detected as an electrical failure of the device, and the user could be informed by were more
extensive,
their mobileforphone
example, for time.
in real 12 months,
Alongitwith
would beifpossible
this, to obtain the
the measurement annual were
campaign variability
more
of each electrical device from the Deep Learning model generated.
extensive, for example, for 12 months, it would be possible to obtain the annual variability
Use and application in NCRE programs (Non-Conventional Renewable Energy):
of each electrical device from the Deep Learning model generated.
Electronic circuit improvements can be made to obtain greater resolution of the total
Use and application in NCRE programs (Non-Conventional Renewable Energy):
Electronic circuit improvements can be made to obtain greater resolution of the total en-
ergy consumption budget with all the necessary home appliances. This knowledge of the
energy density of total annual consumption in the home may allow determining how
Energies 2024, 17, 1452 17 of 19

energy consumption budget with all the necessary home appliances. This knowledge of
the energy density of total annual consumption in the home may allow determining how
much energy should be obtained or generated through NCRE and/or stored in batteries,
allowing intelligent and efficient energy management, carrying out billing of the electrical
distribution network to zero, and if there is a surplus, it can be injected into the network.
Having a positive difference could lead to the sale of energy to the distribution company,
satisfying the electrical consumption needs of a home and saving energy savings.

Author Contributions: Introduction, V.H.-F.; Literature review, V.H.-F., F.U.-V. and D.C.; Methodology
and system development, V.H.-F., F.U.-V., C.E.-I. and L.G.-S.; Experimental development, J.T.-R. and
F.A.-R.; Discussion of results, F.U.-V.; Conclusions, V.H.-F., D.C. and L.G.-S.; Future works, F.U.-V.
and C.E.-I. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: https://drive.google.com/drive/folders/1AkS88ZBL543VhOhUxBJzu-
INN0LfweH2?usp=drive_link (accessed on 16 December 2023).
Acknowledgments: Thanks to the Metropolitan Technological University, in particular, to the PIRD
Digital Radio Communication Research Program, our alma mater and to our loved ones, for all the
support provided during this research.
Conflicts of Interest: The authors declare no conflicts of interest.

Abbreviations and Symbols

The following abbreviations and symbols are used in this manuscript:
HEMS Home Energy Management System
HAN Home Area Network
NILM Non-intrusive Load Monitoring
ILM Intrusive Load Monitoring
IoT Internet of Things
WSN Wireless Sensor Network
SH Smart Home
SP Smart-Plug
SS smart-socket
AMR Automatic Meter Reading
F1 F1 Score measure of predictive performance
PLC Power Line Carrier
VAC Voltage Altern Current
SGD Stochastic Gradient Descent
SLA Service Level Agreement
NCRE Non Conventional Renewable Energy
vϕ Statistical representation of the behavior of the time series data
C Classification of an appliance

References
1. Caldera, M.; Hussain, A.; Romano, S.; Re, V. Energy-Consumption Pattern-Detecting Technique for Household Appliances for
Smart Home Platform. Energies 2023, 16, 824. [CrossRef]
2. Lin, Y.-H.; Tsai, M.-S. An advanced home energy management system facilitated by nonintrusive load monitoring with automated
multiobjective power scheduling. IEEE Trans. Smart Grid 2015, 6, 1839–1851. [CrossRef]
3. Hassan, T.; Javed, F.; Arshad, N. An empirical investigation of V-I trajectory based load signatures for non-intrusive load
monitoring. IEEE Trans. Smart Grid 2013, 5, 870–878. [CrossRef]
4. Ridi, A.; Gisler, C.; Hennebert, J. A Survey on Intrusive Load Monitoring for Appliance Recognition. In Proceedings of the 2014
22nd International Conference on Pattern Recognition, Stockholm, Sweden, 24–28 August 2014; pp. 3702–3707.
5. Ulloa, F.; García-Santander, L.; Carrizo, D.; Hurtado, C. Towards a Home Energy Management Model Through a Coordinator of
Smart Sockets. Latv. J. Phys. Tech. Sci. 2018, 55, 35–43. [CrossRef]
6. Pawar, P.; Vittal, K.P. Design of smart socket for power optimization in home energy management system. In Proceedings of the
2017 2nd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT),
Bangalore, India, 19–20 May 2017; pp. 1739–1744.
Energies 2024, 17, 1452 18 of 19

7. Villanueva, M.L.G.; Dumlao, S.M.G.; Reyes, R.S.J. Appliance recognition system for ILM using AGILASx—Dataset of common
appliances in the Philippines. In Proceedings of the 2016 Cloudification of the Internet of Things (CIoT), Paris, France, 23–25
November 2016; pp. 1–5.
8. Cannone, C.; Hoseinpoori, P.; Martindale, L.; Tennyson, E.M.; Gardumi, F.; Croxatto, L.S.; Pye, S.; Mulugetta, Y.; Vrochidis, I.;
Krishnamurthy, S.; et al. Addressing Challenges in Long-Term Strategic Energy Planning in LMICs: Learning Pathways in an
Energy Planning Ecosystem. Energies 2023, 16, 7267. [CrossRef]
9. Ulloa-Vásquez, F.; García-Santander, L.; Carrizo, D.; Heredia-Figueroa, V. Intelligent electrical pattern recognition of appliances
consumption for home energy management using high resolution measurement. IEEE Lat. Am. Trans. 2021, 20, 326–334.
[CrossRef]
10. Suryadevara, N.K.; Biswal, G.R. Smart Plugs: Paradigms and Applications in the Smart City-and-Smart Grid. Energies 2019,
12, 1957. [CrossRef]
11. Cruz, C.; Palomar, E.; Bravo, I.; Gardel, A. Towards Sustainable Energy-Efficient Communities Based on a Scheduling Algorithm.
Sensors 2019, 19, 3973. [CrossRef] [PubMed]
12. Khalifeh, A.; Darabkh, K.A.; Khasawneh, A.M.; Alqaisieh, I.; Salameh, M.; AlAbdala, A.; Alrubaye, S.; Alassaf, A.; Al-HajAli, S.;
Al-Wardat, R.; et al. Wireless Sensor Networks for Smart Cities: Network Design, Implementation and Performance Evaluation.
Electronics 2021, 10, 218. [CrossRef]
13. Alam, T. Cloud-Based IoT Applications and Their Roles in Smart Cities. Smart Cities 2021, 4, 1196–1219. [CrossRef]
14. Jaradat, A.; Lutfiyya, H.; Haque, A. Appliance Operation Modes Identification Using Cycles Clustering. arXiv 2021,
arXiv:2101.10472.
15. Obushevs, A.; Oleinikova, I.; Mutule, A. Demand side management platform for HAN flexibility estimation with agent control.
In Proceedings of the 2016 13th International Conference on the European Energy Market (EEM), Porto, Portugal, 6–9 June 2016;
pp. 1–5.
16. Ullah, M.; Javaid, N.; Khan, I.; Mahmood, A.; Farooq, M. Residential Energy Consumption Controlling Techniques to Enable
Autonomous Demand Side Management in Future Smart Grid Communications. In Proceedings of the 2013 Eighth International
Conference on Broadband and Wireless Computing, Communication and Applications, Compiegne, France, 28–30 October 2013;
pp. 545–550.
17. Saleem, M.U.; Usman, M.A.; Politis, C. Design, Deployment and Performance Evaluation of an IoT Based Smart Energy
Management System for Demand Side Management in Smart Grid. IEEE Access 2022, 10, 15261–15278. [CrossRef]
18. Chen, Y.-C.; Ko, Y.-L.; Peng, W.-C. An Intelligent System for Mining Usage Patterns from Appliance Data in Smart Home
Environment. In Proceedings of the 2012 Conference on Technologies and Applications of Artificial Intelligence, Tainan, Taiwan,
16–18 November 2012; pp. 319–322.
19. Guillemin, S.; Ha, D.L. Fuzzy and parametric method for self-configuration of home energy manager system. In Proceedings of
the 2013 IEEE Grenoble Conference, Grenoble, France, 16–20 June 2013; pp. 1–6.
20. Devlin, M.A.; Hayes, B.P. Non-Intrusive Load Monitoring and Classification of Activities of Daily Living Using Residential Smart
Meter Data. IEEE Trans. Consum. Electron. 2019, 65, 339–348. [CrossRef]
21. Fahim, M.; Kazmi, S.M.A.; Khattak, A.M. ApplianceNet: A neural network based framework to recognize daily life activities and
behavior in smart home using smart plugs. Neural Comput. Appl. 2022, 34, 12749–12763. [CrossRef]
22. Codispoti, J.; Khamesi, A.R.; Penn, N.; Silvestri, S.; Shin, E. Learning from Non-experts: An Interactive and Adaptive Learning
Approach for Appliance Recognition in Smart Homes. ACM Trans. Cyber-Phys. Syst. 2022, 6, 1–22. [CrossRef]
23. Du, S.; Li, M.; Han, S.; Shi, J.; Li, H. Multi-Pattern Data Mining and Recognition of Primary Electric Appliances from Single
Non-Intrusive Load Monitoring Data. Energies 2019, 12, 992. [CrossRef]
24. Gajowniczek, K.; Zabkowski,
˛ T. Data Mining Techniques for Detecting Household Characteristics Based on Smart Meter Data.
Energies 2015, 8, 7407–7427. [CrossRef]
25. Yu, Z.; Fung, B.C.M.; Haghighat, F. Extracting knowledge from building-related data—A data mining framework. Build. Simul.
2013, 6, 207–222. [CrossRef]
26. Gao, Y.; Yang, H. Non-Intrusive Load Identification by Fuzzy Cluster Analysis Based on Active Power. In Proceedings of the 2012
IEEE PES Asia-Pacific Power and Energy Engineering Conference, Shanghai, China, 27–29 March 2012; pp. 1–4.
27. Ridi, A.; Gisler, C.; Hennebert, J. ACS-F2—A new database of appliance consumption signatures. In Proceedings of the 2014 6th
International Conference of Soft Computing and Pattern Recognition (SoCPaR), Tunis, Tunisia, 11–14 August 2014; pp. 145–150.
28. Jiang, L.; Luo, S.; Li, J. Intelligent electrical event recognition on general household power appliances. In Proceedings of the 2014
IEEE 15th Workshop on Control and Modeling for Power Electronics (COMPEL), Santander, Spain, 22–25 June 2014; pp. 1–3.
29. Lima, W.S.; Souto, E.; Rocha, T.; Pazzi, R.W.; Pramudianto, F. User activity recognition for energy saving in smart home
environment. In Proceedings of the 2015 20th IEEE Symposium on Computers and Communication (ISCC), Larnaca, Cyprus, 6–9
July 2015; pp. 751–757.
30. Sun, Q.; Li, H.; Ma, Z.; Wang, C.; Campillo, J.; Zhang, Q.; Wallin, F.; Guo, J. A Comprehensive Review of Smart Energy Meters in
Intelligent Energy Networks. IEEE Internet Things J. 2015, 3, 464–479. [CrossRef]
31. Salihagic, E.; Kevric, J.; Dogru, N. Classification of ON-OFF states of appliance consumption signatures. In Proceedings of the
2016 XI International Symposium on Telecommunications (BIHTEL), Sarajevo, Bosnia and Herzegovina, 24–26 October 2016;
pp. 1–6.
Energies 2024, 17, 1452 19 of 19

32. Klemenjak, C.; Elmenreich, W. On the applicability of correlation filters for appliance detection in smart meter readings. In
Proceedings of the 2017 IEEE International Conference on Smart Grid Communications (SmartGridComm), Dresden, Germany,
23–27 October 2017; pp. 171–176.
33. Gupta, A.; Murarka, P. A Ubiquitous Sensor Network for Domestic Energy Monitoring and Energy Consumption Optimization.
In Proceedings of the 2012 IEEE International Conference on Green Computing and Communications, Besancon, France, 20–23
November 2012; pp. 58–65.
34. Baier, C.R.; Hernández, J.C.; Wheeler, P. Measurements, Predictions, and Control in Microgrids and Power Electronic Systems.
Sensors 2023, 23, 4038. [CrossRef]
35. Ben Slama, S.; Mahmoud, M. A deep learning model for intelligent home energy management system using renewable energy.
Eng. Appl. Artif. Intell. 2023, 123, 106388. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.