Vous êtes sur la page 1sur 6

9th IFAC symposium on Control of Power and Energy Systems

9th Institute of Technology
9th IFAC symposium
December symposium
9-11, 2015.
on Control
of Power and
Indiaof Power and Energy
AvailableEnergy Systems
online Systems
at www.sciencedirect.com
Indian Institute
Indian Institute of Technology
of Technology
December 9-11, 2015. Delhi, India
December 9-11, 2015. Delhi, India
IFAC-PapersOnLine 48-30 (2015) 415–420
Overview of non-intrusive load monitoring and identification techniques
Overview of
of non-intrusive
non-intrusive load
load monitoring
monitoring and
and identification
identification techniques
E.J. Aladesanmi* KA Folly*.
E.J. Aladesanmi*
E.J. Aladesanmi* KA
KA Folly*.

* Electrical Engineering Department, University of Cape Town,
Private Bag **Rondebosch
Electrical 7701, (e-mails:
Engineering ereolaa@yahoo.co.uk;
Electrical Engineering Department, University
Department, University of Komla.Folly@uct.ac.za)
of Cape
Cape Town,
Private Bag Rondebosch 7701, (e-mails: ereolaa@yahoo.co.uk; Komla.Folly@uct.ac.za)
Private Bag Rondebosch 7701, (e-mails: ereolaa@yahoo.co.uk; Komla.Folly@uct.ac.za)
Abstract: Load monitoring and identification is a method of determining electrical energy consumption
and operation
Abstract: Load
Load condition of individual
monitoring and appliances
and identification
identification is based
is aa method
methodon theof analysis of composite
of determining
determining electrical load measured
electrical energy
energy from
and main power
operation meter
condition in
of a building.
individual They
appliances can supply
based on information
the analysis such
of as
and operation condition of individual appliances based on the analysis of composite load measured from type of
load load, electricity
measured from
the main
main power
powerdetail and in
meter theaarunning
in building.
building. conditions
They cancanofsupply
the appliances
supply information
information to both
such as consumer
as type
type of and the
of load,
load, utility.
The information
consumption canand
detail andbetheused
the to formulate
running conditions
conditions load of plan
of strategies to
the appliances
the appliances forboth
to optimal
both energy utilization.
the consumer
the consumer and the
and Load
the utility.
The techniques
information can be canused generally
to be
formulate grouped
load into
plan intrusive
strategies and
for non-intrusive
The information can be used to formulate load plan strategies for optimal energy utilization. Load energy load monitoring.
utilization. Load
Intrusive load
monitoring monitoring
techniques canprovides
can generally
generally accurate
be results and
be grouped
grouped intowould
into allowand
intrusive andeach individual appliance's
non-intrusive load energy
load monitoring.
Intrusive load to be communicated
monitoring provides to a
resultshub.and However,
would this
allow method
Intrusive load monitoring provides accurate results and would allow each individual appliance's energy is costly
individual because
appliance's of the
consumption to be communicated to a central hub. However, this method is costly because ofnon-
number of
consumption equipment
to be to be
communicated manufactured
to a and
central installed.
hub. This
However, has
method the
is introduction
costly because of of the
number load
of monitoring
equipment to system.
be Non-intrusive
manufactured and load monitoring
installed. This is
has cost effective
number of equipment to be manufactured and installed. This has prompted the introduction of non- theand convenient
introduction of means
of load monitoring
intrusive load since system.
load monitoring
monitoring it requires
system. lower equipment
Non-intrusive load due to fewer
load monitoring
monitoring is components
is cost
cost effective
effective andto install
and and smaller
convenient means
of loadrequirements.
load monitoring
monitoring sinceThis it
since itpaper is concerned
requires lower
lower equipment with thedue
equipment overview
due to fewerofcomponents
to fewer different load
components to identification
to install
install and
and smallerand
space techniquesThis
requirements. for energy
This paper management
paper is concerned focusing
is concerned with the
with theonoverview
overview load monitoring.
of different
of different load identification
load identification andand
monitoring techniques
techniques for
for energy
energy management
management focusing
focusing on non-intrusive
on identification,
non-intrusive load
load monitoring.
© 2015, IFAC
Keywords: E (International
nergy management, Federation
intrusive ofload
monitoring, Control)
load Hosting by Elsevier Ltd. All
non-intrusive loadrights reserved.
Keywords: E nergy management, intrusive load monitoring, load identification, non-intrusive
Keywords: Energy management, intrusive load monitoring, load identification, non-intrusive load monitoring. load monitoring.

 2012). In this paper, different load identification and
2012). In techniques
2012). In this paper,fordifferent
this paper, energy management
different load are reviewed
load identification
identification and
Intrusive load monitoring technique monitor appliances based and analysedtechniques
monitoring focusing on
techniques fornon-intrusive
for energy load monitoring.
energy management
management are
are reviewed
on supervisory
Intrusive load
Intrusive monitoringand
load monitoring data acquisition
technique (SCADA).
monitor appliances
monitor appliances based
basedTo and analysed
and analysed focusing
focusing on
on non-intrusive
non-intrusive load
load monitoring.
on monitor appliances, sensors
on supervisory control and data acquisition (SCADA). To
supervisory control and data are
acquisition installed
(SCADA). at each
adequately be monitored,
monitor appliances,
appliances, and sensors
once theare
sensors areappliance
installed at
at each
each 2.
appliancea to
appliance tosensor
be message, and
be monitored,
monitored, it immediately
and once
once the records the
the appliance
appliance load
data and
receives delivers them to the data centre
receives a sensor message, it immediately records the load
a sensor message, it immediately for further
records analyses.
the load
2.1 Intrusive Load Monitoring (ILM)
data intrusive
data and
and delivers
delivers load
them to
to the data system
the data centre
centre for is further
for comprehensive,
further analyses.
analyses. 2.1 Intrusive
2.1 Intrusive LoadLoad Monitoring
Monitoring (ILM) (ILM)
systematic, andload convenient method The Intrusive load monitoring system is a common metering
The intrusive
intrusive load monitoring
monitoring systemof is
system ismonitoring
comprehensive, load.
system that measures the energysystem
consumption of an appliance
systematic, the cost of installation and maintenance of The
The Intrusive load
load monitoring is
is aa common metering
systematic, and convenient method of monitoring load.
and convenient method of monitoring load. by
system that
measures the meters
to each
of an
in the
However, devices
However, the costsuch
the cost of asinstallation
of meters or sensors
installation may be a too
and maintenance
and maintenance of
of system that measures the energy consumption of an appliance
expensive. In addition, household.
by Therefore, it requires
by connecting power meters to each appliance in
connecting power meters entering
to each the house,
appliance thus
in the
measuring devices
devices such
such as as the
meters intrusive
or sensors
be aa tootoo system is referred to asit
technique household. Therefore, itintrusive. It provides
requires entering
entering theaccurate
house, thusresults,
thus the
expensive. is In
expensive. Intoo addition,
addition, the tointrusive
the implementload
intrusive loadin an ordinary
monitoring household.
requires the house, the
technique without a two-way communication device such as system isis referred
referred to high costs and
as intrusive.
intrusive. a complex
It provides
provides installation
accurate results,
technique is too complicated to implement in an ordinary
is too complicated to implement in an ordinary system
which usually
It accurate results,
smart meter.
household Also, customer acceptance
household without a two-way communication device such
without a two-way communication is another
device issue
however, imposing
imposing high
high costsand
costs anddata
and storage units
aa complex
complex for the
smart hinder
meter. the
Also, effective
customer implementation
acceptance is anotherof feedback
issue that house
which households
usually requires concerned.
wiring andIntrusive
which usually requires wiring and data storage units for
data load
storage monitoring
units for the
smart meter. Also, customer acceptance is another issue that techniques can be direct or indirect monitoring techniques
could hindersincethesome customers
effective will not allow of
implementation the feedback
intrusion house households concerned. Intrusive
house households concerned. Intrusive load monitoring load monitoring
could hinder the effective implementation of feedback (Yoshimoto,
in their
programs, privacy
since (Du
some customers
et al., 2010). Future
programs, since some customers will not allow the intrusion
will not allowload-monitoring
the intrusion
techniques can can2000),
be direct (Zeifman,
direct or and monitoring
or indirect
indirect Roth, 2011).
monitoring Direct
systems monitoring
(Yoshimoto, techniques which are also
(Yoshimoto, 2000), (Zeifman, and Roth, 2011). Direct
2000), (Zeifman, and known
Roth, as
2011). physical
in their should
their privacybe(Du
privacy based
(Du et on
et al.,the
al., strategies
2010). Future
Future forload-monitoring
minimizing the
intrusive signatures measure
systems of
systems monitoring
should be baseddevices
be based on the and
on the cost effectiveness
strategies for
for minimizing
minimizing of the
monitoring techniques
techniques whichtheare
which areelectrical
also known characteristics
known as
as physical of
monitoring systems. This has led to the introduction of each appliance's
intrusive signatures power demand
measure the (Marceau,
the electrical and
electrical Zmeureanu,
characteristics of
number of monitoring devices and cost effectiveness of
of monitoring devices and cost effectiveness ofnon-
the intrusive
appliance's power
demand (Marceau,
and Zmeureanu,
monitoring load monitoring
systems. This system
has led (NILM)
to the
monitoring systems. This has led to the introduction of non- (Zoha
introductionet al.,of2012),
non- 2000),appliance's
each (Breed, and power Delport,
demand 2001) . The physical
(Marceau, and intrusive
(Du et al.,
intrusive load2010),
load (Ehrhardt-Martinez,
monitoring system
system (NILM)
(NILM) et al, 2010).
(Zoha et Non-
al., 2012),
et al., 2012), 2000),
2000), (Breed,
signatures can beand
(Breed, and Delport,
generated a small
2001) .. The
The physical
device attached
physical intrusive
to the
(Du load monitoring techniques
(Du et al., 2010), (Ehrhardt-Martinez, et al, 2010). Non-
et al., 2010), (Ehrhardt-Martinez, make
et use
al, of
2010). a single
Non- power
signaturescordcan of
be an appliance
generated by a for
small measuring
signatures can be generated by a small device attached todevice the
attached energy
to the
intrusive load meter
load that is installed
monitoring techniques
techniques at make
the electrical
make use
use ofof aa service
single consumption
power cord by
of the
an appliances.
appliance Whenever
power cord of an appliance for measuring the energy measuring the appliance
the energy is
recording meterfrom that
meter where
that is the whole
is installed
installed at thehousehold
at the electrical energy
electrical service
service switched
consumption on, the
by device
the sends
appliances. signal
consumption by the appliances. Whenever the appliance is
Whenever to thethedata collector
appliance is
entrance from fromand where time of
where the usewhole
the wholeof each appliance
household energy
energyare indicating
switched on,on,the the
the device state
device sendsofsignal
sends the appliances.
signal to
to the dataThe
the data power
consumption Besides, NILM
consumption and time of use of each appliance are
and time of use isof low
each cost (in
appliance terms
are drawn by the
indicating the appliance
the operating can
operating statebeof
state ofcalculated
the appliances.by measuring
appliances. The
The power the
monitored. and maintenance),
Besides, NILM
NILM is is itlowdoescost
low not require
cost (in
(in terms
termsan electromagnetic
drawn field
appliance generated
can be by
drawn by the appliance can be calculated by measuring the
by the calculated the flow
by of
measuring current
implementationinto the customer’s
implementation and maintenance), it does not require an
and maintenance), premisesit does whennot require
monitoring an through the wire.
electromagnetic field This
field technique
generated by
by the theprovides
flow of of accurate
energy consumption
intrusion into
into the of different
the customer’s
customer’s appliances
premises when(Zoha
when et al.,
monitoring measurement,
through the the but it is not
wire. Thiscosttechnique
This effective (Zeifman,
technique provides and
provides Roth,
energy consumption
consumption of of different
different appliances
appliances (Zoha (Zoha et et al.,
al., measurement,
measurement, but it is not cost effective (Zeifman, and Roth,
but it is not cost effective (Zeifman, and Roth,
Copyright 2015 IFAC
2405-8963 © 2015, IFAC (International Federation of Automatic Control)415Hosting by Elsevier Ltd. All rights reserved.
Peer review©under
Copyright 2015 responsibility
IFAC of International Federation of Automatic
415 Control.
Copyright © 2015 IFAC
10.1016/j.ifacol.2015.12.414 415
December 9-11, 2015. Delhi, India E.J. Aladesanmi et al. / IFAC-PapersOnLine 48-30 (2015) 415–420

2011). On the other hand, indirect intrusive load monitoring the system is in operation and has not been overcome by
(IILM) techniques measure non-electrical characteristics, radio or customer inference. Apart from the aforementioned
from which each appliance's power demand is inferred. There applications, NILM system applications include load
are three forms of indirect intrusive load monitoring forecasting and rate forecasting (Ehrhardt-Martinez, 2010).
technique: appliance tagging, ambient sensors and
conditional demand analysis. Appliance tagging is the The main idea behind the NILM system is to determine the
modification of smart appliance such that a tag displays a energy consumption of individual appliances in a building
unique signal when the appliance turns on or off. based on the analysis of the aggregated data measured from a
single meter outside the building (Zoha et al., 2012). The
total aggregated measurement of the household energy
2.2 Non- intrusive Load Monitoring (NILM) consumption of the individual appliance from the main meter
can be estimated according to equation (1) (Zoha et al.,
Non-intrusive load monitoring is a convenient means of 2012), (Wang and Zheng, 2012):
determining the energy consumption and the state of
operation of individual appliances based on analysis of the n
aggregate load measured by the main power meter in a pt   i 1
pi (t )  et (1)
building. NILM is a process of analysing changes in the
voltage and current going into a building and deducing what where;
appliances are used in the building as well as their individual pt is the aggregated household energy consumption, pi(t) is
energy consumption. It is called non-intrusive because it the individual appliance energy consumption of at a given
does not require intruding into the house or consumer period of time and et is the measurement errors and line loss.
premises when measuring the power consumption of different Electrical appliances display a specific or unique
appliances. Smart meters with NILM technology are used by characteristic during the normal working cycle. This
utility companies to survey the specific uses of electric power characteristic is known as appliances signatures, which may
in different homes. NILM is considered a low cost alternative contain information about values of electrical parameters
to intrusive monitoring techniques. The idea of analysing the during steady-state or transient states. This information
power flow to determine household appliances and report on enables proper monitoring and identification of a specific
their operating condition started when George W. Hart was appliance from the aggregated household energy
collecting and analysing load data as part of a residential consumption measurement. The power signal from active
photo-voltaic system studied (Ehrhardt-Martinez, 2010). The appliances aggregate on the main meter is shown in Figure 1.
basic monitoring principle is to recognize a step change in Domestic appliances can be classified into the following
active and reactive power in the total load produced by categories based on the operational state or working principal
altering the operating state of the different customer’s of the appliances: two-state appliances, multi-state
appliances. Recently, Électricité de France (EDF) developed appliances; continuously-varying power appliances;
a new approach which consists of Hidden Markov Model permanent consumer appliances (Zoha et al., 2012),
(HMM) in an attempt to recognize the logical and (Zeifman, and Roth, 2011), (Baranski, and Voss, 2003),
chronological switch-on and off of different appliances. An (Norford, and Leeb, 1996), single measurement (Zeifman,
HMM consists of states and transitions between appliances and Roth, 2011).
switched on and off (Phial, 1998). Monitoring of individual
utility customers for the purpose of energy audit is an
important application of NILM system. A NILM can be
installed temporarily at the customer’s request in order to
analyse the characteristic of the appliances which can be used
in suggesting ways of reducing consumption and costs. A
second audit is often valuable to confirm the savings
resulting from conservation measures. Another application of
NILM is a power monitoring for appliance failure analysis or
security purposes. Faulty appliances can be detected by their
unusual power consumption or duty cycles. For instance,
irregularity in power consumption of refrigerator which was Fig.1: Power vs time plot of the total appliance data from
ON almost all of the time can easily be detected and replaced. Two-state appliances are those appliances that have only two
In terms of security, a vacation home which is unoccupied for states of operations either ON or OFF at a given point in
a long period of time can be monitored at a single point. The time. Examples of such appliances are table lamp, light bulb,
monitor could be programmed to automatically generate and toaster and water pump. Multi-state appliances are the
report appliances usage above or below the specified appliances that have more than two operating states. For
thresholds and send notification to the owner automatically instance, the appliances with on, off and standby modes are
via phone message. Furthermore, application of NILM multi-state appliances. They are also known as a Finite State
involves the verification of demand side management control. Machine (FSM). Examples of appliances include washing
Many electric utilities install appliance controller on machine, stove burner and dishwasher. Continuously-varying
deferrable load throughout their customer base to shed them power appliances are the appliances that do not consume a
during the period of peak power usage. NILM can verify that

December 9-11, 2015. Delhi, India E.J. Aladesanmi et al. / IFAC-PapersOnLine 48-30 (2015) 415–420 417

constant power. The monitoring and identification of such operational conditions by analysing the electrical signals (Du
appliances are usually the most difficult one since they et al., 2010). Residential load monitoring and identification
cannot be described by well-defined states. Examples include was first proposed in the 1980s by George Hart (Hart, 1992).
light dimmers and power tools. Permanent consumer The operating schedules of individual loads or groups of
appliances are the appliances that remain permanently ON loads are determined by identifying the times at which
with nearly constant active and reactive power consumption. electrical power measured change from one steady-state
Examples of this category of appliances include smoke value to another. This steady-state change is known as
detectors and telephone set. The Different appliances types events. The events of appliances usually correspond to the
based on their energy consumption pattern are shown in appliances either being turning ON or OFF and are denoted
Figure 2. by their magnitude and sign of real and reactive power
(Laughman, C. et al., 2003).

3.1 Appliance Signatures for Energy Disaggregation

Most appliances are identified by their distinctive

characteristics under certain condition. Appliance signature
is said to be a significant parameter of the total load that
provides information about the nature and operating
condition of individual appliance in the load (Zhang, et al
2011). Non-intrusive load monitoring system (NILMS) infers
Fig. 2. Different Appliances Types Based on their Energy individual loads of ON or OFF by checking signatures that
Consumption Pattern (Zoha, 2012) supply information on their activities. When an appliance
switched ON and OFF power, sample changed and a new
Data acquisition stage is a stage of obtaining aggregated load steady-state power level is established. The difference
measurements from the household at a given time interval so between the two state power levels is called an event. Event
that distinctive load patterns can be identified. During this signature can be fundamental frequency or harmonic
stage, voltage and current measurements are obtained and frequency signatures (Ehrhardt-Martinez, 2010), (Norford
processed to produce power metrics (real and reactive and Leeb, 1996).
power). The steady-state or transient changes in the pre-
processed signals, corresponding to the operation of 3.2 Fundamental Frequency Signatures
individual appliances are detected (Barges, et al., 2010).
Power meters are used to collect and measure the total At the utility, fundamental frequency can be used as
household load based on the time interval specified by the appliances signature by measuring the power, current or
researcher. When an appliance switched ON and OFF, power admittance of the total load and by considering the step
samples changed, and a new steady power level is change as signatures. The utility voltage varies over time and
established. This is characterized by a change in real and the actual voltage can vary within 10% of the nominal
reactive power. Appliance identification can be made by voltage supply. A linear appliance (i.e., the relationship
using an algorithm and comparing the energy consumption of between voltage and current is linear) plugged into the
each appliance with known features of a typical appliance in varying voltage source will also draw a current within the
the data (Barges, et al., 2010). varying voltage. The power consumption will then vary by
20%. Fundamentally, in NILM, changes detected in the total
load must provide information about events within the load.
3. LOAD IDENTIFICATION SYSTEMS Thus, the variations in power consumption do not provide an
ideal signature of the connected appliance for reasons
Vital information that enables effective energy management external to the load. Consequently, to abrogate the alteration
and consumption in the residential sector is provided through of voltage, the linear model suggests that admittance is better
accurate and reliable domestic appliance identification and than power and current as a signature. Admittance depends
monitoring (Du et al., 2010). However, apart from the lack of on the voltage of a linear appliance and is additive when the
effective coordination system between the loads and the appliances are wired in parallel. The load admittance Y (t) is
building management system, inefficient and feeble given by equation 2:
intelligent load identification and monitoring techniques has
made it difficult to identify and recognize the majority of the P (t )
Y (t )  (2)
electrical appliances that are connected to residential 2
V (t )
buildings (Du et al., 2010). By identifying different types of
loads and their power consumption from the current and Where; P (t) and V(t) are measured power and r.m.s. voltage.
voltage signatures, one can provide useful information for However, proper admittance is not the best choice of
load management system and therefore improve the system appliances signature, since it lacks engineering intuition
efficiency. Electrical appliances often present unique about the values to expect and their units. Therefore,
characteristic in the electrical signals (voltage, current and admittance in normalized power is preferred. The admittance
power) (Zoha et al., 2012). Such characteristics provide a in normalized power is given by equation 3:
viable means of identifying the type of appliances and their

December 9-11, 2015. Delhi, India E.J. Aladesanmi et al. / IFAC-PapersOnLine 48-30 (2015) 415–420

variation in active and reactive power can be used to identify

different types of loads. As the condition of the appliance
 120 
2 changes, the magnitudes of active and reactive power also
Pnormal ( t )  120 Y ( t )    P ( t ) (3) change. These values are then compared with the predefined
 V (t )  database values of load power. The Load can be identified
through the position in P-Q plane as shown in Figure 3. From
Equation 3 depicts the power if the utility provides a steady Figure 3, it can be seen that many appliances are located
voltage and the load obey a linear model. The value of 120 around the horizontal axis. The loads that are far from each
corresponds to the nominal voltage in the USA power other in the plot can be identified only by using reactive and
distribution system (Ehrhardt-Martinez, 2010). active power.
3.3 Harmonic Frequency Signatures

Harmonic currents produced by the appliance can be used as

a signature to identify the appliance. A linear model suggests
that the utility voltage waveform is directly proportional to
the current response. For instance, if the voltage waveform is
sinusoidal, the current response will also be sinusoidal,
although many appliances are obviously nonlinear in this
respect. At higher frequencies, notable current are generated
by many power electronics devices. Apart from the resistive
heaters and incandescent lights, virtually all appliances
produce mixed harmonic currents. Harmonic current
signatures might be very useful in identifying certain Fig.3: Real and reactive power of electric loads in residential
appliances that are too similar to distinguish using real and buildings in the P-Q plane (Perez, MNV, 2011)
reactive power (Ehrhardt-Martinez, 2010), (Nakano, 2006).
Signatures for appliance identification can generally be Using this approach, the success rate of loads identification
grouped into: steady-state signature and transient-state of large residential loads can be greater than 80% (Du et al.,
signature (Du et al., 2010). In this paper, only steady-state 2010). Other approaches that were proposed for appliance
signature will be discussed. identification based on changed in active and reactive power,
include the temporal motif mining approach (Shao, et al.,
3.4. Steady-State Signatures
2012), and a hidden Markov model (HMM) based on
The steady-state signatures of appliance are obtained from combined load model (Kolter, and Jaakkola, 2012). The
the difference between steady-state properties of the temporal motif mining approach is based on the steady-state
appliance operating conditions, estimated as the difference analysis and detects repeated sequences of power changes.
between the operating levels of the connected state (Ehrhardt- The purpose of the method is to identify the shortest repeated
Martinez, 2010), (Haasz, and Madani, 2014), (Perez, MNV, sequence of power changes, adding up to zero and
2011). The steady-state based techniques include techniques corresponding to full operation of cycle of each appliance.
based on the active and reactive power; techniques based on The hidden Markov model (HMM) identifies the presence of
the current waveform characteristics and harmonic and individual loads from total power consumption. The method
techniques based on V-I Trajectory. Due to space limitation, is used to obtain daily operating schedule (or time-of-day
the V-I Trajectory technique will not be discussed here. energy usage) and daily power consumption of household
Interested reader can read (Du et al., 2010), (Ehrhardt- appliances. The model can be used to identify
Martinez, 2010). thermostatically controlled automatic appliances (such as
fridge, air conditioner and water heater), fixed operation
3.4.1. Techniques Based on Active and Reactive Power humanly activated appliances (such as dishwasher and
washing machine) and usage dependent humanly activated
Active power is also called real power. It is the power devices (such as computer, microwave and printer). Since the
capable of performing useful work. It is also said to be the individual and combined power consumption waveforms of
conversion of electrical energy into another form of energy these appliances are composed of different steady-state power
such as mechanical work or heat. It is motivated by resistive levels and a steady-state power level is a stationary stochastic
loads where current and voltage components are in phase. process, the HMM based approach is a suitable modelling
Reactive loads such as coils and capacitors consume no approach to describe and to identify these appliances due to
electrical energy as they store it and return it back into the the strength of HMM which lies in the fact that its models are
source. This power is called reactive power and it generates a combination of stationary stochastic processes. The states
magnetic and electric fields. Reactive power is consumed of the HMM are used to model the stationary stochastic
due to the presence of reactance in a circuit such as motors, process and the state transition probability is used to describe
transformers or solenoids (Perez, MNV, 2011). The real the transition between the processes. Furthermore, (Kolter
power and the reactive power are measured and recorded and Jaakkola, 2012), (Ghahramani and Jordan, 1997)
such that the proposed Factorial Hidden Markov Model (FHMM)

December 9-11, 2015. Delhi, India E.J. Aladesanmi et al. / IFAC-PapersOnLine 48-30 (2015) 415–420 419

approach which is an extension of HMM. The approach can and power electronic loads. Current harmonics are suitable
be used to solve complex problems with multiple time series. for identifying power electronic loads since they are
In spite of the widely use of active and reactive power in nonlinear and generate current harmonics (Du et al., 2010),
appliances identification, the technique still has some (Jing, and Honggeng, 2012), (Prasad and Semwal, 2013),
limitations. Firstly, some appliances with the same power (Akbar and Khan, 2007), (Sultanem, 1991), (Cole and
consumption or low power appliances cannot be identified by Albicki, 2000), (Xu, and Dong, 2012). In order to improve
the technique. However, according to (Du et al., 2010), the the identification efficiency and accuracy, current harmonics
efficiency of load identification using active and reactive are often combined with real and reactive power methods and
power based technique can be improved using rule-based transient power methods for load identification (Laughman,
method. The algorithm uses meter data (measured energy C. et al., 2003).
consumption data), with assumptions about the consumer
attitude. In a scenario where two or more loads have similar
power demand levels, the algorithm uses decision analysis
approach to distinguish between them, based on the
assumptions regarding the usage of these appliances, such as
the time of day or the length of usage. Beside the
aforementioned limitation, another limitation to load 4.1. Advantages of NILM
identification based on real and reactive power is that the
Compared to intrusive load monitoring, NILM has a number
technique is based on steady-state power consumption. Thus,
of important advantages. Some of these advantages include
it cannot be used in real-time. However, some loads in
low cost of installation and maintenance due to fewer
residential buildings do not yield reliable steady-state
components to install, maintain and remove, greater
measurements (Du et al., 2010), (Lam, et al., 2006).
reliability and smaller space requirements. The ease of
installation will allow more appliances to be monitored in
3.4.2. Technique Based on Current Waveform Characteristic
more homes, providing broader data and in many cases more
and Harmonics
accurate data. In addition, unlike conventional intrusive
systems which are usually restricted to a certain number of
The best sets of information to identify appliances include
appliances, NILM system is not restricted or limited by
current and voltage, although it is not practically viable to
communication bandwidth, it only depends on the
compare the waveforms because of the computation
sophistication of the algorithms and hence can decompose the
limitation and sensitivity issues. Also, abstracting features
total load (Yoshimoto, 2000a, b) Also, being less expensive,
from the waveform for appliance identification is a challenge.
it can be placed at many sites, thereby reducing the biases
Thus, peak current, average current and r.m.s. current, current
that result from small samples. Furthermore, it can be used by
timing have been selected for load identification (Du et al.,
customers who would not allow the intrusion of power utility
2010). The current features in the appliance database are
into their premises. Finally, non-intrusive load monitoring
transmitted to a home server, which compares the feature of
equipment and wiring may provide a constant utility presence
known appliance with the received feature to identify an
to the customer resulting in a change in energy consumption
appliance’s model and activity. However, the method is
habits (Ehrhardt-Martinez, 2010).
generally affected by noise, calculation timing and variation
in voltage. Also, the technique cannot identify correctly the 4.2. Disadvantages of NILM
appliances with the same power level (Jing and Honggeng,
2012), (Ito, et.al., 2004). Current harmonic is mostly used for
identifying appliances with similar energy consumption Continuously variable appliance such as light dimmers and
(Simivasan, et al., 2006), (Yoshimoto, et al., 2000), (Morie, appliances that operates constantly like clocks cannot be
and Suga, 1991). Current waveforms are easy to be calculated detected by NILM technique. Also, NILM cannot distinguish
and represented. Thus, in order to improve the identification between electrically identical appliances such as two burners
efficiency and accuracy, they are often combined with P-Q of the same size on an electric stove. However, these
plane methods and transient power methods for load limitations can be alleviated by harmonic current signatures.
identification. Current harmonics can be calculated using Another disadvantage of NILM is that it has greater potential
measured or collected data with a high frequency of about 8, for undetected error. There is a greater potential for the
000 kHz sampling rate. Fast Fourier transforms (FFT) and reported data to contain significant errors since the total load
Short-time Fourier Transform (STFT) of current waveforms is disaggregated in software rather than the hardware of
are also used to analyse time-varying harmonic current. For separate sensors and communication channels. Moreover, the
single-phase loads, real and reactive power correspond to the software may not recognize unusual appliances that are not in
envelop in phase and quadrature current draw by the load the database. NILM has not been experimented on multistate
relative to the voltage. The short time Fourier transform appliances such as dishwashers and heat pumps (Ehrhardt-
computes estimates of the real, reactive and higher frequency Martinez, 2010).
components of the current (Laughman, C. et al., 2003). In
addition, power factor (PF) and harmonic distortion in the 5. CONCLUSIONS
current can also be used to identify loads in its steady-state
signature. PF can distinctly identify a purely resistive loads This paper presents an overview of various methods of
monitoring and identification of electrical appliances in

December 9-11, 2015. Delhi, India E.J. Aladesanmi et al. / IFAC-PapersOnLine 48-30 (2015) 415–420

residential areas. Based on the review, it is observed that no Lam, H.Y., Ting, K.H., Lee, W, K., Fu G.S.K. (2006). An Analytical
Understanding on Voltage-Current Curve of Electrical Load,”
single method can successfully monitor and identify
International Conference on Electrical Engineering (ICEE), 2006
appliances in a household. Although non-intrusive load Lam, H.Y., Fung, G. S. K., Leeb, W. K. (2007). A Novel Method to
monitoring is the most cost effective in terms of Construct Taxonomy of Electrical Appliances Based on Load
implementation and maintenance, it has several limitations Signatures. IEEE Trans. on Consumer Electronics, pp.653-660.
Laughman, C., Lee, K., Cox, R., Steven Shrew, Lee, S., Norford, L., and
such as not being able to detect directly continuously variable
Armstrong, P. (2003). Power Signature Analysis. Power and Energy
appliance such as light dimmers and appliances that operates Magazine, IEEE, Vol.1, Issue 2, DOI:
constantly like clocks. In addition, it cannot distinguish 10.1109/MPAE.2003.1192027, PP.56-63.
between electrically identical appliances and has greater Marceau, M.L., R. Zmeureanu, R. (2000). Non-Intrusive Load
Disaggregation Computer Program to Estimate the Energy
potential for undetected error.
Consumption of Major End Uses in Residential Buildings,”
PERGAMION, Energy Conversion and Management, Vol. 41,
Acknowledgment Morie H., and Suga, S., (1991). Power System Harmonics Prediction with an
This work is based on research supported partially by the Artificial Neural Network. IEEE International Symposium on Circuit
and System, Vol.2, pp.1129-1132.
National Research Foundation of South Africa UID: 85503; Nakano, Y. (2006). Non-Intrusive Electric Appliances Load Monitoring
UID: 83977. System Using Harmonic Pattern Recognition Performance Test
Results at Real Households. The Fourth International Conference
REFERENCES on Energy Efficiency in Domestic Appliances and Lighting
(EEDAL), London, pp. 477-488.
Akbar, M., and Khan, D. Z. A. (2007). “Modified Non-Intrusive Appliances
Norford, L.K., Leeb, S.B. (1996). Nom-Intrusive Electrical Load Monitoring
Load Monitoring for Nonlinear Devices,” IEEE International
in Commercial Buildings Based on Steady-State and Transient
Multitopic Conference, INMIC, pp.1-5.
Load-Detection Algorithms. ELSEVIER, Energy and Building Vol.
Baranski M., and Jürgen Voss, J. (2003). Non-Intrusive Appliances Load
24, pp. 51-64.
Monitoring Based on an Optical Sensor. IEEE Proceeding
Perez, MNV (2011). A Non-Intrusive Appliance Load Monitoring System
Power Tech Conference, Bologna, Vol.4, pp.1- 8.
for Identifying Kitchen Activities. A Thesis Submittedd in Partial
Barges, M., Goldman, E., Matthews, H.S., Lucio Soibelman, L. (2008).
Fulfillment of The Requirement for The Degree of Electrical
Training Load Monitoring Algorithms on Highly Sub-Metered Home
Engineering and Telecommunicateions, Department of
Electricity Consumption Data. Tsinghua Science and
Automation and System Technology Alto University School of
Technology, ISSN 1007-0214 65/67, Vol.13, pp.406-411.
Electrical Engineering , Espoo, Finland.
Barges, M.E., Goldman, E., Matthews, H.S., and Lucio Soibelman, L.
Phial, H. (1998). Non-Intrusive Appliance Load Monitoring Systems Based
(2010). Enhancing Electricity Audits in Residential Buildings with
on a Modern kWh-Meter,” VTT Energy, Thesis Submitted for Official
Nonintrusive Load Monitoring. Journal of Industrial
Examination for The Degree of Licentiate in Technology,
Ecology, Vol. 14, No.5, pp.1-15, Oct. 2010.
Breed, J.A., and Delport, G.J. (2001). Non-Intrusive Load Monitoring of
Prasad, R.S., and Semwal, S. (2013). A Simplified New Procedure for
Residential Appliances. Proceeding of the 7th Conference on
Identification of Appliances in Smart Applications. 2013 IEEE
Domestic Use of Energy, pp.33-38.
International System Conference (SysCon2013) Proceedings,
Cole, A. and Albicki, A. (2000). Non-Intrusive Identification of Electrical
Loads in a Three- Phase Environment Based on Harmonic
Shao, H., Marwash, M. and Ramarkrishnam, N. (2012). A Temporal Motif
Current. Proceedings of the 17th IEEE Conference on Instrumentation
Mining Approach to unsupervised Energy Disaggregation:
and Measurement Technology IMTC2000 Vol.1, pp.24-29.
Application to Residential and Commercial Buildings. 1st
Du, Y., Du, L., Lu, B., Harley, R. and Habetler, T (2010). A Review of
International Workshop on Non-Intrusive Load Monitoring, pp, 1-7.
Identification and Monitoring Methods for Electric Loads in
Simivasan, D. Ng, W.S and Liew, A.C. (2006). Neural- Network Based
Commercial and Residential Buildings. In Proc. of Energy
Signature Recognition for Harmonics source Identification. IEEE
Conversion Congress and Exposition (ECCE), 2010, IEEE D01: 10
Transaction on Power Delivery pp. 398-405.
1109/ECCE; 2010.5618423, pp. 4527-4533.
Sultanem, F. (1991). Using Appliance Signatures for Monitoring Residential
Ehrhardt-Martinez, K. Donnelly, K.A. and John. A. (2010). Advanced
Loads at Meter Panel Level. IEEE Transactions on Power Delivery
Metering Initiatives and Residential Feedback Programs:
Vol.6, pp1380-1385.
A Meta-Review for Household Electricity-Saving Opportunities.
Wang Z. and Zheng, G. (2012). Residential Appliances Identification and
Technical Report E105 for American Ccouncil for Energy Efficient
Monitoring by a Non-Intrusive Method,” IEEE Transaction on
Ghahramani Z. and Jordan, M.I. (1997). Factorial Hidden Markov Models.
Smart Grid, Vol.3, No. 1, pp. 80-92.
Machine Learning Vol.29, pp.245-273.
Xu, W., and Dong, M. (2012). Tracking Energy Consumptions of Home
Haasz V., and Madani, K. (2014). Advanced Data Acquisition and
Appliances Using Electrical Signature Data. IEEE General
Intelligence Data Processing. Published by River Publisher, Wharton,
Meeting on Power and Energy, pp.1-5.
TX, USA, chapter 3, pp.31- 58.
Yoshimoto, K., Nakano, Y., Amano, Y., and Kermanshahi, B. (2000a). Non-
Hart, GW. (1992). Non-Intrusive Appliances Load Monitoring”, Proceeding
Intrusive Appliances Load Monitoring System Using Neural Networks.
of the IEEE, Vol. 80. Issue 12, PP.1870-1891.
IEEE Proceedings of International Symposium on Circuits System,
Ito, M., Uda, R., Ichimura, S., Tago, K., Hoshi, T. and Matsushita, Y.
Vol. 2, pp.183-194.
(2004). A method of appliances detection based on Feature Power
Yoshimoto, K., Nakano, Y., Amano Y., and Kermanshaki, B. (2000b). Non-
Waveform. International Symposium on Appliance and the
Intrusive Appliances Load Monitoring System Using Neural
Internet Tokyo Japan, pp.291-294.
Network. ASCEEE Summer Study on Energy Efficiency in
Jing L., and Honggeng, Y. (2012). The investigation of Residential Load
Buildings, Pacific Grove, CA, USA, pp. 1-12.
Identification Technologies. Power and Energy Conference
Zeifman, M., and Roth, K. (2011). Non-Intrusive load Monitoring: Review
(APPEEC), 2012 Asia Pacific, DOI: 10:1109/APPEEC. 2012.
and outlook. IEEE Transactions on Consumer Electronics, Vol.57,
6307565, pp.1-4.
No.1, PP.76-84.
Kamat, S. R.. (2004). Fuzzy Logic Based Pattern Recognition Technique for
Zhang, P., Zhou, C., Stewart, B.G., Hepburn, D.M., Zhou W., and Yu, J.
Non-Intrusive Load monitoring. 2004 IEEE Region 10 Conference on
(2011). An Improved Non-Intrusive Load Monitoring Method for
TENCON, Vol.3, pp.528-530.
Recognition of Electric Vehicle Battery Charging Load, ELSEVIER,
Kolter, J. Z. and Jaakkola, T. (2012). Approximate Interface in Additive
Energy Procedia Vol. 12, pp.104-112
Factorial HMMs with Application to Energy Disaggregation,”
Zoha, A., Gluhak, A., Imran MA., and Rajasegarar, S. (2012). Non-Intrusive
Proceeding of the Fifteenth Conference on Artificial
Load Monitoring Approaches for Disaggregated Energy Sensing: A
Intelligence and Statistics, pp.1472-1482.
Survey,” Open Access Sensor ISSN 1424-8220, pp.16838-16866.