Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

journal homepage: www.elsevier.com/locate/rser

Non-technical loss analysis and prevention using smart meters MARK

Tanveer Ahmad
School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan, 430074, China

A R T I C L E I N F O A BS T RAC T

Keywords: In the countries such as Pakistan, for analyzing the losses and techniques in the power distribution and for
Non-technical loss mitigating, are the two active areas of research which is spread globally for increasing the accessibility of power
Advanced metering infrastructure irrespective of installing future generation equipment. As, the Technical Loss and the Non-Technical Loss are
Support vector machine accounted by the total energy losses. They both are also referred as TL and NTLs. In terms of the non-technical
Linear regression analysis
losses there are major financial losses for the utility companies present in the countries that are in the
developing stage. NTLs is the major cause for the additional losses and also it includes the part of damaging the
network that includes the infrastructure and network reliability reduction. This paper is subjected to
investigating the non-technical losses in terms of the power distribution systems. In addition to that, the
consumer energy consumption information is used for analyzing the NTLs from Rawalpindi region from the
different feeder source. The data of Low Voltage (LV) of the distribution network are focused more that consists
of commercial, industrial, residential and agricultural consumers by the use of KWh interval data which is
captured over a month using the smart meter infrastructure. The discussion of this review paper determines
analysis and prevention techniques of NTLs to safeguard from the illegal use of the electricity in the distribution
of electrical power system.

1. Introduction
In the electricity supply companies, the customer illegal use of
electricity are the main problems and the reason for the electrical theft
is due to the Non-Technical Losses (NTLs). These type of losses occur
because of tampering the meter, then the illegal connections, mal-
functioning of the meter, irregularities of billing and because of the
bills that are not paid. The NTLs problem is faced not just by the
developing countries instead it also includes the developed countries
like the United States of America. Taking United States as an example
for the problem of NTLs, the total annual revenue is estimated to range
from 0.5% to 3.5% [1]. Bangladesh, India, Iran, and Pakistan comes
under the category of developing countries, where the percentage of
NTL is average, and it ranges from 10% to 15%, whereas the United
States of America and the United Kingdom falls under the category of
developed countries. Thus, the estimated total annual revenue is
comparatively low in comparison with the developing countries [2].
The losses of worldwide transmission and Distribution (T & D) are
comparatively higher than the total generation capacity of the countries
such as France, Germany and UK. It is known that certain companies
are known for their loss which is over US $20 billion annually, just
because of the theft of electricity. On the other hand, the countries
similar to India faces lose that ranges to US$9 billion annually because,
of the theft of electricity [1]. The observed energy losses in Pakistan are
generated as 17.5% in the year 2012–2013 and the year of 2013–2014
it is observed as 16.9% and 0.6% is accounted for the improvement of
the losses.
The graph represents the line losses that Pakistan faced, which is
shown in Fig. 1 [3]. There are huge range of losses suffered by the
utilities because of the theft of electricity. Therefore, to save from the
effects of losses, it is important to reduce the process of NTL. The
power system components influences the power that is generated,
distributed and transmitted including the appliances of the customer.
The most difficult part of the electricity is the use of illegal electricity
consumption in real time. In such cases, the parameters that are used
to analyze the theft of electricity are, the economic, political, manage-
rial and criminal aspects. Different kinds of methods are described
such as the efficient management of NTLS and various electricity
distribution system are proposed, because of the problem that occurred
in NTLs in the form of electric utilities for securing the revenue. The
use of Advanced Metering Infrastructure (AMI) can reduce the NTLs in
an effective way, in which it makes the fraudulent activities complex
and the detection is easy. But, in such case, the cost will be much
expensive for those meters. For example, they are described in various
sectors like the commercial, residential and agricultural sectors [4–6].
The fraud identification of the research studies on several data mining
techniques and the prediction of it is carried out by the sector called as
electricity distribution sector. They include many types like statistical

E-mail address: tanveer.ahmad.pk1@gmail.com.

http://dx.doi.org/10.1016/j.rser.2017.01.100
Received 6 November 2015; Received in revised form 1 January 2017; Accepted 15 January 2017
1364-0321/ © 2017 Elsevier Ltd. All rights reserved.
T. Ahmad
Fig. 1. Total power system losses including (transmission and distribution) in Pakistan [3].
methods [48], Decision Tress [49], Artificial Neural Networks (ANNs)
[50], and then the Knowledge Discovery in the Database. By getting all
of these the load profiling is one of the most widely used pattern of
electricity and the customer over a period of time demand its pattern.
At the same time, the system called as e-metering will be used for
implementing the processing of the data, and will detect the abnormal
activities that are present at the load profile. Thus, it will indicate the
electricity theft. To identify the energy consumption of a customer, a
smart meter is used for this process, which produces additional
information when compared with the conventional energy meter. The
approach for grouping of illegal consumption and the process of
estimation will be based on the smart meter is proposed [51–56].
2. Power system losses
The losses of electric power system will be affecting the utilities and
it can be categorized into two types:
(1) The technical losses
(2) The Non-technical losses
The difference between the quantity of the energy delivered and the
quantity of the energy recorded is defined as the power losses, and they
are sold to the customers.
PLOSSES = PDELIVERED − PSOLD
The amount of energy can be represented as PLoss and the amount
of energy delivered can be expected as PDelivered. Whereas, the amount
of energy either recorded or sold will be determined as PSold.
2.1. Technical losses
It mainly consists of power dissipation in terms of electrical system
components. The components will be described as the transformers,
transmission and the distribution lines, and energy system’s measure-
ment unit. But, at the same time the losses are in technical terms and
are being calculated during the phases of design and construction.
Their power transmission and distribution networks are also men-
tioned. The values of the technical losses will be known by the utilities
[7–9].
2.2. Non-technical losses
The actions of external are caused by the Non-technical losses and
then these actions are follows: power system’s external actions, mainly
the theft of electricity, customer non-payment regards to the errors in
the meter reading and in maintaining the record. The NTL activities are
contributed and the factors can be shown in [10–13], and it can be
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
categorized into the following forms:
1. Line diversion is unauthorized: stealing by means of bypassing the
meters or the other way of making an illegal connections [13].
2. Line tapping is unauthorized: meters will be tampering and so the
lower rates of the consumption will be recorded by the meters [11].
3. The lack of quality and the meter reading will be inaccurate [13].
4. Electricity bill of the customer are not accurate [10].
5. The techniques are poor for the poor revenue collection [12].
6. With the help of the internal employees, the bill can be arranged
and the irregularities can be sorted, like the lower bills [12].
7. The non-metered supplies are not accurate when estimated. For
example, rail traction and public lighting [12].
8. Faulty meters cause loss [10].
9. Non-payment of electricity bills [12].
10. The protective equipment, cables, conductors and the switchgear
becomes loss if the equipment are damaged [11].
The electrical distribution business will help in the form of
detecting the NTL, and it is said to be the very important task; because
in Spain, represents the percentage of fraud based on the form of
energy and in respect with the total NTLs. The NTLs is about 35–45%.
There are so many fraud for the works and researches and they are
from the literature.
The researched based on fraud and for the detection of NTL are
(1) present in various fields [56–65]. But, the researches that are based on
the detection of NTLs in the power utilities is less though, the
percentage of the NTLs is comparatively greater [66–70]. The research
works are generally based on theory and the types of detection
techniques used are less. The electrical companies adopt to the current
methodology in terms of the NTLs detection. It is basically divided into
two kinds. They are listed below:
(1) The primary type is based on the making of situ inspections for the
wanted users, which is selected from the earlier zone.
(2) The secondary type is based on the user’s study and it contains the
null consumption at a certain point.
The primary issue that is faced in the primary alternative is that, it
wants vast numbers of the inspectors, which results to high cost. The
secondary alternative contains the problem of possibly detecting the
users only with the null consumption. But, for the customers who have
non-null consumption will be quite low when compared with their
consumption. In recent days, the techniques of the data mining [71,72]
will be applied to one or more fields and then the power utility will be
considered as an industry, where it meets its success [73–78].
574
T. Ahmad
Fig. 2. Flow of Power in advanced metering infrastructure.
3. Advanced metering infrastructure
The electrical grid will be named as the smart gird which contains
different variety of operational and energy measures, smart appliances,
smart meters, energy resources that are renewable and the efficiency of
energy resources. The important aspects of the smart grid will be
determined as the electronic power condition and the controlling
process of production and the distribution of electricity. The United
States of America follows a set of policy that supports the modern
nation’s electricity transmission and helps in maintaining the distribu-
tion system. Then, it can create reliability and can protect the
infrastructure of electricity. Further, it can meet the demands of the
growth of the future generation and can help in achieving the following
factors [79]:
(1) Integration of the distributed resources and the deployment of
the distributed resources and the generation, along with the
renewable resources.
(2) Usage of digital information will increase and the control
technology will increase the security, reliability and electric grid’s
efficiency.
(3) Along with the full cyber-security, the dynamic optimization of
the grid operations will take place.
(4) Energy efficient resources and the resources of demand-side, and
the demand response will be incorporated and developed.
(5) Combination of the consumer devices and the smart appliances.
(6) Providing the consumers with control options and timely infor-
mation.
(7) With the use of ‘smart’ technology, it can optimize the consumer
devices and other appliances physically. It is used for metering
the communicating concerns, its operations, and automation of
the distribution.
(8) The combination of advanced electrical storage and the techni-
ques of peak-shaving technologies along with the plug-in electric
and the hybrid electric vehicles will be deployed.
(9) Progress in the communication standards, the operating process
of the appliances and the electric grid connected equipment along
with serving the infrastructure of the grid.
(10) Gathering it and lowering the unnecessary barriers, will result in
adoption of smart grid technology, services and practices.
575
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
AMI stands for Advanced Metering Infrastructure, in which it is
determined as the primary step of process and moves towards the
smart grid. It has been well known for its acknowledgement for power
utilities. By using the advanced metering infrastructure, the automatic
system of the management, the enchanted control mechanism of the
low voltage distribution system can be achieved. They are being
operated for the additional capabilities and the integrated systems
with the use of the distribution automation system.
The smart grid infrastructure contains the integral part named as
the smart meter systems, in means of data collection and the data
communications. It is known for their modern type of the transmission
and the distribution with the aspects of electric grid. By functioning it,
it will be determined as the automatic electric power system and that, it
monitors and controls the activities of the grid, by confirming the
efficient and the reliable process, the electricity flows in two-ways and
the information is shared between the power plants and between all the
points. The smart grid is also used for monitoring the delivery of the
electricity and then it tracks down the consumption of the power with
the smart meters that is used to transmit the energy usage of
information to the utilities through the communication networks. It
will allow the customers to track down their amount of own energy that
is used by them on the internet, with the computer program as a third-
party. There are two ways of nature for the smart meter systems and it
allows sending the commands to perform it using the grid infrastruc-
ture devices, referred as distribution switches and the recloses, and
these are able to provide the energy for more reliable system, and this is
defined as the Distributed Automation (DA).
For enhancing the power quality, the AMI will be having the
capability. Between the smart maters and the central control stations
there will be a two way communication. By the use of these commu-
nications they can send the billing information to the customers. The
smart meter can be used to control the individual household loads with
their ability. By the benefit of micro girds and the AMI’s network, we
can possibly get the pattern of improved efficiencies and the moderat-
ing energy usage [14,15].(Fig. 2).
3.1. Smart meter
Electrical energy is recorded by the use of the smart meters and
then simultaneously it also sends the data back forth to the central
server for the purpose of monitoring and analysis of the data. The main
T. Ahmad
functions of smart meters, which are followed by the measurement of
the power flow due to the bi-directional form with the smart meters.
The main functions are recording the energy consumption data and to
give an interval between the reads, link with the communication,
connecting and disconnecting, remotely programming and theft detec-
tion. The consumers that are described will be wasting the energy and
they will make use of the energy in large amount of their sanctioned
load. The consumers are based on their functions namely, home area
network and the commercial and industrial consumers. Smart meter
will be used for the controlling process of the energy that is misused.
For certain cases, there will be certain limit for the power supply, when
it exceeds the smart meter, it will cut off the power supply. Then, it will
be restoring the power within its prescribed limits. The energy
management of the government goals will be able to accomplish using
smart meters. At present times, the distribution companies will be
facing the challenge for changing the conventional meters with the
smart meters. It is also used for updating the system. It has the biggest
advantage of receiving the energy consumption data continuously and
quickly and the processing, transmitting process will be received
continuously. When the distributed companies are being changed with
the smart meters, they will be able to find and locate the region and
then it will be controlled by the use of power consumption. Further, it is
able to find where the power consumption is very high. With the use of
the smart meter we can reduce the cost of the meter reading and the
collection of bill. Due to this feature, it can also improve the accuracy of
the bill and can reduce the theft of the error. With the daily usage, the
smart meter can be used to record the usage of the electricity and the
gas, with addition to it, also it can have the possibilities of recording the
consumption of the water. As delivered in Fig. 3 the architecture of the
smart meter will be having an energy meter, harmonic sensor, circuit
breaker. International Control Station (ICS) will be the central proces-
sing unit of the registered meter called smart meter. And also it will
used for controlling the components of the smart meter. Smart meter
will be allocating the portion for the energy mater. With the use of this
the amount of consumed electricity will be recorded. After receiving the
request from the ICS, the circuit breaker will be disabling the supply of
the electricity. For detecting the presence of the harmonics the
harmonic sensor function is used. There will be a work station present
for the wireless transceiver and the control system and it will designed
by the (ECS) external control system. By the use of this workstation, we
can compute the issues of the control signals and the non-technical
losses. The smart meter will be having wireless transceiver, and it is
used for sending and receiving the data and controlling the signals,
which is among the ECS, ICS. The ECS will be having the ability to
Fig. 3. Proposed new model of the smart metering infrastructure.
576
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
manage and control the supply of the electricity or any of the home
appliances. There will be a connected source between the harmonic
signals and the distribution transformer called Hybrid Filter. They are
used for the process of preventing the power system components
damage from the unwanted harmonic signals. It consists of two filters
namely active and passive filter. By the use of four fully control able
IGBTs can build the active filter. The combination of inductors and the
capacitors will be the passive filters [16–22].
3.1.1. Smart meter benefits
The smart meter benefits for the customers are:
A. When an outage occurs, the smart meter will allows the detection
of the faster outage and the service for restoration.
B. The customers will be provided with the greater control of their
own consumption of the electricity. When they available with the
functions of time-based rates and the range increases, due to the
different pricing plans that are available to the customers and provid-
ing them with more chances of monitoring their consumption of the
electricity and the bills obtained.
C. The smart meters which are produced to the customers will be
having the access for prior’s day electricity, and the usage will be
through the utility website.
D. Smart meter will be helping the environment for reducing the
need for building the power plants. To meet the high demand, the
utilities will be avoiding the use of the peaker plants, the benefits of the
environment will be determined as the peaker plants and typically it
will having the higher greenhouse gas and some other air missions.
E. It will be increasing the privacy because of the information about
the usage of the electricity and it can be relayed according to the
automatic utility process of the billing purposes without on-site visits
by the use of the utility it can check the meter.
F. Smart meters are the first preferences for creating the smart grid.
For the use of the smart grid every aspects of the industry will be
applied by the digital technologies, and from the generation, to the
distribution, transmission and to the interface of the customer. By the
help of the sense of the smart grid, it can be observed that what is
happening to the energy flows and should keep it in the balance and
then improve their reliability.
3.1.2. Smart meters worldwide deployment
The interface between the smart grids and the consumers will be
named as the smart metering. In Fig. 4, it will be delivering the process
for installing the smart meters for measuring the power generation and
the automation of distribution system as per the world markets of year
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

further steps like setting up the accurate form for the billing purpose of
the customer. After the installation of the smart meter, it is always
ready for the operation that has to be performed and it will be
proceeded automatically with the network system. There will be a
notification generated for the customer regarding the failure of the
installation, no presence of the user and the notification will be
delivering the time for the customer for later installation process [81].

4. Theft detection techniques

Fig. 4. Installed capacity of the smart meters world-wide, world market: 2012–2015
[23].
(2012–2015). The Europe will be considering them self as the part for
implementing the maters to 80% of the total population by the 2020
[20].
3.1.3. Smart meter installation
The process of actual installation of the smart maters will be the
model for planning the installation for normally. The part considered, if
it is done correctly, it can be able to lead to the installation smooth with
the low errors and the issues from the customers, delaying time for the
installation [81]. The aspects for the installation will be safely and can
be followed as below:
1. For utilizing the wires- (NESC) The National Electric Safety Code.
2. For the wiring process in the home- (NEC) The National Electric
Code.
3. For the codes of the Electricity Metering- ASNI C12.1
4. Codes for the local building.
The beginning steps of the installation will be involving in the
assessment for the access to the meter location and provides the safety
for the equipment’s that are existed.
After the establishment for the proper access, the following actions
will be included:
There are drawbacks which are major for the process of implement-
ing the policies existed, then following the electricity theft, and it
includes corruption and weak infrastructure. As a unit of the utility
company, the rules and guidelines for controlling the electricity theft
are implemented for reducing the electricity theft. The corrupt employ-
ees will be in the same system and will be displaying the least
dedication. For finding the various types of electricity theft, there are
some theft detection techniques that are used on the power distribution
system. By using all these features the NTLs can be detected.
4.1. Harmonic generator deployment for theft detection and
prevention
The paradigm shift will be our proposed approach from the method
called conventional method. It will be used for the process of finding
the illegal customer, by the use of the physical observation on the
distribution feeder. Otherwise, in the evaluating part the load pattern
will be generated for the customers. The observed Fig. 5 will be
showing the operations in the form of the flow chart for the proposed
systems. The operations that are carried out in a normal way, the smart
meters are presented in the household’s transmit instantaneous values
of the power consumption. The power factor and current for the (ECS)
are placed in the utility company. The ECS will be generating the total
non-technical losses from the values that are being received. The ECS
will be sending the control signals for the internal control station of the

1. For the safety issues, damage and diversion, it will be checking the
meter location.
2. The meter form type and the service provider will be verified for the
meter data.
3. Verifying the correct address, GPS location and meter number by the
use of the premise information.
4. For the purpose of the new installation, the customer premise
information should be updated

The housing unit split will be recorded approximately by the

national demographics [80]. The housing unit will be 74% for the
single family and the 26% for the multi-family, and the percentage will
be different from the state to state [80]. By concluding this, there will
be wide majority for the installation process of the smart meters and it
will be to the homes of single family with the designs of single meter.
Generally the meter base will be mounted to the exterior wall surface,
where the house will be attached with the service entrance. For the
multi-family dwelling units the multiple meters to a few locations will
be consolidated using the designs of the gang meter socket. Usually the
gang socket will be placed in the meter rooms, in the wall outside of the
apartment buildings or it can be placed in the basement of the long
rising apartment buildings. Though the multifamily installations will be
more complicated than the single family installations. But the process
of the installation steps will be same for them. For an extra function the
two processes will be designed to show the safety concerns and the
physical access, to see that the smart meter is installed properly and
safely and the information will be generated and delivered for the Fig. 5. Flow chart for detection of electricity theft using harmonic generator [56].

577
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

present smart meter to cut off the supply of electricity to the perfect

customer. This will take place in case of the computed non-technical
losses, which are higher than the 5% of the energy that is distributed.
Coming to the process, the illegal consumers are separated from the
genuine consumers. And the process of the legal consumers are being
isolated, and the illegal consumers will continue the process of drawing
energy from the grid. Just before the power cut in the neighborhood,
the generator called harmonic generator will be processed to on and off
for seconds. To produce the distribution feeder from the waste
harmonic components the function harmonic generator is used. By
using it, the customers who are illegal will be affected in the
performance of their home appliances. After the process of operation
on the harmonic generator, the signals of the ECS will be giving the
notification for the ICS of all the smart meters in order to the restoring
process of the power supply to the genuine customers. All this process
continues, after the harmonic generator is switched off [24,56].

4.2. Support vector machine

The vapnik is the one, who invented the support vector machines.
They are divided into the set of related supervised learning methods.
With the respect to the output, the support vector machines will
analyze the data given and can be able to recognize the pattern or data
trend in which it respects with the output. For this process, the
classification of the statistical data and the regression analysis is done
with the use of the SVMs [27,28]. The set of rules is determined by
receiving the training data and these generated model is developed by
the training algorithm of SVM. Basically, the SVMs will be developing a
set of hyper planes and a single hyper plane in a dimensional space of
high or infinite, and it is based on the classified data. Considering a
difference between the data points which is classified as it is achieved
and for describing a largest distance to the shortest training data points
of any type of class in the hyper plane. When the separation margin is
high, the classifier will be termed as minimal in the traces of the
generalization of error. For the past years, there are some applications
which are generated by the SVMs [29]. The applications are data
mining and face recognition. In case, if the SVM is to be tested and
trained, the noise data must be developed according to the registered
criterion. The development should be based on the approximate
consumption of the energy with their pattern. Before the process of
selecting the consumption energy data, the data must be considered
and the parameters should be inserted. The SVM model are:

1. Based on the locations of geographical (east, north, west, center,

south)
2. Seasons that will be in every year (winter, summer)
3. Customer classification (small, large) depending upon the area they
live.

Basically, the SVM is used for analyzing the consumption of the

energy in terms of the data and it will be used for the process of
recognizing the consumer energy consumption in terms of their present
behavior. The consumer load profile information is used by the SVM,
and is also used for exposing the behavior that is abnormal and highly
correlated with the terms of (NTL). Based on the overall consumption
of the energy, the customers are grouped in some forms depending on
their size of the house, and the electrical appliances they use. Fig. 6. Flowchart for support vector machine [25].
For the illustrating process, Fig. 6 will be generated. The illustrating
process will be the operational terms of the SVM. The operational the quick energy consumption. It will be representing the class in which
flowchart for the theft detection system is adapted in terms of energy the particular instance will be present. By concluding it, the consumers
during the given interval of the time. The energy consumption data that will be categorized into three classes based on the rules given below:
is collected will be the series of energy consumption values that occurs
instantaneously. The support vector will not support the format of this 1. Classifying under the class D:
present data so, the data has to be preprocessed. For the purpose of The approximate consumption pattern must be followed by the
training, a single part of the data is used. Hence, the input file for the energy consumed by the customers.
data used for training will be the values of the output and the values of For the two hours in 24-h period or the 96 inputs, it can have

578
T. Ahmad
zero consumption read.
The six individual zero-energy consumption inputs must be
produced. Then, for the process of the addition it will include the
customers who are all having zero-energy consumption.
2. Classifying under the class S:
The energy consumption pattern must be followed approximately
by the customer’s energy consumption, and can proceed with minor
changes or putting more than three inputs of the zero energy
consumption in one hour or without any repetition.
With not more than the two clients, it can have the two zero
energy consumption.
More than the given amount of six individual instances of the
zero energy consumption will not be proceeding to the next step out
of 96 inputs that are represented by 24 h in a day.
3. Classifying under class-I:
The approximate energy consumption for the customer will be
facing a small changes in the customer’s energy consumption profile.
By including the repetitions, there should not be more than two
non-continuous zero inputs.
The electricity consumption that is received should be changed to
other format, so only it can be a compatible format for the training
data. Based on the locations of the geographical, the data must be used
before the training, in case it belongs to the weekdays or the weekend,
customer range based on the load capacity and in which year of the
season the data is represented. In the database the data is being
transmitted into the central control station where the database is
located. For the purpose of training SVM model the training data is
being used and it also used for testing the data and also for process of
testing it and detecting the consumption that is named as illegal. The
energy consumption will be continuous for only the customer profile
that is genuine. And by checking the energy consumption pattern, the
customer will be named as the genuine customer and it will be sent to
the database. Suppose the suspicious profile found of the customer,
then that customer will be checked. If all the rules that are specified are
violated, the customer will be categorized the D class. And immediately
the customer can be inspected, is due to the high range of the illegal
consumption, if the profile of that customer doesn’t fall under the class,
and it doesn’t matter the customer is large or the medium, but they are
being tested for the regulations that are illustrated for classifying under
S. The customers who are very small, medium and large are divided
and shown in Table 1:
The profile of the customers which are not being under the S class,
the losses of the calculated overall distribution will be more than the
5% and then the class I will be generated after the profile of the
customer is rechecked in terms of the rules. After the passage of the
distribution transformer the losses of the distribution will be less than
the 3–5%. The illegal consumption will be occurred, when the
distribution losses are more than their range. The customers who are
not under the part of the class I will be named as the genuine customer.
Some customers are categorized in the class D are quickly checked. The
customers classified under the S class either they are large or medium,
they will also be checked and they are getting reported to the database
for quick checking in. The outputs of −1,0,1 will be results classified
under the classes D,S,I. Customers who are all named as 1 will be under
the class I, class S will be named for the customers with 0, and the class
D contains the customer generated as −1. The class D will be named as
Table 1
Types of very small, medium and large customers [25].
Consumption Very-Small Medium Large
KWH per Month 00–500.0 500.0–2000.0 2000.0–20000.0
KWH per Day 00–17.0 17.0–67.0 67.0–667.0
KWH per Hours 00–0.70 00.7.0–2.70 2.70–27.70
579
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
the illegal customer because of the stolen of electricity and they are
responsible for the process of billing and moving with that irregula-
rities. In the class S the customers will be involving in theft, and the
customers are viewed for more time and being processed for any fault
behavior [25–35].
4.3. Energy theft detection using central observer meter
The novel methodology is proposed in the system and then the
detection will be automatic for the illegal utilization of the electricity
used in the future distribution networks and then it is equipped with
the infrastructure of the smart metering.
Due to the overcome of the SM infrastructure the countries like
European is moving towards the SM infrastructure from the traditional
electricity metering. The standard that is in Netherlands (NTA 8130)
for the SM will be defined as the basics function and the parties for the
infrastructure of the SM. What so ever, the proper statement of the SM
will define upon the level of functionality that is not fully developed
[82]. In case of the managing process the efficiency for the new services
will be integrating in the future grid, and then communication which is
suitable must be performed. The method that is proposed for the
detection in advance will be the illegal for the use of the electricity with
respect to one of the features in the feasible type. And in which it
should not be the type of form which is omitted in the period for
planning and the process will be in the future standardization for the
SM infrastructure [83,84].
4.3.1. Investigated network configuration
In the future grid, in the Netherlands for stimulating the real
conditions the typical LV grid configuration is considered as a base
simulation of the model, where the cable networks with MV/LV
transformer and 240 customers connected. As per the future grid's
anticipated improvements, DG that is dispatched to the LV grid is
added according a substantial amount. An implementation known as
the “intelligent substations”, is called as future grid's additional feature.
This implementation will have to improve MV and LV network's
flexibility. When compared with the traditional substations, contains
traditional substations, the intelligent substations will combine the
controllable energy storage and it will have the MV/LV transformer,
which contains the automatic tap changer known as “Smart transfor-
mer”. Therefore, the loading and voltage profile in the MV grid will not
be associated directly with the voltage profile in the LV grid. A
monitoring system such as the intelligent MV/LV substation will be
improved, which can help in easy identification of theft in the network
[85]. The information that is received from the SM will be used for
proposing the methodology of detecting the electricity tampering. This
will be useful for automating, then for detecting and in the process of
localization. The distribution network in future will have the infra-
structure of SM. The topology consists of the smart metering infra-
structure and intelligent substation as an important part in the
electricity theft detection that is proposed.
4.3.2. Detection of illegal use of electricity
There are two phases that are involved in the method that is
proposed for detecting the method for electricity tampering attempts.
The initial phase emphasizes on the occurrence criteria of illegal usage
inside the LV network. Whereas, the other phase emphasizes on POC
localization and it is expected have illegal use. For evaluating the illegal
electricity use based on the initial phase, must depend on the
estimation and measurement of balancing the energy at the substation
of MV/LV. In case of the mismatch of the local energy balance with the
actually including the losses and the already measured one, it is
required that the sequence of the localization must be initiated for
verifying the occurrence of the illegal use occurrence inside the network
that is investigated. This illustrated in the below Fig. 7.
Only in the substations, the points of measuring and the data
T. Ahmad
Fig. 7. The flow chart for detection electricity theft using central observer meter [88].
processing will take place in the primary phase. Therefore, the
occurrence criteria must run in all the time frames irrespective of
maximizing the data or the requirements of computing for the
infrastructure of communication. Through SM, the production of data
and the consumption of energy must be present. The reconsideration
for the illegal detection once again is possible only for the localization
sequence that is implemented to define the time frame. In the scenario,
where there is no encountering of accuracy level, it is suggested that the
localization must intend to concentrate on the other time frame [86].
4.3.3. Tampering localization procedure
Generally, the illegal use of electricity takes place only where the
meter is nearby or inside the property. The relays secretly controls the
frequent illegal use of the electricity for complicating the fraud
detection. When the illegal use of electricity theft is encountered it
will be automatically launched, as the substation level shows positive
580
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
results with the balancing of power. The first priority are the cables that
are focused to provide the violating attempt before the meter or on the
terminals of the meter [87]. The tampering attempt's functional
adjustments basically depends on the presumptions and it must be
considered that there is no burden on the currents that are measured
by SMs. On the other hand, the POC's voltage must be measured as the
local values that easily reflect the power drawn from the nearby branch
of POC. The meter's voltage is influenced by the illegal use of the power
from the nearby branches. Hence, the illegal electricity usage is
burdened by the voltages. Every time there is an attempt of tempering,
voltage profile for all POCs connected will reflect it. Thus, Theft
localization present in the finial procedure, refer the Fig. 7, as it shows
the comparison of the voltages that are measured and the estimated for
the localization.
4.3.4. Errors in the influence of measurement
After the measurements for the accuracy in the errors, the expected
data from the SMs are mainly affect the errors in the measurements.
The values in the measurement should contain the standard errors in
the measurement for distribution and also these measurement errors
are attenuated by the extension of the time frame. Still in the process of
stimulation, showing the proper methods the robustness and also it
shows the robustness for the errors that generated randomly. In fact
the results are obtained in the successful manner for the tampering
attempts in the correct positions are found easily. The localization
accuracy that mainly depends on both the accuracy measurement of the
SMs and the time frames in the amount. These two methods are used in
the localization. From the localization, the location in accurate that
does not provide the clean and clear difference. Then the localization
that extended the estimation from one time frame to another time
frame or may be the time frames are aggregated, till the order of the
localization that provides the level of confidence satisfied [88].
Different types of methods in their models are used to find the theft
in electricity. Above methods of mathematical or other models are
shown such as Support Vector Machine Radial (SVM-RADIAL), Multi-
Layer Perceptron (ANN-MLP), Artificial Neural Network and Optimum
Path Forest classifier (OPT) [36,37]. The technique of Multi-layer is
mainly depends on the some of the methods for modelling, which obeys
some of the other methods like tree diagram or Non-Linear Statistical
Modelling (NLSM). The technique for recognizing the supervised
pattern are commonly used. Others like Support Vector Machine and
Artificial Neural Network may be put for the purpose of identifying the
commercial frauds through automatic methods. Anyhow, they bear a
lot from the convergence in slow and burden high computation [38].
According to [40], it is demonstrated that the Optimum-Path Forest
classifier for identifying the losses related to the non-technical ones are
higher than the neural networks but they resemble to the Support
Vector Machines in a faster manner. Though the techniques such as
artificial intelligence are increasing there is some need to focus on the
drawbacks. For instance, the ANN with multi-layer perceptrons (ANN-
MLP) has the capacity to both address the issues in a linear and in a
non-linear way. But, the exception is that it is not possible to do the
non-separable situations [89]. Being a classifier, which is unstable, the
ANN-MLP collections can improvise the performance level of the
classifiers with an unidentified limit [90]. The development of the
vector machines is initiated for resolving the existing issues [91]. To
support this the linearly separable classes in a higher-dimensional
feature space is assumed [93]. The support vector machine's processing
cost surges as the training size and the support vector increases. Being
a binary classifier, there is a need of multiple SVMs for solving the
issues related to the multi-class. The new framework which reduces the
issues of pattern recognition for the graph-based classifiers same as the
Optimum Path Forest (OPF) computation in the feature space that is
induced by the graph are shown [92]. Such classifiers will not obstruct
the tasks related to classification, which is known as the hyper planes
optimization issues whereas, the combinatorial optimum-path compu-
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Table 2 5. Simulation and results

DISCO’s revenue collection efficiency in Pakistan [47].
With the implementation of the linear regression analysis, the
DISCOs 2009– 2010– 2011– 2012– 2013–
2010 2011 2012 2013 2014 spearman rank correlation coefficient test and distribution line mon-
itoring system to detect the theft of electricity with the help of Power
PESCO 71.0% 67.0% 70.0% 78.0% 68.0% Line Communication (PLC) is done which is developed to detect the
HESCO 77.0% 68.0% 60.0% 59.0% 60.0%
theft of electricity on the electrical distribution system. The Matlab
QESCO 86.0% 80.0% 76.0% 41.0% 36.0%
LESCO 98.0% 96.0% 93.0% 98.0% 96.0% software and the SPSS are used for analyzing the performed operations.
GEPCO 98.0% 95.0% 96.0% 99.0% 98.0%
FESCO 99.0% 97.0% 97.0% 97.50% 98.0% 5.1. Linear Regression Analysis
IESCO 98.0% 97.0% 96.0% 93.0% 96.0%
MEPCO 97.0% 96.0% 94.0% 98.0% 97.0%
In the linear equation, the appropriate linear regression applies for
modelling the two variable's relationship according to the data that is
tation are present for the rest of the nodes. Every single node has been observed, where one of the variable is referred as the explanatory
categories as per the strength of the prototypes that are connected variable and the second variable is referred as the dependent variable.
which is defined as the discrete optimal partition for the feature space. The “X” and “Y” variable's average values will be identified using the
Every single prototype acts as the root from the optimum-path tree. regression analysis method. Each variable has a purpose, and it is
There is a technique proposed by Ramos, C. C. O et al called as a determined that the variable “Y“ contains the data that is retrieved
technique which is based on the OPF to identify and to safeguard the from the energy which is produced either with the energy that is
NTLs. The phase of implementation is fast according to [39] and [40]. received or from power house or from the feeder energy that is
It is determined that the OPF has higher chances to identify the delivered. It is determined that the variable “X” provides the energy
electricity theft. On the other hand, they are even higher in accuracy paid back values as a bill format for the utility from customer's side.
when compared to SVM-RBF, ANN-MLP and SVM-LINEAR [41–46]. The following equation provides the correlation which is present
among the two X and Y variables as follows:
σxy
ρ=
σx. σy (2)
4.4. Distribution companies (DISCOs) wise revenue collection
efficiency in Pakistan The above equation contains some of the elements as follows:
“σx” resembles the standard deviation of the variable “X”.
The reasons for the largest contributors of the circular debts are the “σy” resembles the standard deviation of the variable “Y”.
non-payment bills of the electricity used, then the industrial customers Finally, “ρ”, it resembles coefficient of the correlation.
and the commercial customers. The same reasons are not applicable for The above equation can also be written in the below mentioned
all the companies in the world, there are some of the firms that take form:
care of their records which are not possible by some of the firms and x, y
they faced the low revenue collection issues. This is showed in Table 2 ρxy = Cov
σx. σy (3)
[47].
The issues that are being escalated due to the circular debt is The data reflects the dispersion with the co-variance (Cov). Cov is a
because of the below mentioned reasons: function's property to retain its form during the cases when there is a
linear transformation of the variables. Co-variance's formula is pro-
• Theft of electricity is the main reason. vided below. It is a combination of the expectation and the mean.
• The revenue collection of DISCOs is poor. Cov (x, y) = E [(x − x )( y − y )] (4)
• The delay in infusing the cash in the power sector using the fuel
The above equation contains some of the elements as follows:
price methodology.
• Losses related to transmission and distribution. “E” resembles expectation/expected value. “E” is the random
• Long term approval by the court for the Fuel Price Adjustments variable’s weighted average value. The following equation provides
the basic equation for finding sample’s squared standard deviation,
(FPAs).
• The governance is poor. which is also referred as the energy consumption data.
• Tariff determination delay x2 ⎛ x⎞
2

• Contract payment for the subsidy of the Tariff Differential Subsidy σ² = ∑

(TDS) and Karachi Electric Supply Company (KESC).
• Late notification from the Government of Pakistan (GOP).
• Late payments on Tariff Differential by the Ministry of Finance
(MOF).
The federal government doesn’t plays a vital role to pass the
legislation for controlling the theft of electricity, to promote energy
conservation, to strengthen the regulatory process, to achieve trans-
parency and to encourage an open and competitive market. In Pakistan
there are certain issues that are worsening the situation of the circular
debt such as, insufficient budget by TDS, poor tariff terms and
conditions, opposed generation views, poor thermal power plant
efficiency, non-commercial or the non-professional approach for shed-
ding the load, promoting the demand side management and Late
Payment Surcharges (LPS) were GOP neglected by Central Power
Purchasing Agency (CPPA) to Independent Power Producers (IPPs)
[47].
N
− ⎜∑ ⎟
⎝ N⎠ (5)
The Islamabad Electric Supply Company’s energy consumed data is
provided in Table 3.
The analysis to identify the energy theft is conducted in various
parts of the Rawalpindi region by the various feeders.
Fig. 8, projects a scatter plot, which represents that the X-axis with
the customer's monthly metered load denoted as “X” whereas, the Y-
axis represents the feeder section delivered load denoted as “Y” with
the help of the SPSS Statistical software. The Scatter plot displays the
correlation among the utility-supplied load and the total customer-
metered load. In the plot the slope changes refers to the losses based on
the technical front and due to the errors of calibration, which is
projected in Fig. 8 by the slope 1.007. The percentage of technical loss
is corresponded in Fig. 8 by 0.7%. The theft of monthly electricity is
determined as 519.93-MWH by Fig. 8. If there is any data point's
dispersion in the linear regression line then it easily represents the
electricity theft is not constant when a whole month is considered. As

581
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Table 3
Energy consumption data from different feeders in Rawalpindi Region (IESCO) by July 5, 2015.

Sr. # 11-KV Feeder Name Energy- MWh %-Line Losses (c) Increase/Decrease in Losses (a-b)

Received (a) Sold (b)

1 11-KV Out/Going NOGAZI 1398.0 1397.0720 0.00664 0.9280

2 11-KV Out/Going AIR PORT NEW 235.00 235.5240 −0.220 −0.520
3 FEEDER DHAMIAL CAMP 7222.00 7162.320 0.82640 59.680
4 11-KV OUT/GOING ADYALA 10752.650 10768.5450 −0.140 −15.80
5 FEEDER RADIO PAKISTAN NEW 14311.960 14304.9440 0.04900 7.0160
6 11-KV OUT/GOING CHHAB 3322.1150 3326.1430 −0.120 −4.00
7 FEEDER DHOKE FARMAN ALI 11099.690 10480.2870 5.58040 619.400
8 FEEDER GAWALMANDI 8013.060 7941.8340 0.88890 71.2260
9 FEEDER KHADAM HUSSAIN 9663.0 9676.5520 −0.140 −13.50
10 FEEDER NATIONAL PARK 13182.0 13055.9660 0.95610 126.030
11 FEEDER PAKISTAN AIR FORCE 9145.6280 9125.340 0.22180 20.2880
12 FEEDER CHAKLALLA SCHEME-3 15738.950 15512.5880 1.43820 226.360
13 11 kV OUT/GOING MEADIA-II 3170.9520 3175.2380 −0.130 −4.280
14 FEEDER RIVER GARDEN 4187.0750 4185.640 0.03430 1.4350
15 FEEDER SWAN-II 2526.50 2520.2230 0.24840 6.2770
16 FEEDER OUT/GOING SECTOR-H 330.00 331.040 −0.310 −1.040
17 FEEDER OUT/GOING KLM 3060.0 3058.9470 0.03440 1.0530
18 11-KV OUT/GOING AHMEDAL 3787.380 3794.1520 −0.10 −6.70
19 FEEDER RADIO PAKISTAN 7857.3940 7873.7340 −0.200 −16.30
20 FEEDER ENGINEERING.M.E COM. 14601.00 14627.4590 −0.180 −26.40
21 11-KV OUT/GOING AZHAR ABAD 7579.9080 7593.7480 −0.1820 −13.840
22 11-KV OUT/GOING CHAHAN 7850.0 7838.2480 0.14970 11.7520
23 FEEDER WILLAYAT COMPLEX 8689.980 8689.8190 0.00190 0.1610
24 FEEDER OUT/GOING FAUJI/FOUNDATION 8642.20 8629.9110 0.14220 12.2890
25 FEEDER NAWAZISH SHAHEED 2862.880 2848.0250 0.51890 14.8550
26 11-KV OUT/GOING FEEDER HFF- 1 7424.00 7271.70 2.05150 152.300
27 11-KV OUT/GOING FEEDER HFF−3 3064.00 3057.5090 0.21180 6.4910
28 11-KV FEEDER TAXILA 4480.0 4490.0470 −0.22430 −10.040
29 11-KV O/G FEEDER BHATIOT 945.220 939.1710 0.64000 6.0490
30 FEEDER LANI WALA 6171.520 6167.1290 0.07110 4.3910
31 11-KV OUT/GOING FEEDER TANAZA DAM 4069.6060 4068.1090 0.03680 1.4970
32 FEEDER SAJJAD SHAHEED 1954.00 1899.6480 2.78160 54.3520
33 FEEDER OUT/GOING BEHLOT 7826.00 7991.4570 −2.1140 −165.50
34 FEEDER 1- KV OUT/GOING W/COLONY 7901.1980 7889.760 0.14480 11.4380
35 FEEDER 11-KV O/G AMIN ABAD 6815.6340 6815.9730 −0.050 −0.30
36 FEEDER 11-KV OUT/GOING P/COLONY 2602.080 2567.60 1.32510 34.480
37 FEEDER 11-KV INCOM./C NO−2 6386.00 6365.7470 0.31710 20.2530
38 FEEDER 11-KV TENCH BHATHA 1112.5180 1109.430 0.27760 3.0880
39 FEEDER 11-KV AWC HOUSING 5014.10 5001.0250 0.26080 13.0750
40 FEEDER COL. SHER KHAN 9821.5640 9783.6860 0.38570 37.8780
41 FEEDER RASHID MINHAS 2823.2420 2817.5220 0.20260 5.720
42 FEEDER SHAMSABAD 366.0780 366.00 0.02130 0.0780
43 FEEDER 11-KV OUT/GOING HAZRO 836.960 839.7840 −0.30 −2.80
44 FEEDER 11-KV OUT/GOING MALHOWALI 5018.340 5011.4140 0.13800 6.9260
45 FEEDER GHARIBWAL 6632.6310 5633.9260 15.05740 998.700
46 FEEDER 11-KV OUT/GOING KACHARI 826.00 785.0880 4.95300 40.9120
47 FEEDER 11-KV OUT/GOING SHADI KHAN 1979.4950 1978.7010 0.04010 0.7940
48 FEEDER MANSER 590.5250 590.2690 0.04340 0.2560
49 FEEDER OUT/GOING PIND SULTANI 590.5250 7313.5920 00.8192 60.0408
50 FEEDER 11-KV NARRA 7579.080 7529.7140 0.06513 49.0366

mentioned earlier, the slope of the scatter plot provides the losses
based on the technical and calibration errors aspect with the help of
linear regression on the load. Moreover, the y-intercept provides the
straightforward values of the theft.
Centroid=(X , Y )=(5728.746, 5776.86226).
The centroid is a place where whichever regression line fits best will
surely pass this or it should pass from this point. The centroid consists
of the “X” and “Y” variable’s mean value.
The regression line’s equation is as follows:
ˆ =bo +b1 . Xi
Yi
∑ . (Yi − Y )²
b1 = ∑ aggregate entirety of SST Squares is equal to Sum of Squares Regression
∑ (Xi − X )(Yi − Y i )
From the analysis of the regression, the resultant key is the
coefficient of determination, which is denoted as r2.
The understanding is that, the variance’s proportion in the depen-
dent variable is easily predicted with the help of the variable which is
independent.
It is noted that the coefficient of determination each time is just
between 0 and 1.
n ( ∑ XY ) − ( ∑ X )( ∑ Y )
r²=
[n ∑ Y 2−( ∑ Y )²][n ∑ X2 −( ∑ X )²] (8)
So when we led the relapse, the SSE (A relapse chooses the line with
(6) the most minimal aggregate whole of squared forecast mistakes, this
esteem is known as the SSE or Error based on the total sum of Squares)
diminished from SST to SSE (or from 779732138.7 to 14559360). The
(7)
plus Total sum of Squares (SST=SSR+SST) so therefore Sum of Squares
Regression is equal to Total Sum of Squares minus Sum of Squares Error
(765172778.5=779732138.7-14559360). SSR is consider as total square
variations form SSE within the interval and SSE

582
T. Ahmad
Fig. 8. Regression line fitted on customer-metered loads and feeder section load delivered (MWh), July 07, 2015.
765172778.5
Determination Co−efficient = r 2 = = 98.13277413
779732138.7
To assess a one dependent variable from one or more Independent
variable, Relapse methods are used. The two strong positive direct
connections (r2) from X and Y, if r2 is close to point +1. The exact
Positive fit means r2 value meets at point +1. The connections between
the X and Y is described from the above positive fit values, which
indicates that X values increase then simultaneously Y value will
decrease. A connection more noteworthy than 0.8 is by and large
portrayed as solid, though a relationship under 0.5 is for the most part
depicted as powerless. In this direct relapse investigation, r2 is equal to
0.9813227741 indicates X and Y variable linear relationship is
determine as the 98% of the total variation in variable X and the
remaining 1.867225872% of total variation of variable X is undeter-
mined or Theft load Energy utilized information based on 05July 2015,
mentioned in Table 4. In Fig. 9 utilizing SPSS form points are useful to
setting up attractive reaction (forecast of variations from the norm
zone).
Energy utilized data is obtained from connection of X (User-
Metered Contour Layer) and Y (Load based on feeder delivery) that
form a maximum or minimum ridges of large or small values. In Fig. 9
shows a least energy utilization values and largest energy utilization
values form the contour layer, least energy utilization values shown
between 235.5240 and 2145.1570 MWh and largest energy utilization
values shown between 14557.77 and 15512.5880 MWh and also Fig. 9
shows scattered from User-Metered contour layer and Load based on
feeder delivery are indicated as 4000–6000 MWh and 1000–
13000 MWh respectively.
5.2. Spearmans rank correlation coefficient test
The statistical energy value of the singles/monotonic (amount/
capacity) correlation for combined data is obtained from Spearman’s
Relationship Co-efficient. An example it is signified by configuration
compelled takes after −1≤rs≤1.
Spearman’s rank relationship co-efficient (rho or r) is derived as
∑i R (xi ) R ( yi) − n {
n n +1
2
}2
rs=ρ= 0.5 0.5
{∑in R (xi )2 − n { n 2+1 }2 } {∑in R ( yi)2 − n { n 2+1 }2 }
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
Variable X and Variable Y has two ranks as R(xi) and R(yi). Two
(9) Variable’s (A > and B > ) positive correlation is obtained from positive
relationship co-efficient and two variable’s (A > and > B) negative
correlation is obtained from negative relationship co-efficient. A
relationship co-efficient of null (0) determines the correlation between
two variables are zero. Regardless of the possibility that the connection
coefficient is zero a non-straight relationship may exist. Existing
information might ordered already but we need to order our informa-
tion again and again. In Table 4, Test for Spearman rank relationship
co-efficient energy utilization information can quickly get from User-
Metered Contour Layer and Load based on feeder delivery with the
help of ranking.
In Table 4, Load based on feeder delivery and user-metered contour
layer are utilized to find out rs and rank accordingly. Rank 1 will be
least value in mhw based on User-Metered Contour Layer verses, No. 2
rank will be second least value in mhw and remaining ranks will added
accordingly. After that we need analysis the variation of two variables X
and Y and then variance should be multiplied. This is known as
squaring difference rank or squaring.
rs=0. 9890 = 98. 9% (11)
In Table 3.6, we find the positive relationship based on the value of
spearman’s relationship co-efficient rs=0.9890 = 98.9percent rs value is
near to 1, so that the connection is powerful between information of
two sets and information of two sets can attached that will go high
together. If rsis (minus ) − 0.9890 then one set will go high and other set
will go down. 100% minus 98.9% is equal to 1.1% this indicates as
unexplained or theft load energy utilized information based on Table 4.
5.3. Detect the theft power with the help of PLC (Power Line
Communication)
Electricity theft detection using Power Line Communication (PLC)
a narrow band carrier signal is injected into the electric distribution
lines along with power frequency signal (230 V, 50 Hz). Thin Band of
current Conducting Unit has amplifier that will identify unauthorized
consumption of power through fluctuation in the amplifier. MATLAB
software is used to find the simulation results. To product the electrical
belongings in the home from high electricity and current fluctuation
(10)
some methods used by Power Line Communication (PLC), that are
capacitors coupling and traps in the supply line. Fig. 10 shows the
583
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Table 4 values can be selected using the generated frequency produced in the
Ranking of the data between customer metered-load and feeder delivered load. carrier signal. In the distribution line, the combined signal present will
be having the amplitude of 250 V with respect to the addition of power
Load Y-Rank Customer- X-Rank Rank Square
delivered Metered Difference Difference frequency voltage of 230 V with the amplitude frequency signal of 20 V.
by the Load in the Rank The theft less condition will be considered in Fig. 12. The occurrence of
Feeder the output carrier signal with high frequency will be caused by the
Y X di (Y-X) di² coupling capacitor, and distribution line. The occurred signal ampli-
15738.950 50.0 15512.5880 50.0 0 0 tude will be under the condition of low and it is used as the reference
9145.6280 42.0 9125.340 42.0 0 0 term for the detecting process of the energy theft.
13182.0 47.0 13055.9660 47.0 0 0 Fig. 13 will be considered as the theft occurrence in the simulation.
9663.0 43.0 9676.5520 43.0 0 0 The consideration is due to the impedance presence in the circuit also
8013.060 39.0 7941.8340 38.0 1 1
the occurrence of the power frequency signal and the high frequency
11099.690 46.0 10480.2870 45.0 1 1
3322.1150 19.0 3326.1430 19.0 0 0 carrier signal will be present. Then, the loss of the power frequency
14311.960 48.0 14304.9440 48.0 0 0 signal will be involved and with respect to that there will not be changes
10752.650 45.0 10768.5450 46.0 −1 1 in the electricity usage illegally. If any absence occur in the co-
7222.0 30.0 7162.320 30.0 0 0
ordination of the high frequency carrier signal, the main state for
235.0 1.0 235.5240 1.0 0 0
1398.0 9.0 1397.0720 9.0 0 0 causing it will be the impedance with loss of power signal that cause
8642.20 40.0 8629.9110 40.0 0 0 loss in the signal.(Fig. 14).
8689.980 41.0 8689.8190 41.0 0 0 For the simulation process, the RF load is considered. The loading
7850.0 36.0 7838.2480 35.0 1 1 consists of series connection through the distribution line in terms of
7579.9080 34.0 7593.7480 34.0 0 0
the power supply. The theft will be detected after the amplitude of the
14601.0 49.0 14627.4590 49.0 0 0
7857.3940 37.0 7873.7340 36.0 1 1 output is compared with reference in terms of the condition.
3787.380 20.0 3794.1520 20.0 0 0 Fig. 15 will be consisting of two theft load of 100 W and 230 V
3060.0 16.0 3058.9470 17.0 −1 1 lighting bulbs. Then the distribution line load is made to be resistive.
330.0 2.0 331.040 2.0 0 0
The generated load resistance will be calculated using the below
2526.50 12.0 2520.2203 12.0 0 0
4187.0750 22.0 4185.640 22.0 0 0
formula,
3170.9520 18.0 3175.2380 18.0 0 0 R=V^2/P Ω.
6386.0 27.0 6365.7470 28.0 −1 1 R- Load Resistance.
2602.080 13.0 2567.60 13.0 0 0 V- Voltage drop in the load.
6815.6340 29.0 6815.9703 29.0 0 0
P- Consumed power in the load.
7901.1980 38.0 7889.760 37.0 1 1
7826.0 35.0 7991.4570 39.0 −4 16 The principles in the power line communication technique is proves
1954.0 10.0 1899.6480 10.0 0 0 by the simulating process of the results based on the theft and no theft
4069.6060 21.0 4068.1090 21.0 0 0 condition and also it can be more effective for the usage of detecting the
6171.520 26.0 6167.1290 27.0 −1 1
theft in the electricity in the distributing system of power.
945.220 7.0 939.1710 7.0 0 0
4480.0 23.0 4490.0470 23.0 0 0
Comparing with the existing system, the various benefits can be
3064.0 17.0 3057.5090 16.0 1 1 identified in the PLC power theft detection system.(Fig. 16).
7424.0 32.0 7271.70 31.0 1 1 The electricity theft can be identified within the range of 250–
2862.880 15.0 2848.0205 15.0 0 0 500 m, the system used will be providing an efficient and accurate
7579.080 33.0 7529.7104 33.0 0 0
system, when comparing with the existing system.
7374.0 31.0 7313.5902 32.0 −1 1
590.5250 4.0 590.2690 4.0 0 0 With the use of the simple and the cheap components in the
1979.4950 11.0 1978.7010 11.0 0 0 updated system, the implementation can be done easily in the network
826.0 5.0 785.0880 5.0 0 0 of the power system.
6632.6310 28.0 5633.9260 26.0 2 4 The loss in the electricity can be reduced by the system with respect
5018.340 25.0 5011.4140 25.0 0 0
836.960 6.0 839.7840 6.0 0 0
to the power theft limitation produced by the system.
366.0780 3.0 366.0 3.0 0 0 The function of losing the revenue due to the power theft will be
2823.2420 14.0 2817.5022 14.0 0 0 leading to the non-development of the countries. So, using this the
9821.5640 44.0 9783.6806 44.0 0 0 theft can be reduced and the revenue will be increased accordingly. So,
5014.10 24.0 5001.0205 24.0 0 0
the power quality can be improved in the developing countries.
1112.5180 8.0 1109.430 8.0 0 0
∑ di2= 32
6. Conclusion

complete construction designs for PLC (Power Line Communication)
by using MATLAB, ordinary current fluctuation unit and large fluctua-
tion unit are connect together and the transferred to Supply distribu-
tion line by using capacitor coupling and traps for supply line.(Fig. 11).
The series of inductor will be present at the input circuit having
great frequency supply for the purpose of blocking the signal of high
frequency, which affects the source of the supply. For protecting the
oscillator circuit, coupling capacitor and series resistor is being
connected in the higher frequency side. The protection is made for
the supply of the power frequency in it. For providing an impedance
matching, line tuner will have the connection of series of resistor. Once
the supply is simulated, the addition of two signals will take place and
the transmission will be made through distribution line. The carrier
signal will be frequency will be at 50 kHZ. The inductor and capacitor
The losses produced in the non-technical will be due to the actions
of the external power system. They will be consisting primarily of
electricity theft, the loss of $20 billion is estimated in the companies
worldwide. The illegal customers are identified by the use of the
support of vector machines based on their consumption of the energy.
The flowchart is shown for the theft detection with the use of the
harmonic generator. The difference between the imports and the
exports of the energy can be identified by the technique of linear
regression. Table 3 will be showing the electric supply company in the
Islamabad for the analyzing of the fraud detection. In the slope the
changes will be occurred in terms of the technical and calibration
errors. For the analysis of energy consumption, the central observer
technique is being used. The load delivered in the feeder section will be
correlating the customer meter load and utility supply load. As shown
in the slope of 1.007 in Fig. 8 with corresponds to the losses in the

584
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Fig. 9. User-Metered Contour Layer verses load based on feeder delivery.

Fig. 10. Proposed system circuit diagram without using theft load.

Fig. 11. Input signal simulation with the use of distribution line.

585
T. Ahmad Renewable and Sustainable Energy Reviews 72 (2017) 573–589

Fig. 12. Output signal simulation under the less theft condition.

Fig. 13. Output signal simulation under the theft condition with the use of the RL Load.

Fig. 14. Diagram for the proposed system with respect to the RL load.

technical of 0.7%. The analysis of the linear regression and the monthly keeps unchanged or it is theft in the energy consumption. In addition to
electricity will be providing a correlation between the X and Y the system, for detecting and monitoring the electricity theft will be
variables. In the other area the value of 1.534735487% in the variation done by power line communication technique.

586
T. Ahmad
Fig. 15. Diagram for the proposed system with respect to the RL load as 2*100–W bulb.
Fig. 16. Results of the output signal with theft load as 2×100-W lamp.
7. Future work
The controlling and the integrating part of the home appliances
with the smart meter will be using different types of the sensing
elements and detail the study, comparison and analysis of the power
transformers, feeders of 11 kV, distribution transformers and distribu-
tion lines, involvement in the NTLs, that causes the electrical power
system overload or faulty material.
References
[1] Smith TB. Electricity theft: a comparative analysis. Energy Policy
2004;32:2067–76.
[2] Shrestha RM, Azhar M. Environmental and utility planning implications of
electricity loss reduction in a developing country: a comparative study of technical
options. Int J Energy Res 2013;22:47–59.
[3] World Bank Data. Annual Energy Outlook 2014. 〈http://data.worldbank.org/
indicator/EG.IMP.CONS.ZS〉.
[4] Çağlar Y. The causes of electricity theft: an econometric analysis of the case of
Turkey. Util Policy 2015;37:70–8.
[5] Soma SSRD, Lingfeng W, Vijay D, Robert CG. High performance computing for
detection of electricity theft. Int J Electr Power Energy Syst 2013;47:21–30.
[6] Guillermo ST, Juan ET, Pedro PL, Sergio P. Power-aware anomaly detection in
smartphones: an analysis of on-platform versus externalized operation. Pervasive
Mob Comput 2015;18:137–51.
587
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
[7] Davidson IE. Evaluation and effective management of non-technical losses in
electrical power networks. In: Proceedings of the IEEE AFRICON conference in
Africa, South Africa; 2002; p. 473–77.
[8] Juan IG, Carlos L, Iñigo M, Félix B, Jesús B. Improving knowledge-based systems
with statistical techniques, text mining, and neural networks for non-technical loss
detection. Knowl-Based Syst 2014;71:376–88.
[9] Edison AC, Aranha N, Jorge C. Probabilistic methodology for technical and non-
technical losses estimation in distribution system. Electr Power Syst Res
2013;97:93–9.
[10] Nizar AH, Dong ZY, Jalaluddin M, Raffles MJ. Load profiling non-technical loss
activities in a power utility. In: Proceedings of the 1st international power and
energy conference (PECON); 2006; p. 82–7.
[11] Rhona F, Nikki M. Basic concepts for crew resource management and non-technical
skills. Best Pract Res Clin Anaesthesiol 2015;29:27–39.
[12] Francis MM. Adopting electricity prepayment billing system to reduce non-
technical energy losses in Uganda: lesson from Rwanda. Util Policy 2012;23:72–9.
[13] Lucy M, Rhona F, Steven Y, Janet M, Kathy C, George Y. Evaluation of the scrub
practitioners' list of intraoperative non-technical skills (SPLINTS) system. Int J
Nurs Stud 2012;49:201–11.
[14] Balakrishna PA, Rajagopal KB, Swarup KS. Application benefits of distribution
automation and AMI systems convergence methodology for distribution power
restoration analysis. Sustain Energy Grids Netw 2015;2:15–22.
[15] Jorge R, Florentino L, Juan CD, David GM, Linda B, Haramatib , Jeffrey M,
Levskyb , Nada S, Ileana P, Mario JG. Microvascular obstruction detected by
cardiac MRI after AMI for the prediction of LV remodeling and MACE: a meta-
analysis of prospective trials. Int J Cardiol 2016:344–8.
[16] Watral Z, Jakubowski J, Michalski A. Electromagnetic flow meters for open
channels: current state and development prospects. Flow Meas Instrum
2015;42:16–25.
T. Ahmad
[17] Domenico C, Luigi F, Vincenzo P, Antonio P, Alfredo V. Experimental character-
ization of consensus protocol for decentralized smart grid metering. Measurement
2016;77:292–306.
[18] Myriam N, Robin W, Andrew M, Jenkinsc , Jialiang Y, Graeme H, Padraig FL,
Yvonne H, Phil TB, Phil CT. A probabilistic approach to combining smart meter and
electric vehicle charging data to investigate distribution network impacts. Appl
Energy 2015;157:688–98.
[19] Cominolaa A, Giuliania M, Pigab D, Castellettia A, Rizzolid AE. Benefits and
challenges of using smart meters for advancing residential water demand modeling
and management: a review. Environ Model Softw 2015;72:198–214.
[20] Wesley S, Mica E, Joseph S, Rebecca S, Nilmini SS. Using in-home displays to
provide smart meter feedback about household electricity consumption: a rando-
mized control trial comparing kilowatts, cost, and social norms. Energy
2015;90:351–8.
[21] Malte N, Birgit M, Ellen M, Karolin TM. An analysis of smart metering information
systems: a psychological model of selfregulated behavioural change. Energy Res Soc
Sci 2015;9:85–97.
[22] Greta B, Mike D. The ageing population and smart metering: a field study of
householders' attitudes and behaviors towards energy use in Scotland. Energy Res
Soc Sci 2015;9:107–15.
[23] MIT Technology Review 〈http://www.technologyreview.pk/smart-meters-are-they-
here-for-good/〉.
[24] SSSR Depuru, Wang L, Devabhaktuni V, Gudi N. Measures and setbacks for
controlling electricity theft. North Am Power Symp (NAPS) 2010:1–8.
[25] SSSR Depuru, Wang L, Devabhaktuni V. Support vector machine based data
classification for detection of electricity theft. Power Syst Conf Expo (PSCE)
2011:1–8.
[26] Dong J, Wang R. A novel smooth support vector machines for classification and
regression. Comput Sci Educ 2009:12–7.
[27] Nagi J, Mohammad AM, Yap KS, Tiong SK, Ahmed SK. Non-technical loss analysis
for detection of electricity theft using support vector machines. Power Energy Conf
2008:907–12.
[28] Jianhua X. Multi-label Lagrangian support vector machine with random block
coordinate descent method. Inf Sci 2016;329:184–205.
[29] Jyotirmayee D, Kanungo BM. Fault classification in SEIG system using Hilbert-
Huang transform and least square support vector machine. Int J Electr Power
Energy Syst 2012;76:11–22.
[30] Benson M, Mon JW, Isabelle EB, Haina M, Cristian PZ, Giovana BZS, Khader MH,
Jair CS. Predictive classification of pediatric bipolar disorder using atlas-based
diffusion weighted imaging and support vector machines. Psychiatry Res
Neuroimaging 2015;234:265–71.
[31] Sajan KS, Vishal K, Barjeev T. Genetic algorithm based support vector machine for
on-line voltage stability monitoring. Int J Electr Power Energy Syst Vol
2015;73:200–8.
[32] Yong L, Huifeng W, Hong Z, Karsten L. A comprehensive support vector machine-
based classification model for soil quality assessment. Soil Tillage Res
2016;155:19–26.
[33] Nele V, Joaquín D, Chris C, Salvador G, Francisco H. Evolutionary wrapper
approaches for training set selection as preprocessing mechanism for support
vector machines: experimental evaluation and support vector analysis. Appl Soft
Comput 2016;38:10–22.
[34] Tiago P, Tiago MS, Isabel P, Zita V, Hugo M. Support vector machines for decision
support in electricity markets' strategic bidding. Neurocomputing
2016;172:438–45.
[35] Mahmoud RS, Golnaz D. The application of continuous wavelet transform and least
squares support vector machine for the simultaneous quantitative spectrophoto-
metric determination of Myricetin, Kaempferol and Quercetin as flavonoids in
pharmaceutical plants. Spectrochim Acta Part A Mol Biomol Spectrosc
2015;152:443–52.
[36] Nagi J, Mohammad AM, Yap KS, Tiong KS, Ahmed SK. Non-technical loss analysis
for detection of electricity theft using support vector machines. Power Energy Conf
2008:907–12.
[37] Qasim M, Khadkikar V. Application of artificial neural networks for shunt active
power filter control. IEEE Trans Ind Inform 2014;10:1765–74.
[38] SSSR Depuru, Wang L, Devabhaktuni V. Support vector machine based data
classification for detection of electricity theft. Power Syst Conf Expo (PSCE)
2011:1–8.
[39] Ramos CCO, Souza AN, Papa JP, Falcão AX. Learning to identify non-technical
losses with optimum-path forest. In: Proceedings of the IWSSIP 2010-17th
international conference on systems, signals and image processing; 2010; p. 1–4.
[40] Ramos CCO, Souza AN, Papa JP, Falcao AX. Fast non-technical losses identification
through optimum-path forest. Intell Syst Appl Power Syst 2009:1–5.
[41] Monedero I, Biscarri F, Le´on C, Biscarri J, Mill´an R. Midas: detection of non-
technical losses in electrical consumption using neural networks and statistical
techniques. In: Proceedings of the international conference on computational
science and applications; 2006; p. 3984.
[42] Hakim SJS, Abdul RH, Ravanfar SA. Fault diagnosis on beam-like structures from
modal parameters using artificial neural networks. Measurement 2015;76:45–61.
[43] Vasyl K, Gennady P. Efficient variable selection batch pruning algorithm for
artificial neural networks. Chemom Intell Lab Syst 2015;149:10–6.
[44] Darwisha AAA, Hanafya TA, Attiad AA, Habashyd DM, El-Bakrya MY, El-Nahassd
MM. Optoelectronic performance and artificial neural networks (ANNs) modeling
of n-InSe/p-Si solar cell. Superlattices Microstruct 2015;83:299–309.
[45] Fenzaa AD, Sorrentinob A, Vitiello P. Application of artificial neural networks and
probability ellipse methods for damage detection using lamb waves. Compos Struct
2015;133:390–403.
588
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
[46] Dongalea TD, Patila KP, Vanjareb SR, Chavanb AR, Gaikwadc PK, Kamat RK.
Modelling of nanostructured memristor device characteristics using artificial neural
network (ANN). J Comput Sci 2015;11:82–90.
[47] Abbasi EAH, Mehmood F, Kamal M, Baig A. Pakistan Energy Vision 2035.
[48] Fourie JW, Calmeyer JE. A statistical method to minimize electrical energy losses
in a local electricity distribution network. In: Proceedings of the 7th IEEE
AFRICON conference Africa: Technology Innovation, Gaborone; 2004.
[49] Filho JR, Gontijo EM, Delaiba AC, Mazina E, Cabral JE, Pinto JPO. Fraud
identification in electricity company customers using decision trees. In:
Proceedings of the 2004 IEEE international conference on systems, man and
cybernetics; 2004; p. 3730–4.
[50] Galvan JR, Elices A, Munoz A, Czernichow T, Sanz- Bobi MA. System for detection
of abnormalities and fraud in customer consumption. In: Proceedings of the
Electric Power Conference; 1998.
[51] Nizar AH, Dong ZY, Zhao JH. Load profiling and data mining techniques in
electricity deregulated market. In: Proceedings of the IEEE power engineering
society general meeting; 2006; p. 1–7.
[52] Gerbec D, Gasperic S, Smon I, Gubina F. Allocation of the load profiles to
consumers using probabilistic neural networks. IEEE Trans Power Syst
2005;20:548–55.
[53] Jiang R, Tagaris H, Lachsz A, Jeffrey M. Wavelet based feature extraction and
multiple classifiers for electricity fraud detection. In: Proceedings of the IEEE/PES
transmission and distribution conference and exhibition; 2002; p. 2251–6.
[54] De S, Anand R, Naveen A, Moinuddin S. E-metering solution for checking energy
thefts and streamlining revenue collection in India. In: Proceedings of the IEEE
PES transmission and distribution conference and exposition. Dallas, TX; 2003; p.
654–8.
[55] Depuru SS, Wang L, Devabhaktuni V. A conceptual design using harmonics to
reduce pilfering of electricity. In: Proceedings of the IEEE PES general meeting,
Minneapolis; 2010.
[56] SSSR Depuru, Wang L, Devabhaktuni V. Electricity theft: overview, issues,
prevention and a smart meter based approach to control theft. Energy Policy
2011;39:1007–15.
[57] Wheeler R, Aitken S. Multiple algorithms for fraud detection. Knowl Based Syst
2000;13:93–9.
[58] Kou Y, Lu C-T, Sinvongwattana S, Huang Y-P. Survey of fraud detection techniques.
In: Proceedings of the IEEE international conference on networking, sensing and
control, Taiwan. 2004. p. 89–95.
[59] Fawcett T, Provost F. Adaptative fraud detection. Data Min Knowl Discov
1997;1:291–316.
[60] Art M, Ayuso M, Guillén M. Modeling different types of automobile insurance
frauds behavior in the spanish market. Insur Math Econ 1999;24:67–81.
[61] Daskalaki S, Kopanas I, Goudara M, Avouris N. Data mining for decision support
on customer insolvency in the telecommunication business. J Oper Res
2003;145:239–55.
[62] Brause R, Langsdorf T, Hepp M. Neural data mining for credit card fraud detection.
In: Proceedings of the 11th IEEE international conference on Tools with artificial
intelligence; 1999; p. 53–61.
[63] Kirkos E, Spathis C, Manolopoulos Y. Data mining techniques for the detection of
fraudulent financial statements. Expert Syst Appl 2007;32:995–1003.
[64] Burge P, Shawe-Taylor J. Detecting cellular fraud using adaptative prototypes. In:
Proceedings of the AI approaches to fraud detection and risk management; 1997; p.
9–13.
[65] Cabral J, Pinto J, Linares K, Pinto A. Methodology for fraud detection using rough
sets. In: Proceedings of the IEEE international conference on granular computing
(GRC); 2006; p. 246–9.
[66] Jawad N, Keem SY, Sieh KT, Syed AK, Malik M. Nontechnical loss detection for
metered customers in power utility using support vector machines. IEEE Trans
Power Deliv 2010;25:1162–71.
[67] Yap KS, Hussien Z, Mohamad A. Abnormalities and fraud electric meter detection
using hybrid support vector machine and genetic algorithm. In: Proceedings of the
third IASTED international conference: advances in computer science and tech-
nology, Phuket, Thailand: Iasted press; 2007.
[68] Cabral J, Pinto J, Gontijo EM, Reis J. Fraud detection in electrical energy
consumers using rough sets. In: Proceedings of the IEEE international conference
on systems, man, and cybernetics; 2004; p. 3625–9.
[69] Cabral J, Pinto J, Martins E, Pinto A. Fraud detection in high voltage electricity
consumers using data mining. In: Proceedings of the IEEE/PES transmission and
distribution conference and exposition (T & D); 2008; p. 1–5.
[70] Sforna M. Data mining in power company customer database. Electr Power Systs
Res 2000;55:201–9.
[71] Jiang R, Tagiris H, Lachsz A, Jeffrey M. Wavelet based features extraction and
multiple classifiers for electricity fraud detection. In: Proceedings of the transmis-
sion and distribution conference and exposition 2002: Asia pacific; 2002; p. 2251–
6.
[72] Kantardzic M. Data mining: concepts, models, methods and algorithms. 2nd ed.
Wiley-IEEE Press (August 16, 2011).
[73] Witthen I, Frank E. Data mining–practical machine learning tools and techniques
with java implementations. New York, San Mateo, CA: Morgan Kaufmann, Acad
Press; 2000.
[74] Inigo Monedero. Editorial. Recent advances in data mining. Eng Appl Artif Intell,
2006; p. 19.
[75] McCarthy J. Phenomenal data mining. Commun ACM 2000;43:75–9.
[76] Ramos S, Vale Z. Data mining techniques application in power distribution utilities.
In: IEEE/PES transmission and distribution conference and exposition; 2008; p.
1–8.
T. Ahmad
[77] Valero S, Ortiz M, Senabre C, Gabald´ on A, Garcia F. Classification filtering and
identification of electrical customer load pattern through the use of self-organizing
maps. IEEE Trans Power Syst 2006;21:1672–82.
[78] Othman ML, Aris I, Othman MR, Osman H. Rough-set-and-genetic-algorithm
based data mining and rule quality measure to hypothesize distance protective relay
operation characteristics from relay event report. Int J Electr Power Energy Syst
2011;33:1437–56.
[79] Energy Independence and Security Act of 2007 (EISA-2007), which was approved
by the US Congress in January 2007, and signed to law by President George W.
Bush in December 2007.
[80] Historical Census of Housing Tables, Units in Structure. U.S. Census Bureau,
Housing and Household Economic, Statistics Division; December 16, 2005.
[81] Smart meters and smart meter systems: a metering industry perspective, a joint
project of the EEI and AEIC Meter Committees, March 2011. Edison Electric
Institute, 701 Pennsylvania Avenue, N.W. Washington, D.C. 20004-2696.
[82] MacDonald F. Smart metering: the importance of creating a shared vision - start
with a clear view of the end game. Smart metering -making it happen; 2009; 1–15.
[83] Neenan B, Hemphill R. Societal benefits of smart metering investments. Electr J
2008:32–45.
[84] Torriti J. Demand response experience in Europe: policies, programmes and
implementation. Energy 2009:1–9.
[85] Abdulrahman H, Bauer P, Enslin JH. Smarttrafo and Flicker. In: Proceedings of the
589
Renewable and Sustainable Energy Reviews 72 (2017) 573–589
power systems conference and exposition; 2006; p. 803–9.
[86] Netherlands Normalisation-instituut. Netherlands Technical Agreement NTA 8130,
Minimum set of functions for metering of electricity, gas and thermal energy for
domestic customers; 2007.
[87] Dick AJ. Theft of electricity-how UK electricity companies detect and deter. In:
Proceedings of the European Convention on Security and Detection; 1995; p. 90–5.
[88] Kadurek P, Blom J, Cobben JFG, Kling WL. Theft detection and smart metering
practices and expectations in the Netherlands. In: Proceedings of the Innovative
Smart Grid Technologies Conference Europe (ISGT Europe); 2010; p. 11662014.
[89] Haykin S. Neural networks: a comprehensive foundation. India: Prentice Hall;
1994.
[90] Kuncheva LI. Combining pattern classifiers: methods and algorithms. New York,
NY, USA: Wiley-Interscience; 2004.
[91] Reyzin L, Schapire R. How boosting the margin can also boost classifier complexity.
In: Proceedings of the 23th international conference on machine learning. New
York, NY, USA: ACM Press; 2006; p. 753–60.
[92] Papa J, Falc A, Suzuki C. Supervised pattern classification based on optimum-path
forest. Int J Imaging Syst Technol 2009;19:120–31.
[93] Boser B, Guyon I, Vapnik V. A training algorithm for optimal margin classifiers. In:
Proceedings of the 5th workshop on computational learning theory. New York, NY,
USA: ACM Press; 1992; p. 144–52.