Talal Saleh
Modern Fault Diagnosis in Power Systems Basedon 5G Networks
School of Technology and Innovat_ion
Master’s thesis in Smart Energy
Master of Science in Technology
Vaasa 2022
2VAASAN YLIOPISTOSchool of Technology and Innovat_ionsAuthor: Talal SalehThesis t_itle: Modern Fault Diagnosis in Power Systems Based on 5G NetworksDegree: Master of Science in TechnologySupervisor: Petri Va¨lisuo, Mohammed Elmusrat_iYear of graduat_ion: 2022 Number of pages: 59ABSTRACT:The future power system will be dynamic, requiring intelligent control, reliable protect_ion,and fast communicat_ion. Modern concepts in power systems, such as smart grids, involve bi-direct_ional power flow and two-way communicat_ion. Convent_ional protect_ion schemes andfault diagnosis methods are unsuitable for future power systems. This study proposes a mod-ern fault diagnosis that integrates 5G’s reliable communicat_ion and AI. 5G’s URLLC, mMTC,and edge comput_ing can bring significant advantages to the applicat_ions of power systems.In this study, a concept of intelligent fault diagnosis is proposed, which ut_ilizes 5G networkand AI. This work is divided into two main sect_ions. The first sect_ion develops an ML-basedpower system protect_ion model on MATLAB, and the second sect_ion deals with simulat_ing 5Gcommunicat_ion network on OMNeT++. ML algorithm developed for power system protect_ionachieved fault detect_ion accuracy of 99% and isolated faults within 7ms. The standalone 5Gnetwork without edge comput_ing server achieved round trip network latency of 20 ms.
Keywords: 5G Network, Modern Fault Diagnosis, SVM, ML, MATLAB/Simulink, OMNeT++, Smart Grid
3Contents
List of Figures 5
List of Tables 6
1 Introduct_ion 9
1.1 Mot_ivat_ion 9
1.2 Research object_ives 11
1.3 Thesis structure 11
2 Communicat_ion roles in power system and 5G networks 13
2.1 Communicat_ion Roles 13
2.2 Grid Infrastructure and Communicat_ion Technologies 14
2.2.1 Wired and Wireless Technologies 16
2.2.2 Fibre opt_ics 16
2.2.3 Power line communicat_ion 17
2.2.4 ZigBee 17
2.2.5 WLAN 18
2.2.6 WiMAX 18
2.2.7 Cellular communicat_ion technology 18
2.3 Fif_th Generat_ion Networks (5G) 20
2.3.1 5G Architecture 21
2.3.2 Mult_i-Access Edge Comput_ing and 5G 24
2.3.3 5G in Power Systems 24
3 Modern Fault Diagnosis 27
3.1 Protect_ion issues and challenges 28
3.1.1 Blinding Protect_ion 30
3.1.2 Sympathet_ic Tripping 30
3.1.3 Reach of Distance Relay 30
3.2 Evolut_ion of protect_ion schemes 30
43.3 Machine Learning and Power Systems 32
3.3.1 Machine Learning in Power System Fault Diagnosis 35
3.3.2 ML algorithm in Modern Fault Diagnosis 38
3.4 ML based Power system protect_ion 38
3.4.1 Datasets 39
3.4.2 Machine Learning Algorithm 40
3.4.3 MATLAB/Simulink Protect_ion model 41
4 System Simulat_ions and Results 43
4.1 5G Simulators 43
4.1.1 Synthet_icNET Simulator 43
4.1.2 OMNeT++ 44
4.1.3 NS-3 45
4.1.4 MATLAB/Simulink 45
4.1.5 OPNET 46
4.2 Power System Test Case with 5G Standalone Architecture 46
4.3 Results 48
4.3.1 5G Communicat_ion Network Results 49
4.3.2 ML based Power System Protect_ion Results 50
5 Conclusion 54
Bibliography 55
5List of Figures
Figure 1 5G Key technologies Yrjola and Jet_te (2018) 10
Figure 2 NIST conceptal grid infrastructure Lo´pez, Moura, Moreno, and Ca-
macho (2014) 15
Figure 3 Cellular network evolut_ion 19
Figure 4 Cellular network main component Peterson and Sunay (2020)) 21
Figure 5 5G with UE connect_ivity process a) BS detect and connect with UE
b) BS establish control plane connect_ion between UE and Core c) BS estab-
lish tunneling d) Tunneled over SCTP/IP and GTP/UDP/Ip e) UE handover
f) UE mult_ipath transmission Peterson and Sunay (2020) 23
Figure 6 5G Network Architecture in power system Cosovic, Tsitsimelis, Vuko-
bratovic, Matamoros, and Anton-Haro (2017) 26
Figure 7 Faults in AC overhead transmission system Eskandarpour and Kho-
daei (2018) 27
Figure 8 Microgrid protect_ion issues, challenges, and protect_ion solut_ion
Patnaik, Mishra, Bansal, and Jena (2020) 29
Figure 9 Microgrid operat_ion protect_ion schemes Patnaik et al. (2020) 31
Figure 10 IDT relay structure Sepehrirad, Ebrahimi, Alibeiki, and Ranjbar (2020) 32
Figure 11 Power system protect_ion with 5G network and ML 39
Figure 12 Simulink power distribut_ion line model 40
Figure 13 ML based fault sensing and isolat_ion 41
Figure 14 OMNeT++ Test case network with SA 47
Figure 15 OMNeT++ BS and UE connect_ivity Process 48
Figure 16 Packet transfer from UE1 to internet 49
Figure 17 Packet transfer from internet to UE2 50
Figure 18 Packet generat_ion by UE 1 (relay) 50
Figure 19 a) Decision boundaries b) 3D project_ion of target values 51
Figure 20 Simulink-ML load isolat_ion output 52
6List of Tables
Table 1 Communicat_ion technology comparison 19
Table 2 5G Simulator Comparison 46
Table 3 Communicat_ion network parameter and latency) 51
Table 4 Accuracy Scores 52
Acronyms
3GPP Third Generat_ion Partnership Project
5G Fif_th Generat_ion
AI Art_ificial Intelligence
ANN Art_ificial Neural Networks
AR Augmented Reality
B5G Beyond Fif_th Generat_ion
BS Base Stat_ion
CN Core Network
DER Distributed Energy Resource
DG Distributed Generator
eMBB enhanced Mobile Broad Band
eNB eNodeB
EPC Evolved Packed Core
gNB gNodeB
HAN Home Area Network
7HIF High Impedance Fault
ICT Informat_ion and Communicat_ion Technology
IED Intelligent Electronic Device
MEC Mult_i-Access Edge Comput_ing
mMTC Massive Machine Type Communicat_ion
ML Machine Learning
NAN Neighbourhood Area Network
NG-Core Next Generat_ion Core
NIST Nat_ional Inst_itute of Standards and Technology
NSA Non-Standalone
PCA Principal Component Analysis
PD Protect_ion Device
PLC Power Line Communicat_ion
QoS Quality of Service
RAN Radio Access Network
SA Stand-Alone
SVM Support Vector Machine
TLA Three Let_ter Acronym
UE User Equipment
UPF User Plane Funct_ion
URLLC Ultra-Reliable Low Latency Communicat_ion
8VR Virtual Reality
VM Virtual Machine
WAN wide area network
WiMAX Worldwide Interoperability for Microwave Access
WLAN Wireless Local Area Network
91 Introduct_ion
To fight rising global warming challenges, increasing energy demand, issues of power sys-
tem reliability and stability, the concept of smart and reliable power system is introduced
and realized as a solut_ion. Power systems are gradually moving toward future and intelli-
gent power systems. Tradit_ional power systems are centralized and stat_ic in nature with
radial configurat_ion Hussain, Nasir, Vasquez, and Guerrero (2020). Basic architecture con-
sists of generat_ing units (electricity generators) at one end and loads (consumers) on the
other. Power flow in a tradit_ional power system is unidirect_ional and system topology
is usually same all the t_ime. On the other hand, future power system can be consid-
ered as more decentralized power system in which many distributed energy resources
(DERs) such as wind, solar, and energy storages are integrated into the grid. For the fu-
ture power systems to be environmentally friendly and producing green energy, many re-
newable generators are integrated. Other than producers and consumers, future power
system creates roles such as prosumers and aggregators. Increase in penetrat_ion of DERs
gives rise to many challenges in power system protect_ion, control, reliability, and stabil-
ity Cisneros-Saldana, Samal, Singh, Begovic, and Samantaray (2022). Bidirect_ional power
flow, inert_ia less generat_ion, and changing network topology are some of the challenges
which needs to be considered.
1.1 Mot_ivat_ion
Due to the changing power system dynamics and increasing bidirect_ional power flow,
modern protect_ion schemes are needed. Frequent distributed generators connect_ion
and disconnect_ion may affect set_t_ings of protect_ion system and protect_ion devices Telukunta,
Pradhan, Agrawal, Singh, and Srivani (2017)Coster, Myrzik, and Kling (2010). Bidirect_ional
power flow in future power system may cause increase in fault tripping of protect_ion de-
vices. Similarly, at Microgrid level sympathet_ic tripping and false operat_ion of impedance
relay can raise serious concerns regarding the protect_ion of microgrid.
10
To manage the future power system challenges, not only improved protect_ion schemes
are needed but advance informat_ion and communicat_ion technologies (ICTs) can play a
vital role. Recent development in wireless 5G communicat_ion and its features such as
ultra-reliable low latency communicat_ion (URLLC) and massive machine type commu-
nicat_ion (mMTC) can be of great advantage to the challenges of future power systems
Patnaik et al. (2020).
Figure 1. 5G Key technologies Yrjola and Jet_te (2018).
Future smart communit_ies will have millions of devices connected and communicat_ing
with each other. Integrat_ion of ICT in power grid will allow devices to have two-way
communicat_ion which will enable grid’s self-healing capability and act_ive part_icipat_ion
of customers Liu, Yang, Wen, and Xia (2021). Different possible communicat_ion technolo-
gies including ZigBee, WLAN, Cellular and fibre-opt_ic can be used as a communicat_ion
medium. Fibre-opt_ic can be a solut_ion, but its economic factor is one of the concerns.
However, af_ter looking from the economic perspect_ive and considering requirements of
future power system regarding massive device communicat_ion, low latency applicat_ion,
and possible plug-and-play opportunity, 5G technology is seen as a potent_ial solut_ion.
11
1.2 Research object_ives
The object_ive of the research is to study 5G role in power system and more specifically
in protect_ion of future power system. By literature it is known that tradit_ional protec-
t_ion schemes are not suitable for future power systems with bidirect_ional power flow
and increasing DER integrat_ion Hussain et al. (2020) Patnaik et al. (2020) Telukunta et al.
(2017) Gomes, Coelho, and Moreira (2019) Tet_teh and Awodele (2019). Therefore, mod-
ern protect_ion schemes need to be developed that can tackle protect_ion challenges. In
this research modern fault diagnosis scheme is proposed which ut_ilizes wireless 5G as an
essent_ial component of the protect_ion scheme. The research object_ives can be catego-
rized into following points: -
1. Power system applicat_ion are low latency applicat_ions. Primary object_ive of the
research is to compute 5G network latencies. With the help of Simu5G simulator,
power system protect_ion test case model is built, and latencies are computed.
2. Developing non-MEC based network design for user equipment (UE) communica-
t_ion with 5G
3. Studying possibility of modern fault diagnosis based on Art_ificial Intelligence (AI).
4. Developing power system fault dataset and Machine Learning (ML) algorithm for
fault diagnosis.
5. Deploying ML algorithm in Simulink and test_ing by generat_ing art_ificial fault.
1.3 Thesis structure
Forthcoming thesis is divided into three chapters. Chapter 2 introduces role of communi-
cat_ion in power systems. It focuses on communicat_ion challenges and possible commu-
nicat_ion technologies that could be adopted. The chapter also presents details about 5G
technology, its key features, and how it can solve future power system challenges.
12
In chapter 3 the concept of modern fault diagnosis is presented. The chapter also reviews
tradit_ional protect_ion schemes and progression to modern protect_ion schemes. Since the
proposed concept is based on ut_ilizing AI, the chapter also discusses ML algorithm and
fault classificat_ion technique. Implementat_ion of developed algorithm on Simulink is also
discussed in the chapter.
The final chapter 4 is based on computat_ion of 5G test case latencies. In opening part in-
troduct_ion of 5G simulators are presented. A network design for modern fault diagnosis,
without MEC server, is developed on simu5G simulator. In the end of the chapter results
of Machine learning based fault isolat_ion and latencies are presented.
13
2 Communicat_ion roles in power system and 5G networks
Shif_ting from tradit_ional grid to a smarter grid require improvement in the power and
communicat_ion infrastructure. Current power grid can be described as a unidirect_ional
power flow system with a limited communicat_ion capabilit_ies in small segments Wang,
Xu, and Khanna (2011). A smarter grid-based infrastructure will allow bidirect_ional power
flow and two-way communicat_ion between power grids and electricity customers. It is
important to ment_ion that to achieve a smarter grid, one of the key solut_ions is to design
and develop smart grid communicat_ion infrastructure.
This chapter highlights the importance of communicat_ion in future power systems and
presents possible communicat_ion technologies available. The chapter presents 5G net-
work as a solut_ion to future power system challenges, specifically in low latency applica-
t_ion and massive machine type communicat_ion.
2.1 Communicat_ion Roles
In a power system operat_ion, control, and management, communicat_ion infrastructure
plays a crit_ical role. Tradit_ional power system ut_ilizes communicat_ion infrastructure in
limited boundaries. Ericsson (2002) classified communicat_ion role of tradit_ional power
system into three categories: -
1. Real-t_ime operat_ional communicat_ion: - It includes teleprotect_ion and power sys-
tem control of a tradit_ional grid. In case of teleprotect_ion the minimum latency of
a communicat_ion system should be less than 12-20 ms Ericsson (2002). Whereas
communicat_ion role in power system control includes communicat_ion in supervi-
sory control (SCADA) or energy management system (EMS).
2. Grid administrat_ive operat_ional communicat_ion: - This category of communicat_ion
role does not demand strict real-t_ime communicat_ion in a power system. The role
14
of communicat_ion might include administrat_ion of substat_ion cameras, ident_ifying
fault locat_ion, and power grid security systems.
3. Administrat_ive communicat_ion: - Which includes telephony or email communica-
t_ion.
For these communicat_ion roles as well as other possible future communicat_ion roles it
is important that the communicat_ion media and the communicat_ion technology used is
capable to fulfil the requirements of these communicat_ion roles. Moreover, the focus of
this research work falls in the area of real-t_ime operat_ion communicat_ion, and specifically
for communicat_ion in power system protect_ion. In upcoming sect_ions a brief introduct_ion
and comparison of the communicat_ion technologies is presented.
2.2 Grid Infrastructure and Communicat_ion Technologies
In a future power system, communicat_ion infrastructure will be a key component for re-
liable, efficient, and intelligent grid infrastructure. The concept of smart grid in future
power system will have an addit_ional digital layer added to it Deshpande and Raviprakasha
(2015) . A tradit_ional power system is divided into four domains namely electricity gener-
at_ion, transmission, distribut_ion, and customers. Energy flows unidirect_ionally between
these four domains. Usually in a tradit_ional grid there is no communicat_ion link between
all these four domains.
Nat_ional Inst_itute of Standards and Technology (NIST) presented a conceptual model of
smart grid in 2014. According to NIST a future power system can be divide into seven do-
mains, with an addit_ional communicat_ion or digital layer added to the grid infrastructure.
These seven domains include four physical layers namely generat_ion, transmission, distri-
but_ion, customer, and an addit_ional three upper domain layers providing an informat_ion
communicat_ion infrastructure along with electricity services Liu et al. (2021). These three
layers are named as electricity markets, service providers (aggregators, retailers or other),
15
and operat_ions (managers of power flow).
Figure 2. NIST conceptal grid infrastructure Lo´pez et al. (2014).
A tradit_ional grid with the physical network can be considered as a body with which en-
ergy flows from one end of the network to the other end. However, tradit_ional grid lacks
reliable and flexible communicat_ion infrastructure. The addit_ion of communicat_ion layer
can be considered as adding the soul and mind of the power network. With the inte-
grat_ion of advance informat_ion communicat_ion technology future power grid will have
improved power quality, high reliability, and addit_ion of intelligence in the power net-
work with real-t_ime informat_ion interact_ion, load management, electricity trading and
many other services Liu et al. (2021).
According to Gungor et al. (2011), for informat_ion flow between power network two types
of infrastructure is needed: - i) Flow of informat_ion from sensors and electrical appli-
ances to smart meters, and ii) Informat_ion flow between smart meters and ut_ility data
centres. However, other possible infrastructure might include, for example, infrastruc-
ture for millions of Intelligent Electronic Devices (IED) communicat_ing with each other
and infrastructure to integrate edge comput_ing in power system.
Developing communicat_ion infrastructure will require reliable communicat_ion technolo-
16
gies. Different communicat_ion technologies are available with medium of communicat_ion
as wired communicat_ion or wireless communicat_ion. Each communicat_ion technologies
have their own pros and cons but select_ing which technology is suitable for grid appli-
cat_ion requires the technologies to be studied from technical, environmental, and eco-
nomical perspect_ives. In upcoming sub-sect_ions an overview of possible communicat_ion
technology for future power system is presented by concisely going through the technical
and economic perspect_ives of the technology.
2.2.1 Wired and Wireless Technologies
In wired technology communicat_ion between devices or transfer of data from one device
to other is carried out through a wired-based technology. Two of the wired technology
are commonly used in grid communicat_ion namely power line communicat_ion (PLC) and
Fibre-opt_ic communicat_ion.
On the other hand Wireless technology, as the name highlights, forms a wireless network
and allows devices to communicate with each other. It is an alternat_ive of wired tech-
nologies which does not require a wired connect_ion and offers a plug-and-play opt_ion.
Different wireless technologies are available which can be used for future grid applica-
t_ions including ZigBee, Wireless local area network (WLAN), Worldwide Interoperability
for Microwave Access (WiMAX), and cellular networks.
2.2.2 Fibre opt_ics
Fibre-opt_ics communicat_ion uses opt_ical fibre cables for communicat_ion. Informat_ion
travel through fibres by means of light pulses. Fibre-opt_ic is a reliable media of communi-
cat_ion with high data rate and less communicat_ion interference. In smart grid fibre opt_ics
are used mainly in wide area network (WAN) and neighbourhood area network (NAN) Liu
et al. (2021). It is mostly deployed for communicat_ion at control centre and substat_ion.
One of the most advantage feature of fibre-opt_ics is that it provides low-latency com-
17
municat_ion with high data rate over a hundred of kilometre free from electromagnet_ic
interference.
However, along with its advantages it also has disadvantages. Fibre-opt_ics are very ex-
pensive in terms of deployment and maintenance when compared with other wireless
technology. Other disadvantage of fibre-opt_ics is that in future power system millions of
devices will need to communicate with each other and fibre-opt_ics cannot introduce plug-
and-play feature. For communicat_ion of millions of devices many opt_ical fibres would be
deployed, which will not be a pract_ical and economical solut_ion.
2.2.3 Power line communicat_ion
Other type of wired-communicat_ion technology is power line communicat_ion (PLC). PLC
ut_ilizes exist_ing power lines to communicate between devices by transferring data over
the power lines. The data can be transferred with this technology as narrowband PLC
(between 3-5 kHz frequency) with low data rate or broadband PLC (between 2 – 259 MHz)
with medium data rate Liu et al. (2021). Benefit of PLC technology is that addit_ional lines
are not needed, and data can travel over exist_ing power lines. Thus, being economical
and offering a plug-and-play opt_ion. However, power lines have high electromagnet_ic
interference. Transfer of data through these power lines undergoes high noise and signal
interferences. Moreover, number of devices connected to the power line and distance
between transmit_ter and receiver, both has negat_ive impact on the signal quality Gungor
et al. (2011).
2.2.4 ZigBee
ZigBee technology can provide wireless communicat_ion in power system with ultralow
power consumpt_ion. It is based on IEEE 802.15.4 standard. It has many pract_ical usages
in industries, home and buildings automat_ion, health care system and in electricity net-
works as well. Despite being economical communicat_ion technology, some of the con-
18
straints for ZigBee technology includes less security, less distance coverage, high latency,
and less devices connect_ion, which limits its applicat_ion in power system.
2.2.5 WLAN
WLAN is a wireless local area network technology which is based on IEEE 802.11 stan-
dards and is used mainly to perform point-to-point or point-to-mult_ipoint communicat_ion
within a home area network (HAN) Liu et al. (2021). Advantages of WLAN network in-
cludes low cost of deployment, plug-and-play opt_ion, latency in 15 ms, and medium-high
data rate. However, disadvantages of WLAN include less area coverage, signal distort_ion
by high-voltage equipment or signal interference, and up to 250 device connect_ions for
a single router.
2.2.6 WiMAX
WiMAX technology is capable to communicate similar to WLAN but with higher data rate,
thousands of device connect_ion, and longer distance communicat_ion. WiMAX is based on
microwave access technology and follows an IEEE 802.16 wireless standard. It can provide
point-to-mult_ipoint connect_ion and can be adopted as NAN or WAN. Disadvantage of
WiMAX includes high deployment cost and high latency as compared to other wireless
technologies.
2.2.7 Cellular communicat_ion technology
Cellular networks are also one of the technology available to play its role in power sys-
tem communicat_ion. Due to its already exist_ing infrastructure in most of the areas and
network easily accessed, it can be an economical as well as pract_ical solut_ion. Cellular
network evolved from 1G (first generat_ion) to all the way up to 6G. These networks evolu-
t_ions are governed by 3rd Generat_ion Partnership Project (3GPP) standards, which unites
19
7 internat_ional telecommunicat_ion standards including ARIB, ATIS, CCSA, ETSI, TSDSI, TTA,
and TTC.
Figure 3. Cellular network evolut_ion.
Cellular network infrastructure consists of four major parts including radio access net-
work (RAN), core network (CN), operat_ion and management, and user equipment (UE).
Its base stat_ion deployed at various locat_ions gives access to cellular networks and allows
devices to communicate with ease. Moreover, recent advancement of 5G and B5G net-
works, including high data rate, mMTC and URLLC, can be of key importance for many
smart grid applicat_ions. However, every technology has limitat_ion, cellular networks also
have some limitat_ion. 5G networks can travel shorter distances and are significantly in-
terrupted by physical objects such as huge building, walls, and towers. 5G networks also
have other vulnerabilit_ies such as ident_ificat_ion at_tacks and bat_tery drain at_tacks against
cellular devices Shaik, Borgaonkar, Park, and Seifert (2019).
Table 1. Communicat_ion technology comparison.
Technology Data rate Coverage Latency CostFibre-Opt_ic Very high up to 100 km 3 us/km HighPLC High 5 km 5 ms MediumZigBee (IEEE 802.15.4) Very high 70 m 3 ms MediumWLAN (IEEE 802.11ac) High 70 m 10 ms LowWiMAX Medium 30 km 50 ms High5G Very high Few hundred meters less than 1 ms Medium
20
2.3 Fif_th Generat_ion Networks (5G)
5G is a fif_th-generat_ion technology for cellular networks. 4G (LTE-Advanced) was capa-
ble to provide high data rate and network performances, but cellular networks had some
challenges which required further development. One of the main reason for develop-
ment of 5G was to introduce cellular networks for industry and informat_ion of things
(IoT) applicat_ions.
With an addit_ional requirement of high data rate, industry demanded possibility of large
devices interconnect_ivity and low latency communicat_ion over a wireless medium. Thus,
it led to an introduct_ion of triangle of 5G, present_ing enhanced Mobile Broad Band (eMBB),
mMTC, and URLLC Figure.1. eMBB deals with the applicat_ions which require high data rate
such as streaming 8K videos, virtual reality (VR), and augmented reality (AR). For smooth
running of these applicat_ion, it also requires low latency communicat_ion. Whereas mMTC
in 5G will allow around millions of devices to connect and communicate with each other
within a single cell.
Internat_ional telecommunicat_ion union (ITU) set requirements for 5G in Internat_ional mo-
bile telecommunicat_ion-2020 (IMT-2020). According to the requirements the peak data
rate for uplink was set to be 10 Gb/s and for downlink 20 Gb/s. Latency for eMBB was set
4ms and for URLLC 1ms. Moreover, 5G would have mobility up to 500km/h in rural eMBB.
Control plane latency was set around 10-20 ms. However, in current crowded and scarce
spectrum, the requirement of data rate of 20 Gb/s is not feasible in low-band spectrum
therefore 5G needed higher band spectrum to achieve this requirement.
Cellular network uses radio waves as a medium to transmit data. 5G network supports
frequency spectrum in the region of mmWave, low-band, and mid-band. mmWave lies in
frequency range of 24-72 GHz and can provide data rate up to 1-2 Gb/s Shahinzadeh et al.
(2020) at shorter distances. For longer distance 5G uses a mid-band frequency spectrum
which falls in the region of 2.4-4.2 GHz.
21
2.3.1 5G Architecture
Cellular network main components can be divided into two systems: RAN and Mobile
core Peterson and Sunay (2020). RAN is responsible of managing radio spectrum by assur-
ing that radio spectrum is used efficiently and provides required quality-of-service (QoS)
to the users. Different sub-components are combined to build up RAN, main component
of RAN is base stat_ion also referred as eNodeB (eNB) for 4G and gNodeB (gNB) for 5G.
Mobile core, on the other hand, provides connect_ion to internet for data and voice ser-
vices, ensures the QoS of the connect_ivity is met, track user’s mobility to avoid service
interrupt_ion, and monitors user’s billing and charging Peterson and Sunay (2020). Mobile
core is referred as Evolved Packed Core (EPC) in 4G and Next Generat_ion Core (NG-Core)
in 5G. Core act as bridge between RAN and the internet. Moreover, mobile core is also
further part_it_ioned into control plane and user plane.
Figure 4. Cellular network main component Peterson and Sunay (2020)).
To interconnect gNB with mobile core a Backhaul Network is used to establish a con-
nect_ion between RAN and mobile core. Backhaul is typical wired connect_ion. It is an
important part of the RAN; however, it is implementat_ion choice, and it is not prescribed
to implement Backhaul network according to 3GPP standard Peterson and Sunay (2020).
22
The connect_ivity services of cellular network are provided to the UE. A UE can be a moving
device or stat_ionery object, depending upon the type of applicat_ion. Previously UE were
name assigned to mobile phones and tablets, but recently objects such as car, drones,
industrial machines, intelligent electronic devices (IEDs), robots and many other devices
are also referred as UE when these are connected to cellular networks. These UE connects
to a base stat_ion when a wireless channel is established for a UE. The connect_ion remains
established unt_il channel is released if UE remains idle for a period of t_ime. In second step
each base stat_ions forwards a signaling traffic af_ter base stat_ion establishes “3GPP control
plane” connect_ivity with the UE and Mobile core. In third step for each act_ive UE, base
stat_ion establishes tunnels. Af_ter tunnel is established, base stat_ion forwards packets of
control plane and user plane between Core and the UE. However, these packets can be
tunneled over SCTP/IP and GTP/UDP/IP Peterson and Sunay (2020).
In case of a moving UE, a base stat_ion also coordinates a sof_t handover of the UE between
neighboring base stat_ions. However, during UE handover, base stat_ion might also coordi-
nate with mult_iple base stat_ions for a mult_ipoint transmission to a UE. The ent_ire process
of connect_ion of UE to handover can be seen in Figure5.
A 5G network can be deployed in mult_iple ways which can be summarized into Peterson
and Sunay (2020): -
1. Stand-Alone (SA) 5G
2. Non-Standalone (NSA) (4G + 5G RAN over EPC)
3. Non-Standalone (4G + 5G RAN over NG-Core)
These different 5G deployment methods allow transit_ion from 4G technology to 5G tech-
nology. In NSA over EPC, 5G base stat_ion is deployed along with 4G base stat_ion. 5G base
23
Figure 5. 5G with UE connect_ivity process a) BS detect and connect with UE b) BS establishcontrol plane connect_ion between UE and Core c) BS establish tunneling d) Tunneledover SCTP/IP and GTP/UDP/Ip e) UE handover f) UE mult_ipath transmission Petersonand Sunay (2020).
stat_ion is responsible for carrying user traffic and the 4G base stat_ion forwards control
plane traffic between UE and 4G Mobile Core. Non-Standalone over NG-Core uses 5G
Mobile core along with 4G base stat_ion. On the other hand, stand-alone has only 5G
technology deployed in the field.
24
2.3.2 Mult_i-Access Edge Comput_ing and 5G
Mult_i-Access Edge Comput_ing (MEC) is considered as an emerging technology for 5G and
Internet of Things (IoT). With MEC technology 5G can bring cloud comput_ing at the edge
of the RAN. Cloud comput_ing at the edge of RAN will provide computat_ional and storage
resources at RAN Liu Y, Peng, Shou, Chen, and Chen (2020) which will result in increased
computat_ion and reduce latency for the end users. In case of a cloud comput_ing far from
edge, the data would have to travel all the way to the cloud and return, which increases
the latency of communicat_ion.
MEC typical features includes proximity, low latency, high bandwidth, real-t_ime insight
into radio network informat_ion, awareness of user’s locat_ion, mobility support, and more
security and privacy protect_ion Liu Y et al. (2020). It can be a solut_ion to many applicat_ions
which either require low latency communicat_ion or are t_ime-sensit_ive applicat_ions such
as Augmented Reality (AR), Virtual Reality (VR), smart grid, and power system protec-
t_ion. However, to pract_ically integrated MEC with 5G and IoT some key technologies are
required namely Cloud Comput_ing, Sof_tware-Defined Network, Network Funct_ion Virtu-
alizat_ion, Informat_ion Centric Networking, Virtual Machine (VM) and containers, Smart
Devices, Network Slicing, and Computat_ional Offloading Liu Y et al. (2020).
This research study proposes a concept of ut_ilizing MEC servers of 5G for power system
fault diagnosis. More details are discussed in chapter 3.
2.3.3 5G in Power Systems
Authors in Garau et al. (2017) presented distributed monitoring and control of smart grid
based on 5G and 4G network. According to authors a smart grid architecture can be
categorised into centralized architecture, hierarchical architecture, and decentralized ar-
chitecture. The authors conducted research on centralized network management (with
LTE) and distributed network management (with 5G) and compared between the two
25
management systems during fault management in smart grid.
A communicat_ion network performance in power system can be evaluated based on
energy consumpt_ion, communicat_ion cost, processing requirement, bandwidth require-
ment, traffic and financial cost. Authors in Garau et al. (2017) highlighted benefits of de-
centralized network management over centralized network management by performing
analysis of these scenarios by combining DIgSILENT PowerFactory, Omnet++, MATLAB,
and Pyhton sof_tware tool.
Mostly power system protect_ion and control schemes are developed with wired commu-
nicat_ion technologies, however this could be challenging, economically and technically,
for future power systems. Authors in Hovila et al. (2019) introduced 5G applicat_ion in line
different_ial protect_ion. To validate communicat_ion performance and protect_ion system a
test setup was developed by the authors featuring wired (fibre connect_ion) and wireless
(5G) communicat_ion technology. Wireless technology included 5G test network located
in Espoo, Finland with core network and cloud network located in R&D premises of Nokia.
On the other hand, the line protect_ion scheme was the protect_ion scheme developed by
VTT’s Microturbine laboratory. By a connect_ion with 4G/5G modem R-SV and R-GOOSE
messages were sent over mobile network.
A 5G based architecture control and protect_ion applicat_ion of smart distribut_ion grid, us-
ing edge comput_ing, is also proposed in literature Zerihun and Helvik (2019). According to
the proposed architecture in Figure ??, the aim is to virtualize and transfer some control
funct_ion at the edge of cloud in 5G. Logical nodes in IEC 61850 are proposed to be kept
near field devices (sensors and actuators) and decision making funct_ionalit_ies on protec-
t_ion IEDs are moved to edge cloud. For substat_ion protect_ion end-to-end network slice
is taken into considerat_ion which is isolated from rest of network. According to authors,
for establishing 5G communicat_ion network architecture, main components required in-
cludes Field Devices, Radio Links, gNB, Edge Server, VM, Virtual Machine Monitor (VMM),
SDN Controller and communicat_ion link.
26
Figure 6. 5G Network Architecture in power system Cosovic et al. (2017).
5G and B5G networks are evolving communicat_ion technologies that are considered suit-
able for real-t_ime and mission-crit_ical applicat_ions of future smart grids. 5G’s mMTC and
edge comput_ing provide a suitable environment for smart grid to implement distributed
monitoring and control applicat_ions Cosovic et al. (2017). Authors in Cosovic et al. (2017)
presented detailed discussion of benefits of 5G environment based distributed state es-
t_imat_ion in smart grids. The authors concluded that 5G technology is capable of provide
an ideal arena for development in future distributed smart grid services.
27
3 Modern Fault Diagnosis
In a normal condit_ion, power systems operate under balance condit_ions but due to an
insulat_ion breakdown between conductors or physical contact due to an accident can
result in imbalance in power system dynamics which is referred to as occurrence of a
fault. Faults can be divided into symmetrical and asymmetrical faults. In 3-phase AC
power system fault can occur between phase-ground (P-G), Phase-Phase (P-P), or 3-Phase
fault. Similarly, DC faults in DC power system can be represented as pole-to-pole or pole-
to-ground fault Figure 1. Such faults change the dynamics of power system and causes
disturbances in the system stability. To minimize the harmful impact of faults on the
system stability and equipment life, protect_ion schemes and protect_ion devices are used
to isolate the fault before any damage has occurred.
Figure 7. Faults in AC overhead transmission system Eskandarpour and Khodaei (2018).
For modern concepts of smart grid such as integrat_ion of micro-grid which involves bidi-
rect_ional power flow and two-way communicat_ion, convent_ional protect_ion schemes or
fault diagnosis methods cannot be considered as suitable for future power systems. Con-
vent_ional protect_ion schemes are suitable for tradit_ional power systems which involves a
unidirect_ional power flow and unchanging network architecture throughout power sys-
tem. In case of bidirect_ional power flow and varying network typologies, modern fault
28
diagnosis needs to be studied.
Micro grids are important fragments of smart grid which can also be considered as a
small-scale design of large power systems. For the customer to be an act_ive part of the
smart grid, micro-grid allows the integrat_ion of DERs at distribut_ion level Patnaik et al.
(2020). It does not only increase the reliability of the system but allows the part_icipa-
t_ion of prosumers in energy market. A microgrid can generate enough power that could
supply energy to its nearby load. It can be operated in grid-connected or Islanded mode.
During its grid connected operat_ion it exchanges power with the main grid. Excessive
power produced by the microgrid is exported to the main grid and when the microgrid
has deficiency of power, it imports electricity from the grid. Hence, this leads to a bidi-
rect_ional power flow in a microgrid.
An intelligent or smarter grid can be considered as future power system which integrates
Intelligence, modern communicat_ion, and cyber technologies (including cyber physical
system & cyber security). Art_ificial Intelligence technique provides powerful tools for
fault control and fault diagnost_ic in future smart grids Bose (2017). In this chapter a power
system fault diagnosis is presented by using Machine Learning, a sub-branch of Art_ificial
Intelligence (AI). In this research, by integrat_ing AI and 5G cellular network, modern fault
diagnosis scheme based on 5G network is proposed.
3.1 Protect_ion issues and challenges
The bidirect_ional power flow disturbs the protect_ion set_t_ing of the system and protect_ion
devices (PDs). The distributed generators in microgrids, contribut_ing to fault current, can
cause irregular operat_ing t_ime of the PDs which will lead to loss in protect_ion coordina-
t_ion. Moreover, the increasing number of DGs of different sizes will generate fault current
of different levels which might also disturb protect_ion coordinat_ion. This dynamic nature
of smart grid with DG integrat_ion and islanding/grid-connected topology can have an
impact on current magnitude and direct_ion which might result in miss-coordinat_ion of
29
protect_ion devices Patnaik et al. (2020).
To develop microgrid’s reliability and stability, protect_ion of microgrid is an important el-
ement. Tradit_ional or convent_ional protect_ion schemes are not suitable to protect future
power systems. In Figure 8 the author has demonstrated all possible microgrid related
issues, challenges, and protect_ive solut_ions. Some of the protect_ion challenges which
are related to microgrid grid-connected mode of operat_ion are further discussed in the
following sub-sect_ions: -
Figure 8. Microgrid protect_ion issues, challenges, and protect_ion solut_ion Patnaik et al. (2020).
30
3.1.1 Blinding Protect_ion
During a fault if the locat_ion of DG is in the middle of fault point and feeding stat_ion, the
fault current which is sensed by the upstream protect_ion device will be of lower level.
Due to which the upstream protect_ion device will not respond to the fault and thus will
result in delay tripping or no tripping at all. This blinding protect_ion usually occurs at high
impedance areas Patnaik et al. (2020).
3.1.2 Sympathet_ic Tripping
In this type of tripping the protect_ion scheme loses its select_ivity and leads to isolat_ion of
connected DG unit or healthy feeder. One such case would be elevated level of fault cur-
rent init_iated from DG due to high resist_ive triple-line-ground fault at load feeder Patnaik
et al. (2020).
3.1.3 Reach of Distance Relay
Impedance relay is triggered when it senses a fault within its reach (maximum distance).
In case when DG is connected to the system, the distance relay might not operate on
its assigned zone. Because when the fault occurs downstream of DG control center the
upstream relay will detect extremely high impedance than real impedance which affects
relay grading, and the relay does not operate in the specified zones Patnaik et al. (2020).
3.2 Evolut_ion of protect_ion schemes
Relays and circuir breakers are two main components of the protect_ion system which are
the brain and the body. Relay act as a brain which senses disturbances and signals the
body (PD) to take certain act_ions.
31
In the paper Tet_teh and Awodele (2019) the author presented an overview of protect_ion
schemes of power system and their evolut_ion throughout the t_ime. The author discussed
earlier works such as classificat_ion and ident_ificat_ion of three-line currents using fuzzy
logic, fault classificat_ion and ident_ificat_ion on transmission line using wavelet and fuzzy
based systems, Integrated protect_ion system based on fuzzy inference rules, and adapt_ive
mult_i agent protect_ion relay coordinat_ion technique.
MG protect_ion schemes can be divided into three branches namely different_ial, distance
and over-current protect_ion Venkataramanan and Marnay (2008). However, Patnaik et
al. (2020) presented classificat_ion of all microgrid protect_ion strategies and protect_ion
coordinat_ion method that are present in literature Figure 9
Figure 9. Microgrid operat_ion protect_ion schemes Patnaik et al. (2020).
In recent work, an intelligent non-model-based different_ial relay is presented for con-
trolled Islanding of microgrids Sepehrirad et al. (2020). One of the issues of different_ial
protect_ion is that it requires real-t_ime adapt_ive and select_ive based different_ial relaying
for fault detect_ion and tripping Sepehrirad et al. (2020). The proposed intelligent De-
cision Tree based different_ial relay can ident_ify proper tripping t_iming with respect to
32
different microgrid operat_ion and topologies. According to the proposed scheme for Is-
landing scenario Figure 10, the electrical voltage and current signal are processed during
faulty events. By using intelligent decision tree algorithm, effect_ive signals are selected
to be used for different_ial protect_ion relay design.
Figure 10. IDT relay structure Sepehrirad et al. (2020).
The future power system will be of dynamic nature which will require intelligent control
& protect_ion and fast communicat_ion. It is important to ment_ion that for the protect_ion
of smart grids, grid intelligence, and fast & reliable communicat_ion will play a key role.
Recent development in cellular communicat_ion and its features such as ultra-reliable low
latency communicat_ion(URLLC)and massive machine type communicat_ion (mMTC) can
be of great advantage to improve the communicat_ion lag and reliability of the protect_ion
system Tet_teh and Awodele (2019).
3.3 Machine Learning and Power Systems
Machine learning (ML) is a branch of Art_ificial Intelligence. As the name suggest, in ML
the machine algorithm tries to learn from data or by experience. Based on that learning
33
the ML model predicts or takes necessary act_ions. Primary purpose of adopt_ing ML in
different areas of applicat_ion is that it provides automat_ic learning from a raw data and
produces results which can be implemented in decision making process Miraf_tabzadeh,
Foiadelli, Longo, and Paset_t_i (2019).
Machine learning tasks can be divided into four types namely, i) Supervised learning ii)
Unsupervised learning iii) Semi-supervised and iv) Reinforcement learning. Main differ-
ence between these four learning is that in supervised learning the output (target value)
is given to the model. Whereas, in case of unsupervised and reinforcement learning out-
put are not shown to the model. Reinforcement learning further ut_ilizes the concept
of reward and punishment to train the model without the need of input/output data.
Semi-supervised learning on the other hand contains some labelled output data and large
amount of unlabeled data. To adopt which task depends on the data and the research
problem.
Supervised ML can be applied to problems which require classificat_ion or regression. Clas-
sificat_ion deals with categorizing data into output classes and predict_ing the class of data.
Classificat_ion can be in the form of alphabet-based classes or numerical class. Later is spe-
cial type of regression in which data is classified in number classes. ML classificat_ion can
be either linear or non-linear classificat_ion depending up-on the data distribut_ion. Dif-
ferent type of classifiers such as determinist_ic or probabilist_ic classifiers can be used. An
example of classificat_ion problem could be to predict whether the image is an image of
“oranges” or “mangoes”. It can also be binary classificat_ion where the data belongs to a
class “0”, “1” or “2”.
Regression, on the other hand, in its simplest definit_ion can be defined as a method to
predict numerical output from a given data. It is also known as curve fit_t_ing in which the
algorithm tries to fit data with the curve. The idea behind regression is to fit available data
to a certain formula which will be good for data modelling; or by rephrasing it, in regres-
sion an approximate mathemat_ical model is build which is suitable for the input/output
relat_ionship. Based on the mathemat_ical modelling of data, regression can be divided
34
into Parametric regression and Blackbox regression. One example of regression problem
could be to forecast energy demand.
Growing applicat_ion of machine learning can be seen in power system control, secu-
rity, forecast_ing, fault diagnosis, and many other applicat_ions. Alimi, Ouahada, and Abu-
Mahfouz (2020) Presented a comprehensive review of machine learning techniques de-
veloped for power system stability and security applicat_ion. The research highlighted
ML-based methodologies, achievements, and limitat_ions of classifier design. There are
four major power system security and stability domains where mostly ML techniques are
deployed namely in SCADA network, Power quality disturbance (PQD) studies, Voltage
stability assessment (VSA) and transient stability assessment (TSA).
Different ML algorithms such as Art_ificial Neural Networks (ANN), Support Vector Ma-
chine (SVM), Decision Tree (DT) or other algorithms are used based on their performances.
In Sung and Ko (2015) an interact_ive machine learning integrated load control scheme
is proposed for improving performance and reliability of load scheduler and reducing
peak power demand. To adjust the load scheduler SVM algorithm was used to accurately
predict demand. Lasset_ter, Cot_illa-Sanchez, and Kim (2018) used ML SVM algorithm to
predict whether re-connect_ion of microgrid to the main grid would lead to stability or
not, based on real-t_ime phasor measurement units (PMU) values. For training the ML
model, the author created training dataset for different scenarios by using power sys-
tem dynamic simulator. Based on these values the model was trained and it predicted
whether the power system would lead to instability or stability when a microgrid will be
reconnected to the power system.
In Zhao, Shang, and Sun (2019) the author proposed that by accurate classificat_ion of
power quality disturbances, the power quality of a power system can be improved and
governed by using t_ime-frequency domain mult_i-feature and decision trees. Wavelet
transform and S-transform were performed to extract features from power quality dis-
turbance signals. Based on the extracted features, decision tree classifier algorithm was
applied to accurately classify power quality disturbances.
35
One of the challenging areas of power system is the power outages. In literature ML mod-
els have also been used to accurately predict power outages af_ter disturbances. There
are number of factors involved which can result in power outages such as environmental
factor, power line faults, or equipment damage. In Kankanala, Das, and Pahwa (2014) the
author used ADABOOST algorithm to est_imate power outages based on weather data. It
evaluated the effects of wind and lightning on power outages in overhead distribut_ion
system. In Eskandarpour and Khodaei (2018) power outages due to extreme weather
event, such as hurricanes, was used as one of the features to train a SVM model and
predict power outages.
3.3.1 Machine Learning in Power System Fault Diagnosis
Faults are likely to occur in any funct_ioning system which can damage the system and
make it unreliable for operat_ion. Important thing for system reliability and stability is that
the fault is diagnosed, located, and isolated. Similarly in a very complex power system it
is essent_ial to diagnose the fault and isolate it before any harm to power system stability
and equipment.
In literature many reviews and implementat_ion of AI applicat_ion in fault diagnosis are
present. According to Prasad, Belwin Edward, and Ravi (2018), classificat_ion of faults in
power system can be performed using three techniques: -
1. Prominent Techniques: - They are wavelet approach, ANN approach and Fuzzy logic
approach.
2. Hybrid Techniques: - It includes Neuro-Fuzzy Technique, Wavelet & ANN Technique
and others.
3. Modern Techniques: - Are the techniques implemented for fault analysis in power
system using SVM, Genet_ic Algorithm, PMU-based protect_ion scheme, Principal
Component Analysis (PCA) based framework and many other techniques.
36
Authors in Mishra and Rout (2018) presented a microgrid protect_ion scheme using ML
technique. According to authors the proposed microgrid protect_ion scheme is more ro-
bust than convent_ional overcurrent relay operat_ion for faulty events. In first step three
phase current signal of the respect_ive buses is passed through an empirical mode decom-
posit_ion method. By using Hilbert-Huang transform important features are computed
from decomposed signal, which are then applied to three different ML classifier algo-
rithm namely naive bayes classifier, support vector machine and extreme learning ma-
chine. Output result of classifier represents the detect_ion and classificat_ion of microgrid
faulty events. The proposed protect_ion scheme was tested for different types of faults
(symmetrical, asymmetrical), different microgrid structure (radial, mesh) and microgrid
mode of operat_ions (islanded and grid connected).
ANNs are also one of the classifiers which is applied for detect_ion and classificat_ion of
power system faults in literature. Jamil, Sharma, and Singh (2015) applied feedforward
Neural Network with backpropagat_ion algorithm in the process of three phase power line
fault detect_ion and classificat_ion. Datasets used for training and validat_ing the ANN model
was obtained by simulat_ing on MATLAB/Simulink a model of 400kV/50Hz three phase
power line with two generators located at both ends. The length of the transmission line
was considered 300 km and various types of faults were applied at different locat_ions
of transmission line. The authors presented results for both fault detect_ion and fault
classificat_ion using ANN.
Disturbances caused by fault are crucial to be detected to enhance performance of mi-
crogrids. Panigrahi, Rout, Ray, and Kiran (2018) has introduced a technique for microgrid
fault detect_ion and classificat_ion. For the detect_ion of faults in microgrid, consist_ing of
wind turbine and diesel generator, Wavelet transform, and Wavelet packet transform is
used. To classify the faults a hybrid technique, Neuro-Fuzzy method is implemented.
Analysis and assessment of the proposed approach was conducted on MATLAB/Simulink
environment.
Increase in complexity of power system increases the rate of occurrence of faults in
37
power systems. Therefore, for rapid fault ident_ificat_ion an adequate intelligent systems
are needed which can ident_ify faults. Kumar, Bag, Londhe, and Tikariha (2021) ut_ilized
machine learning algorithm for classificat_ion of faults. Current and voltage values were
taken from IEEE-14 bus standard system which was simulated in MATLAB/Simulink for
normal, symmetrical, and unsymmetrical fault scenarios. By training SVM algorithm dif-
ferent types of faults in power systems are classified with higher accuracy. In the pro-
posed technique, faults are analyzed using transient data, features are extracted with
wavelet packet transform and redundant features are removed to improve the accuracy
of the model.
In Goswami and Roy (2019) MATLAB/Simulink simulated faults were classified. The Simulink
model consisted of 90km power transmission line and the faults were applied at each
10km of line length. Based on the Simulink model, fault datasets with three phase RMS
voltage and current values were generated and total 11 types of faults classes was consid-
ered. Three ML algorithm including Decision Tree, K-Nearest Neighbors and SVM were
applied for the classificat_ion of faults using python and scikit-learn library. According to
the results presented by the authors, SVM algorithm outperformed the other two algo-
rithms
Sarwar et al. (2020) Proposed power distribut_ion network High Impedance Fault (HIF) de-
tect_ion and isolat_ion. Data from voltage and current sensors are used, which are applied
to data driven techniques such as Principal component analysis (PCA), Fisher discriminant
analysis (FDA), and mult_iclass-SVM, to detect and classify HIF. According to authors PCA
can detect HIF but cannot classify HIF. Fisher discriminant analysis can classify and detect
HIFs. However, bet_ter results are obtained using SVM for fault detect_ion and classifica-
t_ion.
Gururajapathy, Mokhlis, and Illias (2017) presented a review of most of the techniques
which are deployed and commonly used in locat_ing and detect_ing faults in power distri-
but_ion system with distributed generators. According to author the fault locat_ion method
can be categorized into convent_ional method and AI-based method. Convent_ional meth-
38
ods require less computat_ional t_ime but are inaccurate for larger power systems. AI-
based method has higher accuracies for larger power systems. Af_ter comparing between
different AI algorithm including ANN, SVM, Fuzzy logic, Genet_ic algorithm (GA), the au-
thors concluded that SVM algorithm is more widely used due to its successive progress in
recent years. But none of any single AI algorithm has the capability to solve all problems
based on specified condit_ions. Moreover, in this research SVM algorithm is considered
for applying fault diagnosis in power system protect_ion model.
3.3.2 ML algorithm in Modern Fault Diagnosis
The concept of modern fault diagnosis is to integrate intelligence and reliable communi-
cat_ion. By using 5G communicat_ion and ML algorithm modern fault diagnosis scheme is
proposed which can be a solut_ion to future power system challenges such as false trip-
ping due to bi-direct_ional power flow. The concept of modern fault diagnosis can be seen
in Figure 11 in which the ML algorithm can be deployed in MEC server. Incoming sensor
values will be given to the ML model deployed in MEC server, which will detect and clas-
sify the occurrence of fault. If fault is detected the MEC server will sent trip signal to the
actuator. Upon receiving a trip signal, an actuator will isolate the faulty region by tripping
the circuit breaker.
3.4 ML based Power system protect_ion
In this research a ML based protect_ion method is proposed to isolate a fault once the fault
is recognized by the ML algorithm. To train a ML model, three phase current and voltage
datasets are generated by a MATLAB/Simulink model. The Simulink model consists of
a 150kV 3-phase power source, 20km power line and load connected on the other end
Figure 5.
39
Figure 11. Power system protect_ion with 5G network and ML.
3.4.1 Datasets
Fault datasets are generated by art_ificially generat_ing different types of faults on the
power line. Four different types of faults namely Ph-Ph-Ph, 3-Ph-G, 2-Ph, and 2-Ph-G
faults were generated. The measurement of Load voltages and currents are exported
into a CSV file to which a machine learning classifier is applied.
The dataset contains 48007 data values with target value “0” or “1” which mean that
either there is no fault (0) or the fault has occurred (1). Total six values of measurement
(RMS current and voltage of each phase) are input data (features of ML algorithm). Before
training the SVM model, the datasets are pre-processed by applying random permutat_ion
and dividing into train, test, and validat_ion sets. The train and test_ing datasets are used
during the training and test_ing phase of ML model. Once the model has been trained and
tested, a separate validat_ion dataset is used to verify the performance of the model.
40
Figure 12. Simulink power distribut_ion line model.
3.4.2 Machine Learning Algorithm
A ML algorithm for power system fault diagnosis is developed in this research using SVM.
SVM is a data driven technique. Due to its generalizat_ion abilit_ies and less vulnerable to
dimensionality, it is used for detect_ion and classificat_ion of faults Sarwar et al. (2020). It
can classify samples by creat_ing hyperplane, also known as decision boundaries, which
separates one class from other Vaish, Dwivedi, Tewari, and Tripathi (2021).
SVM is a kernel method which uses kernel funct_ion to map two points from an original
space into higher dimension Xie, Alvarez-Fernandez, and Sun (2020). In other words, it
uses kernel funct_ion to transform non-linearly separable samples into separable samples
by convert_ing it into higher dimensions which are more likely to be separated Vaish et al.
(2021). In this study, ML SVM classifier algorithm is used to classify the faults.
SVM classifier is developed in MATLAB and Python. For visualizat_ion of data samples
and decision boundaries, Python is used. However, implementat_ion of SVM Classifier
on Simulink is carried out using MATLAB based SVM classifier. The results of both the
classifiers are also presented in Results sect_ion.
For visualizing dataset decision boundaries and possible linear separat_ion, PCA is used
in Python. PCA is a dimensionality reduct_ion or data compression method, however, in
fault diagnosis it operates as pat_tern recognit_ion Vaish et al. (2021).
41
Both the SVM models are developed by using same model parameters. Radial basis func-
t_ion is used by the SVM classifier to ident_ify the decision boundaries for classes. The C
parameter value is kept 1. It informs the SVM model about the amount of flexibility in
misclassifiying training samples. Higher C-value will select a smaller-margin hyperplane
which will limit the amount of support vectors. Conversely, lower C-value will cause SVM-
opt_imizer to have larger-margin in hyperplane. Select_ing C-value depends upon the train-
ing dataset and respect_ive classes.
3.4.3 MATLAB/Simulink Protect_ion model
To classify the fault and isolate the load from faulted line SVM model is developed in MAT-
LAB and applied in Simulink. The SVM Simulink block takes three phase load current and
voltage as input values and predicts the output, which means the block predicts whether
the fault has occurred or not. As soon as the model predicts the occurrence of fault, it
signals the actuator (circuit breaker) to isolate the load. The SVM model can be labeled
as an intelligent relay which senses the fault and signals the circuit breaker to isolates the
load.
Figure 13. ML based fault sensing and isolat_ion.
Since 5G will be used as medium of communicat_ion and the ML model will be deployed
42
at 5G’s MEC server, it is important to study latency of communicat_ion. The next chapter
presents 5G simulators, results of latency and ML based protect_ion of MATLAB/Simulink
model.
43
4 System Simulat_ions and Results
One way for evaluat_ion of network performance is by simulat_ion of communicat_ion net-
works. The results obtained with simulat_ion might not be as realist_ic as in pract_ical cases,
but it gives an overview of how the system will preform and react under certain condi-
t_ions. Cellular networks are complex communicat_ion system in which each component
has an impact on network performances. By sof_tware modeling of these component
and complex systems an analysis can be made about the operat_ion and interact_ion of
such systems in real-world environment, which also minimizes cost and risk involved in
implementat_ion. Therefore, in this chapter simulat_ion of 5G network for power system
protect_ion test case is presented. The chapter also introduces 5G simulators and results
of ML based power system protect_ion and communicat_ion latencies.
4.1 5G Simulators
5G Simulat_ions sof_tware provides an analysis of how the 5G network and component
interact in the real-world environment. By means of 5G simulat_ion sof_tware one can
compute latencies, simulate cellular networks including mmWave/NR and LTE, perform
pathloss and interference modeling of mmWave, ad-hoc network, mult_i-hop network, ve-
hicular ad-hoc network (VANET), antenna design and many other applicat_ions. There are
many 5G simulators available for the researchers, but which simulat_ion sof_tware is suit-
able depends upon the applicat_ion and need of a researcher. Some of the well-known 5G
simulators include Synteht_icNET, Omnet++, ns-3, OPNET, OpenAirInterface, 5G-K, Mat-
lab/Simulink, Vienna-5G and so on.
4.1.1 Synthet_icNET Simulator
Synteht_icNET simulator is developed by AI4networks Lab which is a python-based sim-
ulator and conforms with the 3GPP 5G standard Zaidi, Manalastas, Farooq, and Imran
44
(2020). Since it uses python plat_form, it can integrate ML libraries of python and is also
considered as the first simulator which is built to test AI network automat_ion Zaidi et
al. (2020). It has other key features including detailed handover process implementa-
t_ion, integrat_ing edge comput_ing, propagat_ion modeling, and support for 5G standards.
However, as per the author’s knowledge Synteht_icNET simulator is not an open-source
sof_tware.
4.1.2 OMNeT++
OMNeT++ (Object_ive Modular Network Testbed in C++) is a discrete event network sim-
ulator which is writ_ten in C++ plat_form. Under academic license it is an open-source sof_t-
ware and free to use and modify. OMNeT++ can be used for wireless and wired computer
network simulat_ion. Most recent release of OMNeT++ allows integrat_ion of Python to
perform data visualizat_ion and plot_t_ing outputs. It models a communicat_ion network
by compiling init_ializat_ion file, network descript_ive file, and message file. OMNeT++ al-
lows networks to be modeled by combining reusable network components, also known
as modules. It provides many other funct_ionalit_ies including protocol modeling, model-
ing of queuing networks and distributed hardware systems, performance evaluat_ion of
complex systems, and many other features.
Simu5G is an evolut_ion of SimuLTE 4G simulator which is developed on OMNeT++ frame-
work to simulate 5G networks. It is a library of OMNeT++ and allows simulat_ion of data
plane of mobile core and RAN. Moreover, it is also compat_ible with other libraries of
OMNeT++ including INET (TCP/IP based network modeling) and Veins (for VANETS). For
packets to be send from a UE to gNB or for communicat_ion between vehicles, applicat_ion
layer can be developed by writ_ing the code in C++ language. Moreover, it is also based
on 3GPP specificat_ion, but it does not model control plane funct_ion.
45
4.1.3 NS-3
NS-3 simulator is based on NS-2 and is licensed under GNU GPLv2 license Bouras, Gka-
mas, and Diles (2020). It is also an open-source sof_tware available free for research and
development purposes. It is a discrete-event network simulator and provides a simula-
t_ion engine to perform network simulat_ions. NS-3 is developed on C++ plat_form and uses
Python language. Main features of NS-3 includes C++ & Python script_ing, Virtualizat_ion
support for virtual machines, sof_tware integrat_ion, support for 4G network by LTE pro-
tocol stack, inclusion of evolved packet core, and support for 5G network simulat_ion by
mmWave simulator module. Bouras et al. (2020).
However, some of its weaknesses, ment_ioned by Bouras et al. (2020), are lack of credi-
bility, lack of act_ive maintainers, lack of documentat_ion and community for fixing bugs.
4.1.4 MATLAB/Simulink
MATLAB is a powerful tool and programming plat_form which is used by engineers and
scient_ists to design products and systems. MATLAB is based on its own MATLAB languages
which is matrix-based language. It can be used to analyze data, develop algorithm, create
models and applicat_ions.
Simulink, integrated with MATLAB, is a block diagram environment which can be used for
applicat_ions in the area of automot_ive, aerospace, industrial automat_ion, signal process-
ing and communicat_ion network modeling. 5G Toolbox in Simulink provides opportunity
for modeling, simulat_ion, and verificat_ion of 5G network systems. 5G Toolbox is also 3GPP
compliant which allows funct_ionalit_ies such as simulat_ing effects of RF designs and inter-
ference, configure and simulate 5G NR communicat_ion link, and generate waveforms.
46
4.1.5 OPNET
OPNET (Opt_imized Network Engineering Tools) is also a simulator available at commercial
level and is part of Riverbed Modeler. It is useful in studying communicat_ion applicat_ions,
protocols, and networks. With its GUI a user can develop network topologies and applica-
t_ion layers and implement them using object-oriented programming. OPNET offers three
main funct_ionalit_ies namely modeling, simulat_ing and analysis. Moreover, some of its
strength also includes fast discrete event simulat_ion engine, less simulat_ion runt_ime, and
fast graphical results upon network simulat_ion.
Along with its benefits some of OPNET’s disadvantages are Bouras et al. (2020): -
1. Supports lesser nodes for a single device.
2. In case of long ideal period simulat_ion is inadequate.
3. Powerful GUI but complicated to use.
Table 2. 5G Simulator Comparison.
Simulators License Type Plat_formsSynthet_icNET Simulator - PythonOMNeT++ Open source (for research and academics) C++NS-3 Open source C++ and PythonMATLAB/Simulink Not open source MATLABOPNET Commercial C/C++
4.2 Power System Test Case with 5G Standalone Architecture
In this research work a communicat_ion network is developed for power system protect_ion
test case to compute latencies of 5G standalone network. OMNeT++ is considered in this
study because it is open source, have easy to use GUI and developed Simu5G library to
run modified test cases.
47
Power system protect_ion test case will have two main IEDs: - 1) Sensors (Relays) 2) Actu-
ator (Circuit breakers). With 5G network both the devices can communicate with each
other. Voltage and current values can be sent from an intelligent relay to an actuator
through gNB. The AI algorithm, deployed in the cloud, will be able to ident_ify if the power
system is running in normal condit_ion or if a fault has been detected. If a fault is predicted
by the AI algorithm, cloud will send a fault signal (similar to a GOOSE message) to an ac-
tuator to trip the circuit breaker.
Figure 14. OMNeT++ Test case network with SA.
Above Figure 14 shows a simple architecture of single relay and actuator communicat_ion,
labeled as ue (relay) and ue1 (actuator). Each UE has a separate network interface card
(NIC) through which it connects and communicates with the 5G base stat_ion. Data from
the UE travel all the way through radio waves to base stat_ion, user plane funct_ion (UPF),
router and to the servers. The relay would send measurement of voltage and current in
the form packet to the cloud far from the edge. The server represented in the Figure 14
will have ML algorithm deployed which will process incoming packet measurement values
and predict occurrence of fault. An applicat_ion layer is developed at the server which
accepts incoming packets from relay and send another packet to the actuator containing
48
informat_ion to whether trip the circuit breaker or not.
Simu5G follows similar process of UE connect_ivity as ment_ioned in Figure 5. When sim-
ulat_ion is run, the gNB first tries to detect the UEs and their availability. If UEs are avail-
able for communicat_ion, BS sends a signal and establishes a connect_ion between UEs
and Core. The core in the Simu5G is represented by UPF. Upon successful establishing of
connect_ion the process of tunneling begins. The packets transferred across the network
using SCTP/IP and UDP/IP protocols. Since in power system test case the UEs are stat_ion-
ary object, therefore there would be no handover process and the communicat_ion would
cont_inue in similar fashion.
Figure 15. OMNeT++ BS and UE connect_ivity Process.
4.3 Results
The result sect_ion is divided into two parts. First sect_ion presents the result obtained by
5G SA network simulat_ion by OMNeT++/Simu5G. Second sect_ion present the results of
49
ML algorithm and protect_ion of power system distribut_ion model by ML on MATLAB/Simulink.
4.3.1 5G Communicat_ion Network Results
A SA non-MEC based network is simulated with two UE communicat_ing over 5G network.
Af_ter the connect_ivity has been established between UPF and UEs by gNB, the UE (sensor)
will start to send packets containing informat_ion of voltage & current measurement to the
dest_inat_ion address. However, to send and receive these packets in OMNeT++/Simu5G
the applicat_ion layers of these modules need to be programmed.
The UE (sensor) applicat_ion layer is programmed in a way that it cont_inues to send the
measurement packets to the internet server. The server applicat_ion layer is programmed
to receive the measurement packet, destroy it, and send a new alert packet to the sec-
ond UE (actuator). The second UE (actuator) applicat_ion layer is also developed which
receives incoming packets from the internet server.
Figure 16. Packet transfer from UE1 to internet.
50
Figure 17. Packet transfer from internet to UE2.
Figure 18. Packet generat_ion by UE 1 (relay).
Both the UEs were approximately 150m away from the gNB with carrier frequency of
4GHz and the simulat_ion was run for 20s. The Table 3 below shows the latency obtained
for these specificat_ions for a communicat_ion round from UE 1-to-Internet-to-UE 2.
4.3.2 ML based Power System Protect_ion Results
In this study two SVM ML models are developed. For visualizing the data and decision
boundary, python-SVM model is used and MATLAB-SVM model is used for deploying in
51
Table 3. Communicat_ion network parameter and latency).
Test case Distance from gNB to UE 1 and UE 2 Carrier frequency LatencySA Non-MEC 150m 4GHz 20ms
Simulink to isolate the fault in power system distribut_ion model. In python, the PCA-SVM
based ML model was able to accurately diagnose the occurrence of fault. PCA was applied
to reduce six-dimensional data to two dimension which was passed to SVM algorithm
which accurately classified the diagnosis of power line faults.
By visualizat_ion of data and decision boundaries of SVM classifier, using Principal com-
ponent analysis (PCA) and project_ing target values, it is observed that the data values
cannot be linearly separated in 2-dimension and 3-dimension Figure 19. Therefore, to
non-linearly separate the target with high accuracy, radial basis funct_ion kernel is consid-
ered. In Figure 19a the red dots represents fault target value ”1” and surrounding blue
dots represents no fault class ”0”. Similarly by looking in three dimension space Figure
19b yellow dots are the target value ”1” (fault class) and purple dots shows no-fault region
”0”.
Figure 19. a) Decision boundaries b) 3D project_ion of target values.
By looking at the results, both the models achieved high accuracies. These accuracies are
obtained using SVM scoring funct_ion. Each scores are obtained with three datasets, train-
52
Table 4. Accuracy Scores.
Models Training score Test score Validat_ion scoreMATLAB-SVM 0.9960 0.9987 0.9941Python-SVM 0.9769 0.9840 0.9808
ing set, test set and validat_ion set. Each dataset has separate values and are not mixed
with each other to verify the accuracies. However, the classes are not able to linearly sep-
arate in 2-D and 3-D. Therefore, RBF kernel funct_ion was applied to non-linearly separate
samples. It can be proposed that by increasing SVM dimension to higher dimensions, the
SVM model might be able to linearly separate data samples.
Figure 20. Simulink-ML load isolat_ion output.
53
The SVM model on MATLAB/Simulink was tested by applying different types of line faults
including 3-ph, 3-ph-g, 2-ph, and 2-ph-g. The ML model isolated the load within 7 mil-
liseconds af_ter the art_ificial fault was generated. The fault was generated at 2 second
and the algorithm isolated the load from generators at 2.007 seconds. Moreover, the ML
model was also tested without applying any fault to test maloperat_ion of the ML model
and its was able to accurately ident_ify that the fault has not occurred. When ph-ph fault
is applied the ML algorithm was able to recognize the fault and isolate the load af_ter 6ms
of occurrence of fault. In the Figure 20 Vabc-L, Iabc-L, V-fault and I-fault represents load
voltage, load current, fault voltage and fault current respect_ively.
54
5 Conclusion
To conclude, in this research work a modern fault diagnosis concept is proposed which
ut_ilizes intelligence of ML and reliable communicat_ion of 5G Network. This study work
also highlights the importance of 5G’s key features including URLLC, mMTC and edge com-
put_ing. The research work is divided into two parts:- 1) Developing 5G communicat_ion
network to compute latency 2) Developing ML algorithm for fault diagnosis in power sys-
tem.
A 5G standalone communicat_ion network is developed on OMNeT++ by simulat_ing com-
municat_ion of two UE over gNB. The latency for the designed network was computed to
be 20ms. However, in future, a MEC-based SA 5G network can also be simulated which
will have lower latencies, as per literature.
In second part of the study two SVM models are trained with the same dataset gener-
ated in MATLAB/Simulink. Python-SVM model is used to visualize the dataset and observe
the decision boundaries. Whereas, MATLAB SVM model is deployed in Simulink to iden-
t_ify occurrence of fault and isolate the load. Four different art_ificial faults (3ph, 3ph-g,
2ph and 2ph-g) were generated and the MATLAB ML model was able to isolate the load
within 7 milliseconds. MATLAB-SVM model achieved an accuracy of 99% and PCA-SVM
ML model, which was used to visualize the data, achieved an accuracy of 97%.
One of the future work would also be a pract_ical implementat_ion of proposed modern
fault diagnosis schemes by deploying ML algorithm at 5G’s edge server and providing
IEDs to communicate over 5G network. Modern fault diagnosis can be a step toward reli-
able and stable future power system. Incorporat_ing modern communicat_ion technology
(5G/B5G) and art_ificial Intelligence will allow to take a step toward a more efficient and
smarter power systems. Since tradit_ional protect_ion schemes are inefficient for future
power systems, advance efficient protect_ion schemes are needed which can accurately
diagnose and isolate faults within the defined period.
55
Bibliography
Alimi, O. A., Ouahada, K., & Abu-Mahfouz, A. M. (2020). A review of machine learning
approaches to power system security and stability. IEEE Access, 8, 113512-113531.
Bose, B. K. (2017). Art_ificial intelligence techniques in smart grid and renewable en-
ergy systems—some example applicat_ions. Proceedings of the IEEE, 105(11), 2262-
2273.
Bouras, C., Gkamas, A., & Diles, G. (2020). A comparat_ive study of 4g and 5g network
simulators.
Cisneros-Saldana, J. I., Samal, S., Singh, H., Begovic, M., & Samantaray, S. R. (2022).
Microgrid protect_ion with penetrat_ion of ders - a comprehensive review. In 2022
ieee texas power and energy conference (tpec) (p. 1-6).
Cosovic, M., Tsitsimelis, A., Vukobratovic, D., Matamoros, J., & Anton-Haro, C. (2017).
5g mobile cellular networks: Enabling distributed state est_imat_ion for smart grids.
IEEE Communicat_ions Magazine, 55(10), 62-69.
Coster, E., Myrzik, J., & Kling, W. (2010). Effect of dg on distribut_ion grid protec-
t_ion. In D. N. Gaonkar (Ed.), Distributed generat_ion (chap. 5). Rijeka: IntechOpen.
ht_tps://doi.org/10.5772/8880
Deshpande, R., & Raviprakasha, M. S. (2015). Power system communicat_ions: Recent
trends, technologies and future - a comprehensive literature scan. In 2015 interna-
t_ional conference on energy systems and applicat_ions (p. 606-611).
Ericsson, G. (2002). Classificat_ion of power systems communicat_ions needs and require-
ments: experiences from case studies at swedish nat_ional grid. IEEE Transact_ions
on Power Delivery, 17(2), 345-347.
Eskandarpour, R., & Khodaei, A. (2018). Leveraging accuracy-uncertainty tradeoff in svm
to achieve highly accurate outage predict_ions. IEEE Transact_ions on Power Systems,
33(1), 1139-1141.
Garau, M., Anedda, M., Desogus, C., Ghiani, E., Murroni, M., & Celli, G. (2017). A 5g
cellular technology for distributed monitoring and control in smart grid. In 2017
ieee internat_ional symposium on broadbandmult_imedia systems and broadcast_ing
56
(bmsb) (p. 1-6).
Gomes, M., Coelho, P., & Moreira, C. (2019). Microgrid protect_ion schemes. In A. C. Zam-
broni de Souza & M. Cast_illa (Eds.), Microgrids design and implementat_ion (pp. 311–
336). Cham: Springer Internat_ional Publishing. ht_tps://doi.org/10.1007/978-3-319-
98687-612
Goswami, T., & Roy, U. B. (2019). Predict_ive model for classificat_ion of power system faults using
machine learning. In Tencon 2019 - 2019 ieee region 10 conference (tencon) (p. 1881-1885).
ht_tps://doi.org/10.1109/TENCON.2019.8929264
Gungor, V. C., Sahin, D., Kocak, T., Ergut, S., Buccella, C., Cecat_i, C., & Hancke, G. P. (2011).
Smart grid technologies: Communicat_ion technologies and standards. IEEE Transact_ions
on Industrial Informat_ics, 7(4), 529-539.
Gururajapathy, S., Mokhlis, H., & Illias, H. (2017). Fault locat_ion and de-
tect_ion techniques in power distribut_ion systems with distributed genera-
t_ion: A review. Renewable and Sustainable Energy Reviews, 74, 949-958.
ht_tps://doi.org/ht_tps://doi.org/10.1016/j.rser.2017.03.021
Hovila, P., Syva¨luoma, P., Kokkoniemi-Tarkkanen, H., Horsmanheimo, S., Borenius, S., Li, Z.,
& Uusitalo, M. (2019). 5g networks enabling new smart grid protect_ion solut_ions. In
25th internat_ional conference on electricity distribut_ion. Belgium: AIM - Associat_ion
des Inge´nieurs de Montefiore. (25th Internat_ional Conference on Electricity Distribu-
t_ion, CIRED 2019 ; Conference date: 03-06-2019 Through 06-06-2019) Retrieved from
ht_tp://www.cired2019.org/
Hussain, N., Nasir, M., Vasquez, J. C., & Guerrero, J. M. (2020). Recent developments and
challenges on ac microgrids fault detect_ion and protect_ion systems–a review. Energies,
13(9). ht_tps://doi.org/10.3390/en13092149
Jamil, M., Sharma, S., & Singh, R. (2015, 07). Fault detect_ion and classificat_ion in electrical
power transmission system using art_ificial neural network. SpringerPlus, 4, 334.
Kankanala, P., Das, S., & Pahwa, A. (2014). Adaboost¡formula formulatype=”inline”¿¡tex
notat_ion=”tex”¿+¡/tex¿¡/formula¿: An ensemble learning approach for est_imat_ing
weather-related outages in distribut_ion systems. IEEE Transact_ions on Power Systems,
29(1), 359-367.
Kumar, P., Bag, B., Londhe, N. D., & Tikariha, A. (2021). Classificat_ion and analysis of power
57
system faults in ieee-14 bus system using machine learning algorithm. In 2021 4th inter-
nat_ional conference on recent developments in control, automat_ion power engineering
(rdcape) (p. 122-126).
Lasset_ter, C., Cot_illa-Sanchez, E., & Kim, J. (2018). A learning scheme for microgrid reconnect_ion.
IEEE Transact_ions on Power Systems, 33(1), 691-700.
Liu, Y., Yang, X., Wen, W., & Xia, M. (2021). Smarter grid in the 5g era: A framework integrat_ing
power internet of things with a cyber physical system. Front_iers in Communicat_ions and
Networks, 2. ht_tps://doi.org/10.3389/frcmn.2021.689590
Liu Y, Y., Peng, M., Shou, G., Chen, Y., & Chen, S. (2020). Toward edge intelligence: Mult_iaccess
edge comput_ing for 5g and internet of things. IEEE Internet of Things Journal, 7(8), 6722-
6747.
Lo´pez, G., Moura, P., Moreno, J. I., & Camacho, J. M. (2014). Mult_i-faceted assessment of a
wireless communicat_ions infrastructure for the green neighborhoods of the smart grid.
Energies, 7(5), 3453–3483. ht_tps://doi.org/10.3390/en7053453
Miraf_tabzadeh, S. M., Foiadelli, F., Longo, M., & Paset_t_i, M. (2019). A survey of machine learning
applicat_ions for power system analyt_ics. In 2019 ieee internat_ional conference on environ-
ment and electrical engineering and 2019 ieee industrial and commercial power systems
europe (eeeic / i cps europe) (p. 1-5).
Mishra, M., & Rout, P. K. (2018). Detect_ion and classificat_ion of micro-grid faults based on
hht and machine learning techniques. IET Generat_ion, Transmission & Distribut_ion, 12(2),
388-397. ht_tps://doi.org/ht_tps://doi.org/10.1049/iet-gtd.2017.0502
Panigrahi, B. K., Rout, P. K., Ray, P. K., & Kiran, A. (2018). Fault detect_ion and classificat_ion using
wavelet transform and neuro fuzzy system. In 2018 internat_ional conference on current
trends towards converging technologies (icctct) (p. 1-5).
Patnaik, B., Mishra, M., Bansal, R. C., & Jena, R. K. (2020). Ac microgrid protec-
t_ion – a review: Current and future prospect_ive. Applied Energy, 271, 115210.
ht_tps://doi.org/ht_tps://doi.org/10.1016/j.apenergy.2020.115210
Peterson, L., & Sunay, O. (2020).
Prasad, A., Belwin Edward, J., & Ravi, K. (2018). A review on fault classificat_ion methodologies
in power transmission systems: Part—i. Journal of Electrical Systems and Informat_ion
Technology, 5(1), 48-60. ht_tps://doi.org/ht_tps://doi.org/10.1016/j.jesit.2017.01.004
58
Sarwar, M., Mehmood, F., Abid, M., Khan, A. Q., Gul, S. T., & Khan, A. S. (2020). High
impedance fault detect_ion and isolat_ion in power distribut_ion networks using support
vector machines. Journal of King Saud University - Engineering Sciences, 32(8), 524-535.
ht_tps://doi.org/ht_tps://doi.org/10.1016/j.jksues.2019.07.001
Sepehrirad, I., Ebrahimi, R., Alibeiki, E., & Ranjbar, S. (2020, 04). Intelligent different_ial pro-
tect_ion scheme for controlled islanding of microgrids based on decision tree technique.
Journal of Control, Automat_ion and Electrical Systems, 31.
Shahinzadeh, H., Mirhedayat_i, A.-s., Shaneh, M., Nafisi, H., Gharehpet_ian, G. B., & Moradi, J.
(2020). Role of joint 5g-iot framework for smart grid interoperability enhancement.
In 2020 15th internat_ional conference on protect_ion and automat_ion of power systems
(ipaps) (p. 12-18).
Shaik, A., Borgaonkar, R., Park, S., & Seifert, J.-P. (2019). New vulnerabilit_ies in 4g and 5g cellular
access network protocols: Exposing device capabilit_ies. In Proceedings of the 12th confer-
ence on security and privacy in wireless and mobile networks (p. 221–231). New York, NY,
USA: Associat_ion for Comput_ing Machinery. ht_tps://doi.org/10.1145/3317549.3319728
Sung, M., & Ko, Y. (2015). Machine-learning-integrated load scheduling for reduced peak power
demand. IEEE Transact_ions on Consumer Electronics, 61(2), 167-174.
Telukunta, V., Pradhan, J., Agrawal, A., Singh, M., & Srivani, S. G. (2017). Protect_ion challenges
under bulk penetrat_ion of renewable energy resources in power systems: A review. CSEE
Journal of Power and Energy Systems, 3(4), 365-379.
Tet_teh, B., & Awodele, K. (2019). Power system protect_ion evolut_ions from tradit_ional to smart
grid protect_ion. In 2019 ieee 7th internat_ional conference on smart energy grid engineer-
ing (sege) (p. 12-16).
Vaish, R., Dwivedi, U., Tewari, S., & Tripathi, S. (2021). Machine learning appli-
cat_ions in power system fault diagnosis: Research advancements and per-
spect_ives. Engineering Applicat_ions of Art_ificial Intelligence, 106, 104504.
ht_tps://doi.org/ht_tps://doi.org/10.1016/j.engappai.2021.104504
Venkataramanan, G., & Marnay, C. (2008). A larger role for microgrids. IEEE Power and Energy
Magazine, 6(3), 78-82.
Wang, W., Xu, Y., & Khanna, M. (2011). A survey on the communica-
t_ion architectures in smart grid. Computer Networks, 55(15), 3604-3629.
59
ht_tps://doi.org/ht_tps://doi.org/10.1016/j.comnet.2011.07.010
Xie, J., Alvarez-Fernandez, I., & Sun, W. (2020). A review of machine learning applicat_ions in
power system resilience. In 2020 ieee power energy society general meet_ing (pesgm)
(p. 1-5).
Yrjola, S., & Jet_te, A. (2018, 11). Spectrum for the industrial internet of things - industry needs,
barriers and recommended new models..
Zaidi, S. M. A., Manalastas, M., Farooq, H., & Imran, A. (2020). Synthet_icnet: A 3gpp compliant
simulator for ai enabled 5g and beyond. IEEE Access, 8, 82938-82950.
Zerihun, T. A., & Helvik, B. E. (2019). Dependability of smart distribut_ion grid protect_ion using
5g. In 2019 3rd internat_ional conference on smart grid and smart cit_ies (icsgsc) (p. 51-59).
Zhao, W., Shang, L., & Sun, J. (2019, 12). Power quality disturbance classificat_ion based on
t_ime-frequency domain mult_i-feature and decision tree. SpringerPlus, 4, 1-6.