Download as pdf or txt
Download as pdf or txt
You are on page 1of 63

A

SEMINAR REPORT

ON

“ARTIFICIAL INTELLIGENCE IN POWER SYSTEM”

SUBMITTED BY

M. SHAHBAZ SIDDIQUE CHAUDHARI

(1930331293061)

UNDER THE GUIDENCE OF

PROF. M.F.A.R SATARKAR

DEPARTMENT OF ELECTRICAL ENGINEERING

DR. BABASAHEB AMBEDKAR TECHNOLOGICAL


UNIVERSITY, LONERE, RAIGAD (MS)- 402103
2022-2023

Dr. Babasaheb Ambedkar Technological University, Lonere Page 1


DR. BABASAHEB AMBEDKAR TECHNOLOGICAL
UNIVERSITY, LONERE, RAIGAD (MS) - 402103

DEPARTMENT OF ELECTRICAL ENGINEERING

CERTIFICATE

This is to certify that M. Shahbaz Siddique Chaudhari


(1930331293061) B.Tech. in Electrical Engineering have satisfactorily
completed the seminar report on “Artificial intelligence in power system”
submits as a partial fulfilment of award of B.Tech. in Electrical Engineering of
Dr.Babasaheb Ambedkar Technological University, Lonere, during academic
year 2022-23. This work has been approved for the final submission.

Date: / /

Place: DBATU Lonere

Prof. M.F.A.R SATARKAR


(Guide)

Dr. Babasaheb Ambedkar Technological University, Lonere Page 2


ACKNOWLEDGEMENT

I express my sincere thanks to the Prof. M.F.A.R. SATARKAR, the head of


Department of Electrical engineering Dr. Babasaheb Ambedkar Technological
University, Lonere, for giving the opportunity to study the vocational seminar
topic. I like to thank Prof. M.F.A.R SATARKAR for guiding me right from the
inception till the successful completion I sincerely acknowledge him for
extending his valuable guidance, support for literature, critical reviews of
seminar report and above all the moral support he had provided to me with all
stages of the seminar.

Finally, I would like to add few heartfelt words for the people who were the part
of the seminar in various ways, especially my friends and classmates who gave
me unending support right from the beginning. My family has been the most
significant in my life so far and this part of my life has no exception. Without
their support, persistence and love I would not be where I am today.

M. Shahbaz Siddique Chaudhari

(1930331293061)

Dr. Babasaheb Ambedkar Technological University, Lonere Page 3


ABSTRACT

Now a days , all are becoming digital based , so by this technology an digital
based power station .It controls all the power grid ,circuit breakers and all the
damages are rectified by artificial intelligence . There are on
• Digital technology based on the application of semiconductor high speed
elements
• Intelligent substation applying IT (information technology ) and
• System configuration aimed at high-speed communication.
In all these are demands no man power is need all we need is intelligence and
working knowledge future intelligence control substation protection monitoring
They are with better efficiency and the functional distribution power distribution
is managed communication system that have advantages in terms of high
performance , information-sharing of power distribution. Today’s conventional
apparatus also requires streaming of functions improvements in sensor
technology and standardize interfacing
Electricity is required in every country and state, so if the electricity is required,
then a consistent and good as well as dependable electricity supply is required.
ML is becoming growing and vast in the every fields of the work, such as
electrical powers system, electronic parts, work software’s and tools, and
construction. Artificial intelligence (AI) is becoming increasingly important in
everyday life, including home and business automation. The amount of research
being done on the power system is growing years by years, and today AI is being
used to make power systems smarter. Power systems are based on geography,
weather, or any specific area. Other factors that the power system is reliant on
include the introduction of new technologies, the upgrading and modification of
technology and equipment, energy transmission, and distribution. AI stands for
artificial intelligence, which is a sophisticated approach for assessing and
solving any problem. As a result, it may be used to address a variety of issues
that arise in power systems, such as control, planning, scheduling, and
prediction. The scope of AI for power systems as well as the use of AI in
handling intricate electrical connections
Dr. Babasaheb Ambedkar Technological University, Lonere Page 4
CONTENTS

1 Introduction
1.1 Definition of power system…………………………………………11
1.2 Definition of AI……………………………………………...............12
1.2.1 AI Techniques ………………………………………………………13
1.2.1.1 Definition of Fuzzy Logic…………………………………..............13
1.2.1.2 Definition of Expert System…………………………………………13
1.2.1.3 Definition of Artificial Nural Networks……………………..............14
1.3 Need for AI in power system……………………………….........14-15
2 Application of Artificial Intelligence (AI)
in Power Transformer Fault Diagnosis
2.1.1 Introduction…………………………………………………………16
2.1.2 Artificial Intelligence Diagnosis System (TFDAI)…………………16
• Construction of TFDES Knowledge Base…………………… 17-19
a) Gas chromatography analysis module
b) Exterior inspection module
c) Oil feature test module
d) Insulation preventive test module
e) Comprehensive analysis
f) Coordinator module
2.1.3 Specific features of TFDES……………………………………20-21
2.1.4 Artificial neural network (TFDANN) ……………………………20-21
2.1.4.1 Construction of TFDANN………………………………………..22
2.1.4.2 TFDAI Principle of operation…………………………………….22
• TFDAI operation
• Specific features of TFDA I
2.1.5 Application Example of AI Diagnosis System (TFDAI)…………22-23

Dr. Babasaheb Ambedkar Technological University, Lonere Page 5


2.2 A Novel of Harmonic Analysis in Power Distribution
Network using Artificial Intelligence
2.2.1 introduction……………………………………………………….24
2.2.2 Description of harmonics in power distribution network…………24-26
2.2.3 Effects and methodology for analysis of harmonics………………26-27
2.2.4 Feed forward back propagation ANN algorithm (FFBP-ANN)… ..27
2.2.5 Results and discussion…………………………………………….27-31
• The Power Quality Data
• Feed Forward Back Propagation Algorithm
2.3 Artificial Intelligence techniques for voltage control
2.3.1 introduction………………………………………………………..32
2.3.2 Algorithmic methods………………………………………….......32
2.3.3 Advantages of Algorithmic Methods……………………………
2..3.4 Applications in NGC………………………………………….......33
• Voltage Collapse Monitor (VCM)
• Combined Active and Reactive Dispatch (CARD)
2.3.5 Artificial Intelligence methods………………………………34-36
2.3.5.1 Expert System
2.3.5.2 Artificial Neural Networks (ANNs)
2.3.5.3 Hybrid System
2.3.5.4 Fuzzy techniques for voltage control
2.3.5.5 Genetic Algorithms
2.4 Artificial Intelligence Techniques for Fault Location
Detection in power system
2.4.1 introduction…………………………………………………..37-40
2.4.2 wavelet transform…………………………………………….40-42
2.4.3 islanding detection method…………………………………42-43
2.5 Artificial Intelligence to Fault Diagnosis in Power Grids
2.5.1 introduction……………………………………………………44
2.5.2 Overall Framework of AI Approaches in Power Grid ……….45
Fault Diagnosis

Dr. Babasaheb Ambedkar Technological University, Lonere Page 6


2.5.3 AI Methods for Fault Identification…………………………45-46
2.5.3.1 Expert System………………………………………………..46-47
2.5.3.2 Artificial Neural Network……………………………………47-49
2.5.3.3 Bayesian Network……………………………………………49-50
2.5.3.4 Optimization Method………………………………………50-51
2.5.3.5 Fuzzy Set Theory…………………………………………….51-52
2.5.3.6 Rough Set Theory……………………………………………52-53
2.5.3.7 Petri Net……………………………………………………54-55
2.5.3.8 Information Fusion Technology……………………………55-56
2.5.3.9 Multi Agent System…………………………………………56-57
2.5.4 Trends and Outlooks of AI Application to Power System
Fault Diagnoses ……………………………………………..57-58

3.0 Advantages of AI in power system……………….…59


3.1 Disadvantages of AI in power system………………59
3.2 Current Application of AI in Power System…………59-60
CONCLUSION………………………………………61
REFERENCES……………………………………62-63

Dr. Babasaheb Ambedkar Technological University, Lonere Page 7


List of Figures

Fig 1 Structure of Power System…………………………………………...11


Fig 2 Construction of Knowledge Base……………………………………17
Fig 3 Module Structure of TFDANN……………………………………....20
Fig 4 The ANN diagram of BP 1 Module…………………………………21
Fig 5 Waveform of Fundamental 3rd and 5th Harmonics…………………..25
Fig 6 Harmonically distorted wave form…………………………………..25
Fig 7 The training cross validation and testing error curves for
harmonic currents in phase A ………………………………………30
Fig 8 The training cross validation and testing error curves for
harmonic currents in phase B ………………………………………30
Fig 9 The training cross validation and testing error curves for
harmonic currents in phase C………………………………………..31
Fig 10 Centralized (a) vs Distributed (b) Generation Power Network….......37
Fig 11 Unintentional islanding representation………………………………38
Fig 12 Wavelet sampling representation……………………………………40
Fig 13 Wavelet transform representation……………………………………41
Fig 14 Algorithm logic representation……………………………………….43
Fig 15 Filtering method representation…………………………………...…43
Fig 16 Overall framework based on artificial intelligence fault
diagnosis system ……………………………………………………..46

Dr. Babasaheb Ambedkar Technological University, Lonere Page 8


List of Table

Table 1 Sample DATA (PPM) ……………………………………..22


Table 2 Oil Feature DATA………………………………………….23
Table 3 Insulation Preventive Test DATA ( Temperature 33C ) …...23
Table 3 Predicted Value of Phase A …………………………………28
Table 4 Predicted Value of Phase B …………………………………29
Table 5 Predicted Value of Phase C …………………………………29

Dr. Babasaheb Ambedkar Technological University, Lonere Page 9


Abbreviation Used in this Report

AI Artificial Intelligence
ANNs Artificial Neural Network
TFDAI Artificial Intelligence Diagnosis System
TFDES Expert System for Transformer Fault Diagnosis
TFDANN Artificial Neural Network for Transformer Fault Diagnosis
FFBP-AN Feed forward back propagation ANN algorithm
VCM Voltage Collapse Monitor
CARD Combined Active and Reactive Dispatch
VCES Voltage Control Expert System
GA Genetic Algorithm
FD Fault Diagnosis
ES Expert System
FST Fuzzy Test Theory
RST Rough Set Theory
PN Petri Net
HHT Hilbert-Huang Transformer
MAS Multi Agent System
DL Deep Learning

Dr. Babasaheb Ambedkar Technological University, Lonere Page 10


Chapter One
1. Introduction

1.1 Definition of Power System[10]


An electric power system is a network of electrical components used to supply,
transmit and use electric power. Power systems engineering is a subdivision of
electrical engineering that deals with the generation, transmission, distribution
and utilisation of electric power and the electrical devices connected to such
systems like generators, motors and transformers The power system is a network
which consists generation, distribution and transmission system. It uses the form
of energy (like coal and diesel) and converts it into electrical energy. The power
system includes the devices connected to the system like the synchronous
generator motor, transformer and circuit breaker, conductor, etc the power plant,
transformer, transmission line, substations, distribution line, and distribution
transformer are the six main components of the power system The power plant
generates the power which is step-up or step-down through the transformer for
transmission.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 11


The Transmission line transfers the power to the various substations Through
substation, the power is transferred to the distribution transformer which step-
down the power to the appropriate value which is suitable for the consumers.

1.2 Definition of Artificial Intelligence[1]


Commonly, artificial intelligence is known to be the intelligence exhibited by
machines and software, for example, robots and computer programs. The term
is generally used for developing systems equipped with the intellectual features
and characteristics of humans, like the ability to think, reason, generalize,
distinguish, learn from past experience or rectify their mistakes. It generally
refers to machines or programs with ability to think on an independent level from
their operator to make decisions.
Artificial intelligence is a scientific discipline aiming to research, develop and
simulate of human behaviour and its rules. Artificial intelligence techniques
including Brain Science, Neurology, information technology and various
discipline artificial intelligence techniques including Brain Science, Neurology,
and disciplines such as information technology, widely used in all walks of life,
through to mimic the behaviour of the human brain, developed a way to replace
human brains discover, identify and analyses machine, improve efficiency and
save money. Traditional artificial intelligence includes problem solving and in
the process of development to improve the AI induction and basic search
technology; meanwhile, in logical reasoning and theorem proving, automatic
programming, natural language understanding.
expert system all get great achievements. But as for improving own learning and
performance aspects have no substantive progress. Limitations of traditional
artificial intelligence can only simulate one logical thinking to solve the problem,
and cannot be solved in addition to logical thinking, dynamic thinking and some
problems that cannot be represented into the sequence of symbols. But in recent
years, another kind of intelligent way that artificial neural networks develop fast
and can be combined with the traditional AI

Dr. Babasaheb Ambedkar Technological University, Lonere Page 12


1.2.1 Artificial Intelligence Techniques[2]
❖ Fuzzy Logic
❖ Expert Systems
❖ Artificial Nural Networks (ANNs)
1.2.1.1 Definition of Fuzzy Logic Technique[2]
Fuzzy logic or Fuzzy systems are logical systems for
standardisation and formalisation of approximate reasoning. It is similar to
human decision making with an ability to produce exact and accurate solutions
from certain or even approximate information and data. The reasoning in fuzzy
logic is similar to human reasoning. Fuzzy logic is the way like which human
brain works, and we can use this technology in machines so that they can perform
somewhat like humans. Fuzzification provides superior expressive power,
higher generality and an improved capability to model complex problems at low
or moderate solution cost. Fuzzy logic allows a particular level of ambiguity
throughout an analysis. Because this ambiguity can specify available
information

1.2.1.2 Definition of Experts System Technique[2]


An expert system obtains the knowledge of a human
expert in a narrow-specified domain into a machine implementable form. Expert
systems are computer programs which have proficiency and competence in a
particular field. This knowledge is generally stored separately from the
program’s procedural part and may be stored in one of the many forms, like
rules, decision trees, models, and frames. They are also called as knowledge-
based systems or rule-based systems. Expert systems use the interface
mechanism and knowledge to solve problems which cannot be or difficult to be
solved by human skill and intellect.
This is an interactive and reliable computer-based
decision-making system which uses both facts and heuristics to solve complex
decision-making problems. It is considered at the highest level of human

Dr. Babasaheb Ambedkar Technological University, Lonere Page 13


intelligence and expertise. The purpose of an expert system is to solve the most
complex issues in a specific domain.

1.2.1.3 Definition of Artificial Neural Network Technique[2]


Artificial Neural Networks are biologically inspired
systems which convert a set of inputs into a set of outputs by a network of
neurons, where each neuron produces one output as a function of inputs. A
fundamental neuron can be considered as a processor which makes a simple non-
linear operation of its inputs producing a single output. The understanding of the
working of neurons and the pattern of their interconnection can be used to
construct computers for solving real world problems of classification of patterns
and pattern recognition
They are classified by their architecture: number of
layers and topology: connectivity pattern, feedforward or recurrent. Input Layer:
The nodes are input units which do not process the data and information but
distribute this data and information to other units. Hidden Layers: The nodes are
hidden units that are not directly evident and visible. They provide the networks
the ability to map or classify the nonlinear problems. Output Layer: The nodes
are output units, which encode possible values to be allocated to the case under
consideration.

1.3 Need for AI in power system[2]


Power system analysis by conventional techniques becomes more difficult
Because of
❖ Complex, versatile and large amount of information which is used in
calculation, diagnosis and learning.
❖ Increase in the computational time period and accuracy due to extensive and
vast system data handling.
The modern power system operates close to the limits due to the ever-
increasing energy consumption and the extension of currently existing

Dr. Babasaheb Ambedkar Technological University, Lonere Page 14


electrical transmission networks and lines This situation requires a less
conservative power system operation and control operation which is
possible only by continuously checking the system states in a much more
detail manner than it was necessary. Sophisticated computer tools are now
the primary tools in solving the difficult problems that arise in the areas of
power system planning, operation, diagnosis and design. Among these
computer tools, Artificial Intelligence has grown predominantly in recent
years and has been applied to various areas of power systems. And increase
in the computational ACCURACY and TIME due to vast system data
handling and extensive versatile large amount of information which is used
in calculation diagnosis complex and learning and maintenance of the
system.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 15


Chapter Two
Application

2.1 Application of Artificial Intelligence (AI) in Power


Transformer Fault Diagnosis[3][4]
2.1.1 Introduction[3]
Power transformer is one of the important electric powers equipment’s in power
system. Transformer fault diagnosis is important for overall power system
reliability. Fault diagnosis is basically a process of associating input data
pattern to one or more fault conditions. It is very difficult to establish a physical
module to link faults with input data. Artificial intelligence system can learn
and adapt to statistical distributions, extract essential characteristics from input
data, and require no physical models.
Successful use of various diagnostic criteria based on gas analysis, however,
remains more an art than science, and largely depends on the user's skill in
interpreting the gas data. It is deemed necessary to seek a more reliable method
to detect transformer faults from dissolved gas analysis. Artificial intelligence
system is an ideal tool for such a task

2.1.2 Artificial Intelligence Diagnosis System (TFDAI)[3]


Because the expert experience is important in transformer fault diagnosis, we
have developed the expert system (TFDES) and the artificial neural network
(TFDANN) for transformer fault diagnosis. The practical effectiveness of the
artificial intelligence for transformer fault diagnosis is admitted in the power
system area.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 16


A. Expert System
➢ Construction of TFDES Knowledge Base
The structure of this TFDES is consists of five parts, namely, the transformer
fault diagnosis Knowledge Base, the Data Base, the Inference Engine, the
Interpretation Mechanism and the Man-Machine Interface. Knowledge base the
kernel part of expert system. The knowledge Base of is TFDES is of module
structure as shown in Fig.

Figure 1 Construction of knowledge base


For establishing the knowledge base TFDES emphasizes the gas
chromatography analysis
a) Gas chromatography analysis module
This module, based on the data of chromatography analysis of dissolved gases
in oil, analyses the transformer operation preliminarily with Three Ratios
Method and Characteristic Gas Method. It distinguishes whether the
transformer is "normal", "normally aged", "partially discharging” (including
partial discharge, spark discharge, arc discharge) or "over heated" (including
degree of overheating). The gases investigated are H₂, CH4, C2H6, C2H4, C12,
CO and CO2. Since the Three Ratios Method is usually more accurate than
Characteristic Gas Method, it is used foremost. Only those parts that cannot be
covered by Three Ratios Method are analysed by Characteristic Gas Method

Dr. Babasaheb Ambedkar Technological University, Lonere Page 17


b) Exterior inspection module
With its analysis function, this module can determine whether there is an exterior
or interior imperfection by supervising the noise, oil level, oil temperature, etc.
of the transformer in operation.
c) Oil feature test module
This module is used to analyse the present feature of the insulation oil. The items
analyse are: acidity, resistivity. water content, surface tension, dielectric loss tg
8 breakdown voltage. Based on the analysis of these indices. the insulation oil
can be divided into three categories, i.e., "good", "alert", "bad", and the
corresponding preliminary conclusions can be drawn. For those items that
cannot be tested (sometimes due to the restriction of on-site test conditions), this
system permits default inference. This implies the untested items are considered
as "good".
d) Insulation preventive test module
This module is used to analyse the results of insulation preventive tests. By
comparing the measured values of DC resistances, insulation resistances, leak
age currents and dielectric losses tg 8 (with temperature factor been taken into
account) of high, medium, low voltage three-phase windings of the transformer
with their corresponding historical data, the present condition of the transformer
insulation can be evaluated
e) Comprehensive analysis module
According to the gas chromatography analysis, exterior inspection, oil feature
test and insulation preventive test, this module analyses the transformer
operating conditions. comprehensively and gives the final judgment.
If an exterior fault of the transformer is observed, this fault should be eliminated
and intensive care should be paid. If the quality of the insulation oil is not good,
oil treatment should be carried out and intensive care should be paid.
If an interior imperfection of the transformer is identified. beside the nature of
the imperfection, this module also gives the suggestion like "continue the
operation of the transformer", "stop the transformer operation, make the
necessary tests and interior inspections". etc. to the operating personnel. These

Dr. Babasaheb Ambedkar Technological University, Lonere Page 18


suggestions are based on the incremental rate of abnormal gas generation, the
operating personnel. These suggestions are based on the incremental rate of
abnormal gas generation, the operating history and the maintenance condition of
the transformer (historical data) as well as the present condition of system power
supply (current data).

f) Coordinator module
This module is the principal one of TFDES. It controls and coordinates the work
of other modules. When TFDES works, this module starts the gas
chromatography analysis module first, then exterior inspection module.
According to the results of analyses accomplished by these two modules, it starts
the other two, i.e., the oil feature test and insulation preventive test modules.
Again, according to the results of analyses accomplished by all these four
modules, it starts the comprehensive module lastly

2.1.3 Specific features of TFDES[3]


➢ The knowledge presentation system adopted in TFDES is the most widely
used system nowadays- Production System or Rule -Based System. The
knowledge base is of the module structure. While all the modules are
independent. Such a structure is beneficial for modifying, developing and
renewing this base. It is also favourable for melting the knowledge and
experiment of domain experts into this base and hence expanding its function
timely. All these make the knowledge base maintenance very convenient.
➢ This system makes the full use of the specific features of TURBO-PROLOG
language, accomplishes the objective-driven the objective-driven backward
inference, adopts some fuzzy logic hence can handle some uncertainties.
➢ The data base of this system consists of two parts: gas chromatography
analysis and insulation preventive test data base, dynamic data base. The gas
analysis data base puts the input gas data into files, while the insulation test
data base takes the insulation preventive test data as historical materials that
can be retrieved and utilized by users conveniently. The final conceal

union drawn is not only based on the current input data, but also considers
the historical information vertically and the relevant data horizontally.
Dr. Babasaheb Ambedkar Technological University, Lonere Page 19
Dynamic data base is just a context tree. It stores the intermediate inference
and the final diagnosis results. In case of need, they can be retrieved and
interpreted via interpretation mechanism.

2.1.4 Artificial Neural Network (TFDANN)[3]


2.1.4.1 Construction of TFDANN
The function of artificial neural network is to imitate the information processing
of human brain. Artificial neural networks can teach oneself and adapt non-linear
mappings between input and output.
The development of the neural network model consists of two phases: learning
and diagnosis. In the learning phase, gas analysis data and other test data are
taken from past data of several transformers. Then, the data sets are presented to
the network. The weight matrices and biases are generated by the back-
propagation learning algorithm. In the diagnosis phase, test-data of various
transformers are used to calculate actual output of the artificial neural network.
These values are compared with the desired output. The TFDANN is of module
structure, while all the modules are independent, as shown in Fig 2

Figure 2 Module structure of TFDANN


It consists of five modules, namely, the Characteristic Gas Method (BP), the
Three Ratios Method Module (BP2), the Insulation Oil Feature Test Module
(BP3), the Exterior Inspection Module (BP), and the Comprehensive Analysis
Module (BP).The Characteristic Gas Method Module is based on the data of

Dr. Babasaheb Ambedkar Technological University, Lonere Page 20


chromatography analysis of dissolved gases in oil. An artificial neural network
has been used for this module as shown in Fig 3

Figure 3 The ANN diagram of BP 1 module


There are six feature elements. (i.e., H2, CH4, C2H2, C2H4, C2H6, CO) in its input
layer. A four output nodes network is adopted, with normal, overheating, corona, and
arcing nodes. (Y1~Y4) BP2there are three input nodes (X1, X2, X3) and nine output
nodes (Y1~Y9) in the Three Ratios Method Module. There are three feature elements
(i.e., C2H2/ C2H4, CH4/H2, C2H4/C2H6,) in the input layer. A nine output is adopted,
with one normal and eight fault status nodes. BP3there are six input nodes (X1~X6) and
three output nodes (Y1~Y3) in the Insulation Oil Feature Test Module. There are
six feature elements (i.e., acidity, receptivity, water content, surface Tension,
dielectric loss tg δ, breakdown voltage) in the input layer. A three-output nodes
network is adopted, with good, alert, and bad status nodes. The exterior
Inspection Module (BP4) can determine whether there is an exterior or interior
imperfection by supervising the noise, oil level, oil temperature, etc. of the
transformer. According to the gas chromatography analysis (the characteristic
gas method and three ratios method), insulation oil feature test, and exterior
inspection, the comprehensive analysis module (BP5) analyses the transformer
operating conditions comprehensively and gives the final judgment.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 21


2.1.4.2 TFDAI principle of operation[4]
• TFDAI Operating
As TFDAI operating, first. The transformer measuring dates are input. The
logical judging model of TFDAI would make a preliminary judgment, based on
the characteristic gas contents, the rate of gas generation, historical data and
Current data, etc. If transformer is normal the user would be notified
immediately. If an interior imperfection of the transformer is identified, TFDAI
system stars ES passageway and ANN passageway at once. Two passageways
can be drawn each results separately. The results would be transmitted to
comprehensive reasoning module, to make comprehensive analysis and gives
the final judgment. For those items that cannot be tested (sometimes duo to the
restriction of on-side test condition), TFDAI system permits default reasoning.

• Specific features of TFDAI


➢ TFDAI adopts module structure and two passageways. Whole program use
TURBO-PROLOG and Vc++ languages. It accomplishes Data-parallel
processing.
➢ TFDAI ensures the accuracy of system. It is easy to use and works very fast.
It can satisfactorily meet the requirements of on –side application.
➢ TFDAI provides a friendly men-machine interface. This system provides
both the screen and the printer outputs

2.1.5 Application Example of AI Diagnosis System


(TFDAI)[4]
TABLE Ⅰ
SAMPLE DATA (PPM)

Dr. Babasaheb Ambedkar Technological University, Lonere Page 22


TABLE Ⅱ
OIL FEATURE DATA

TABLE Ⅲ
INSULATION PREVENTIVE TEST DATA (TEMPERATURE 33℃)

Dr. Babasaheb Ambedkar Technological University, Lonere Page 23


2.2 Harmonic Analysis in Power Distribution Network
using Artificial Intelligence[5]
2.2.1 introduction[5]
in electrical power system the continuous supply of electrical power has become
a challenge for the power utilities. the increased usage of switching devices an
modern electronic circuitry by customers often disturbs the availability of quality
supply of power. this is due to the injection of undesired harmonics in power
distribution network. harmonic distortion in the supply of electrical power from
the utility is due to the increased magnitude of the currents generated by the non-
linear loads. thus, it becomes essential for power utility engineers to analyze the
wave shape of electric current drawn by non-linear customer loads. these loads
include modern electronic equipment like super computers, variable speed
drives, modern electronic ballasts, and other equipment which operates on
continuous switching mechanism. power utilities across the board aim to
maintain the voltage with constant amplitude and frequency without any
distortion. for linear customer loads e.g., heaters, incandescent lamps and any
equipment containing only resistive elements, the current wave drawn is also
linear i.e., sinusoidal. however, when the customer’s load gets nonlinear the
current drawn also gets non-sinusoidal which leads to harmonic distortion. for
non-sinusoidal conditions the harmonically distorted waveforms are made up of
harmonic frequencies with different amplitudes

2.2.2 Description of harmonics in power distribution network[5]


Harmonic distortion is found in both the voltage and the current waveforms in power
distribution networks and can be given as under

Equations (1) and (2) give the root mean square values of voltages and
currents for the non-sinusoidal waveforms where Vh and Ih are the amplitude of
voltage and current respectively at the harmonic component h. Fig. 1 below

Dr. Babasaheb Ambedkar Technological University, Lonere Page 24


shows the fundamental sine waveform and also shows 3rd and 5th harmonics on
the same waveform where as Fig. 2 gives the distorted waveform which is the
result of sinusoids of harmonic frequencies with different amplitudes.

Fig. 1. Waveforms of fundamental, 3rd and 5th Harmonic

Fig. 2. Harmonically distorted wave form

Dr. Babasaheb Ambedkar Technological University, Lonere Page 25


The total harmonic distortion (THD) which is a measure of the harmonic
component present in a distorted waveform can be expressed as in 3

where mh is the root mean square value of harmonic component h of the quantity
m.

2.2.3 Effects and methodology for analysis of harmonics[5]


power distribution networks are ideally designed to tackle sinusoidal voltage and
current wave forms. However with the increased usage of modern power
electronic equipment the situation has become difficult for power utility
engineers to maintain supply of quality power to its customers on sustainable
basis. the existing power distribution network design in most cases is only
capable of absorbing harmonic distortion to a certain limit after which the effects
of harmonic distortion become evident in the distribution system. some of the
common effects associated with the harmonic distortion in power distribution
network in an electrical power system are
• over voltage problems
• circuit breaker tripping
• equipment malfunction and failure
• interference with communication
• cable heating
• data recording and metering problems
• insulation failures
over the years numerous techniques, methods and tools have been employed to
measure the harmonic distortion in power distribution network. in recent times
the extensive use of non-linear loads especially in modern industry has made it
quite difficult to achieve accuracy for the measurement of amount of harmonics
generated by customer’s equipment. in such a scenario, fast methods for
measuring and estimating harmonic signals through artificial intelligence
techniques have produced excellent results. different researchers have worked
on the harmonic detection in electrical power system with the help of neural
networks and have proposed active filters to cancel the harmonic current

Dr. Babasaheb Ambedkar Technological University, Lonere Page 26


generated from the specific nonlinear load. power quality data for voltage
unbalance for a power utility of 66/22kv zone substation in victoria, . however,
the analysis reserved only 10% data for training purposes and harmonic
distortion data was not considered. the current research work is an extension of
the previous work and aims to analyse the harmonic distortion for the same
power utility by using intelligent algorithms. the PQ instruments at the
substation monitor PQ parameters at the 22kv bus. the values of voltages and
currents are measured phase-to-phase. the different PQ parameters have been
recorded for a period of 2232 hours along with power factor of the distribution
network.

2.2.4 Feed forward back propagation ANN algorithm (FFBP-


ANN)[5]
The feed forward back propagation algorithm is one of the most widely used
techniques in ann. due to the nonconvergent behaviour of multilayer perceptron
for our available power quality dataset, the FFBP algorithm is proposed. in this
algorithm supervised techniques are employed. the training errors for the
estimated harmonics are calculated using the “least mean squared (LMSE)
technique”. the algorithm is summarised as follows:
a. randomly initializes the weight matrix
b. train the network depending on the initial weight matrix.
c. calculates the LMSE by comparing the network output and the desired output.
d. updates the weight matrix by back propagating the result obtained to reduce
the error.
e. repeats all steps from b to d, to achieve convergence. (in this research
convergence is taken as 0.01)
2..5 Results and discussion[5]
In this research, Artificial Neural Networks are used to efficiently predict
harmonics on the three phases of the power distribution network. The
experimental results obtained in this research are discussed below

Dr. Babasaheb Ambedkar Technological University, Lonere Page 27


A. The Power Quality Data
The Power Quality data of a power distribution network used in this research has
been taken from an electrical distribution company in Victoria, Australia. It
consists of 19 attributes to estimate the harmonics in power distribution network.
Each attribute contains 2192 datasets. The raw average values of PQ data were
used. The Power Quality data consists of average values of different parameters
of power distribution network for consecutive 2232 hours.
B. Feed Forward Back Propagation Algorithm
Twenty percent of the available data was reserved for testing purpose where as
the remaining 80% data was used to train the neural network. A two layer neural
network is used with tan sigmoid in the hidden layer and log sigmoid in the
output layer. The output layer estimates the harmonic values on the three phases
of the power distribution network. The estimated and actual values of harmonics
for phase A are listed in Table I, whereas for phase B and phase C these values
are listed in Table II and Table III respectively.
Table I
predicted values of phase A

Dr. Babasaheb Ambedkar Technological University, Lonere Page 28


Table II
predicted values of phase B

Table III
predicted values of phase C

On the average 94.5% of accuracy was achieved for predicting the harmonic
values of the distribution network. This can help the power utility to attain some
precautionary measures against high values of harmonics. The training, testing
and cross validation error curves are shown in Fig. 3, 4 and 5 respectively. The
harmonics on Phase A achieved convergence in 88 epochs. The harmonics on
Phase B achieved convergence in 67 epochs while on Phase C, the convergence
was achieved in 27 epochs.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 29


Fig. 3. The training, cross validation and testing error curves for harmonic
currents in Phase A.

Fig. 4. The training, cross validation and testing error curves for harmonic
currents in Phase B

Dr. Babasaheb Ambedkar Technological University, Lonere Page 30


Fig. 5 The training, cross validation and testing error curves for harmonic
currents in Phase C

Dr. Babasaheb Ambedkar Technological University, Lonere Page 31


2.3 Artificial Intelligence techniques for voltage control [6]
2.3.1 introduction[6]
In Electric Power Systems. the advantages of reactive power dispatching or
optimisation include
• improved utilisation of reactive power sources hence reduction in reactive
power flows and real losses of the system
• unloading of the system and equipment as a result of reactive flow reduction:
the power factors of generation are improved and system security is enhanced
• reduced voltage gradients and somewhat higher voltages which result across
the system from improved operation
• deferred capital investment of new reactive power sources as a result of
improved utilisation of existing equipment
• for the National Grid Company plc (NGC), the main advantage is reduced out-
of-merit operation.
The objective of paper is to review conventional methods and AI techniques
for reactive power control.

2.3.2 Algorithmic methods[6]


2.3.3 Advantages of Algorithmic Methods
Reactive power control and dispatch are tradition ally considered as
constrained optimisation problems which can be solved via a number of
algorithmic methods. The main advantages of algorithmic methods include:
• optimality is mathematically rigorous in some algorithms:
• problems can be formulated to take advantage of the existing sparsity
techniques applicable to large scale power system
• there exists a wide range of mature mathematical programming technologies,
such as Mixed Integer 0 scale power systems: Programming. Sparse Primal
Linear Programming, Benders Decomposition and Nonlinear Programming.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 32


2.3.4 Applications in NGC
i) Voltage Collapse Monitor (VCM)
VCM is based on a technique due to Flatabo which uses linearised AC power
Jacobian sensitivities. Though VCM is not an optimisation program, it is security
constrained by the intact and outage system states in the following fashion. A
geographical boundary is defined as input to VCM, which then increases loads
at the import side of the boundary and increases generator output at the export
side to maintain power balance. When a generator reaches its reactive output
limit. the Jacobian sensitivities are re-calculated. Voltage collapse is detected by
the change of sign of the Jacobian sensitivity terms as the loads and generation
are increased. The maximum boundary power transfer is determined by the
power transfer at the instant of voltage collapse for each system state. The
smallest boundary transfer among all system states gives the voltage collapse
limit. Recently, investigations have been carried out to improve on the accuracy
of the VCM method as well as enhancements such as the incorporation of
synchronous compensators and on-load tap changers
ii) Combined Active and Reactive Dispatch (CARD)
CARD is another LP-based optimisation program using the sparse dual revised
simplex formulation with constraint relaxation and described by Chebbo and
Irving . It is a fully coupled active and reactive power dispatch constrained by
the intact and the outage system states. The LP in CARD iterates with a fully
coupled Newton-Raphson power flow until convergence. CARD formulates
multiple looking ahead time steps in one LP to model generator ramping rates.
It has voltage collapse indicators at all busbars and maximises the smallest of
these indicators. The objective function of CARD is a weighted sum of
• generation costs;
• reactive power reserve margins;
• load voltages and
• the smallest busbar voltage collapse indicator

Dr. Babasaheb Ambedkar Technological University, Lonere Page 33


2.3.5 Artificial Intelligence Methods[6]
Despite the success of the algorithmic approaches described in the last Section,
there remains a large class of problems that elude complete solution in a
conventional setting. These problems
• require the use of knowledge bases to store human knowledge:
• require operator judgement particularly under practical situations;
• require experience gained over a period of time;
• are characterized by network uncertainty. load variations etc
Hence the use of Artificial Intelligence techniques (Expert Systems. Neural
Networks, Genetic Algorithms, etc) is being considered for voltage control.
Typical applications will be discussed below

4.3.1 Expert Systems


For an expert system. the main advantages are:
• it is permanent and is consistent;
• it can easily be transferred or reproduced and
• it can easily be documented.
Liu and Tomsovic described a prototype voltage/var control expert system
(VCES), its main purpose being to prove the feasibility of the approach for this
particular domain. VCES contained both knowledge of an empirical nature and
knowledge supported by reasoning over cause-effect modelling. Subsequent to
this research, an expert system was deployed at the control centre of the
Portuguese national utility- EDP . The main problem was to solve the voltage/var
problem within the appropriate context of the SCADA/EMS systems. There was
also the need to deal with the problems associated with large, complex systems
and computational efficiency in real-time environments. The experience gained
with the EDP project gave an insight into the effective role of knowledge-based
systems in the EMS environment, and the evolutionary process of integration of
artificial intelligence technology into the power system industry

Dr. Babasaheb Ambedkar Technological University, Lonere Page 34


4.3.2 Artificial Neural Networks (ANNs)
ANN technology and its characteristics have been reported extensively in the
literature and the main
advantages are:
• is fast. robust and adapts to the data
• possesses learning ability and it is appropriate for non-linear modelling
These advantages suggest their use for voltage security monitoring and control.
In a recent paper on an ANN-based voltage collapse investigation. the multi-
layered feedforward perceptron was trained by. a backward error propagation
algorithm. The approach is based using the Minimum Singular Value method.
Though the neural network training is generally. computationally expensive. the
method described took negligible time to evaluate voltage stability once the
network had been trained
Despite these advantages of ANN method. some of the disadvantages of using
this technology include
• its large dimensionality;
• convergence difficulties; selection of the optimum configuration - (and so the
possibility of using Genetic Algorithms (GA) to assist is being investigated):
• the choice of training methodology - the use of GA in training are being
explored;
• the 'black-box' representation of ANNs - they lack explanation capabilities and
so decisions are not auditable;
• the fact that results are always generated even if the input data are unreasonable

4.3.3 Hybrid Systems


It has been suggested that the complementary methodologies of Expert Systems
(ES) and Neural Networks might be combined to advantage in a hybrid system
for voltage collapse monitoring . In this hybrid scheme, the ES carries out high-
level monitoring. diagnosis. or planning, whereas the ANN is used to evaluate
local problems which would otherwise involve great analytical complexity

Dr. Babasaheb Ambedkar Technological University, Lonere Page 35


4.3.4 Fuzzy Techniques for Voltage Control
Fuzzy set theory has also been employed to coordinate control of voltage and
reactive power to enhance voltage security of a power system

4.3.5 Genetic Algorithms


Genetic algorithms (GA) are general purpose optimisation algorithms as
described by Holland . They are distinguished from conventional optimisation
by the use of concepts from population genetics to guide the optimisation search.
A GA operates on a population of individuals each representing a solution to the
problem at hand. By a random process of cross-over and mutation, the GA
combines features of the fittest individuals in a population to breed offspring;
the aim is to preserve the desirable characteristics of individuals from one
generation to the next generation offspring. The advantages of GAS over
traditional techniques are
• it needs only rough information of the objective function and places no
restriction such as differentiability and convexity on the objective function;
• the method works with a set of solutions from one generation to the next, and
not on a single solution. thus, making it less likely to converge on local
minima
• the technique uses a population of solutions each represented by a bit string
in which the flipping of a bit conceptually represents searching through a
hyperplane in the solution space; the search effort is thus spread
simultaneously in many repones of the search space.
• he solutions developed are randomly based on the probability rate of the
genetic operators such as mutation and cross-over; the initial solutions thus
would not dictate the search direction of a GA The use of GA has begun to
be applied to power system optimisation problems, in particular, to reactive
power and voltage optimisation

Dr. Babasaheb Ambedkar Technological University, Lonere Page 36


2.4 Artificial Intelligence Techniques for Fault Location
Detection in Power Systems[7]
2.4.1 introduction[7]
Traditional power systems networks are usually composed of one or more power
generating resources that are located upstream and included in the transmission
network. This type of networks is known as passive distribution power systems.
In order to benefit from small renewable generating resources, the concept of
distributed generation was imposed. For best results, these “green” energy
sources are introduced in the distribution level in power networks. At this level
these power sources have a maximum generating power of a few megawatts.
Usually, alternative power sources are composed of wind generators, solar
power generators, hydro or geothermal energy sources.
The difference between distributed generating power systems and conventional
ones is the fact that there are supplementary generating power sources placed
near the load, forming small microgrids. A schematic comparison between the
two is presented in Fig. 1.

Fig. 1 Centralized (a) vs Distributed (b) Generation Power Network

There are a lot of benefits from introducing distributed generating sources into
the power network. The first one being that these sources are environmentally
friendly. Due to the fact that they are placed near the load means increased

Dr. Babasaheb Ambedkar Technological University, Lonere Page 37


efficiency. Also, by forming these microgrids the need for capacity upgrades for
transmission and distribution is reduces. However, one should consider the
drawbacks of integrating these green power sources into the power network.
Among them the most notable to be tackled are voltage fluctuations, frequency
stabilization and power quality issues. Another aspect to be considered as a
drawback is the oscillating nature of the renewable sources that are highly
dependable on the weather conditions. The purpose of this paper is to provide a
means of islanding detection. Islanding refers to the situation in which
microgrids form as a result of disconnecting the main power source without
stopping the distributed renewable sources. When referring to an islanding
situation one can categorize it as intentional or unintentional. The intentional
disconnection from the main power source would be a means of protection for
the microgrid
and is typically applied when there is a fault on the network that injects
disturbances. Using intentional islanding methods depends on the power
network system configuration. In some cases, this type of protection is not
allowed, and the system stopes the distributed generation when sensing the main
power source disconnection. This configuration relies on the personnel safety
reasons. Most distributed power generation networks are the result of modifying
the traditional power system with minimal device upgrading; thus, the need of
disconnecting the distributed sources. An undetected “island” microgrid is the
“unintentional islanding”. Fig. 2 presents unintentional islanding case when a
fault occurs in the power network.

Fig. 2 Unintentional islanding representation

Dr. Babasaheb Ambedkar Technological University, Lonere Page 38


In order to deal with these situations some standards were created. IEEE 1547-
2003 and IEEE 929-1988 state that an unintentional island must be detected in
maximum 2 seconds and the disconnection of the distributed generation in
islanded microgrids, respectively This explains the need for rapid islanding
detection methods. In the last years literature these methods were divided into
central and local techniques. Some central or remote techniques widely spread
in research literature are intercropping, switch state monitoring and system state
monitoring. All of these methods refer to a monitoring process that uses real time
communication between the network equipment. Local techniques are based on
measurements on the network parameters at the distributed generation level such
as voltage and frequency. These methods were further classified into passive,
active and hybrid techniques based on the action taken when islanding is
detected
One of the most common used hybrid techniques is based on voltage and real
power shift. The passive component is the average voltage signal and the active
component is the real power shift. This method is suitable for configurations
with multiple distributed sources operating in the same power network. Real
power shift needs to be applied in cases where the passive component fails in
islanding detection.
Other hybrid techniques are based on the voltage unbalance and frequency set
point. The advantages of using three methods simultaneously like: voltage
unbalance, positive feedback and total harmonic distortions, is that where one
fails the other may succeed. For each distributed generation the voltage
unbalance method is used because is more sensitive to disturbances and using
the total harmonic distortion the method discriminates between islanding
situation and load switching
For detecting voltage sag/swell one of the most used methods is the dq-
transform. This method is suited for a three-phase power network. First the
phases are transformed into d and q components, then the square root of the sum
of squares of the
components is extracted. The result is compared with a set value and the obtained
result is passed through a low pass filter, thus resulting in the positive sequence
of the voltage. A negative voltage component generated by voltage sag it will
appear as an error in this method. The low pass filter needed for obtaining the
Dr. Babasaheb Ambedkar Technological University, Lonere Page 39
positive value of the voltage is a narrow band filter that can cause delay and
lengthen the response time of the algorithm. In order to detect the voltage
changes the output of the filter is passed through a hysteresis comparator that
generates the detection signal. A high signal indicates a voltage sag/swell
situation; otherwise, the output signal will be a low. The drawbacks are that the
method cannot detect voltage sag lower than a defined depth
The method described in this paper is classified under the signal processing
methods used for detecting the islanding mode. These techniques are regarded
as intelligent artificial methods used to determine the network condition. The
focus of the paper is concentrated on wavelet theory that uses a mathematical
model for small waves called wavelets.

2.4.2 Wavelet transform [7]


As mentioned above a wavelet is a small wave extracted from the signal, this
implies that the studied function must be an oscillating signal. First a “mother
wavelet” is generated from which the wavelets are generated. Usually, a
translation term is used to keep track of the window length as the algorithm
progresses

Fig. 3 Wavelet sampling representation

Most proposed algorithms work with discrete wavelet transform instead of using
the time-domain version of the signal. In the below equation the mathematical
expression of the discrete wavelet transform is presented

Dr. Babasaheb Ambedkar Technological University, Lonere Page 40


Where n and m are used to refer to the translation and dilation of the wavelet, a0
is the dilation step (needs to be grater than one) and b0 represents the translation
step (needs to be grater than zero)

Equation 2 represents the discrete form of (1) while (3) refers to the time-
frequency domain representation of scale location signal indicated by the m, n
indexes. For simplicity and practical implementation, the term “power of two”
is used. Many researchers studying wavelet method applied the dyadic grid
configuration for the dilation and translation steps . Doing so it will result in the
creation of an orthonormal wavelet basis presented in (4).

Hence using the orthonormal basis, one can reproduce the signal using the
inverse form of the discrete wavelet transform and the wavelet coefficients, as
presented in the above mathematical expression. When decomposing the signal
into wavelets it is necessary to include a low pass filter and a high pass filter.
Doing so each signal wave is divided into lower resolution items. Thus, resulting
in the approximation and detail components.

Fig 4. Wavelet transform representation

Dr. Babasaheb Ambedkar Technological University, Lonere Page 41


Approximations are obtained after applying the low pass filter and the details are
the high frequency items. After one iteration the next signal to be considered is
the low frequency component. These steps can then be repeated and each time
an iteration is processed the time resolution is halved because only half of the
sample numbers represent the signal and the frequency resolution is doubled due
to the fact that frequency band spreads only on half the previous one. For a better
understanding of the method the above schematic representation is proposed.

Finding the approximations and details values the above mathematical


expression may be used; where l(k) and h(k) represent the low pass and high
pass filtering values

2.4.3 Islanding detection method[7]


The wavelet transform method presented above represents a time-scale
representation of a waveform using basis functions to scale and translate a
sample into a wavelet. This way both the time and frequency values are
preserved without reducing the resolution. That is why this method is often used
instead of short Fourier transform which uses a fixed window size.
Many studies were carried out in order to determine the best wavelet order to be
considered in order to have accurate results and low number of coefficients for
fast processing in this paper the number wavelet 5 is selected, thus the level of
computation is set to 5. First a brief schematic of the algorithm is presented in
the below representation:

Dr. Babasaheb Ambedkar Technological University, Lonere Page 42


Fig. 5 Algorithm logic representation
In order to continue the presented algorithm, the next figure describes the
filtering process.

Fig. 6 Filtering method representation


As already described the flow of the process follows the voltage signal that
passes through the low pass filter and high pass filter. The results are sub-
sampled, and the resolution is reduced by half regarding the number of samples.
This will make the frequency resolution to double due to the fact that the
frequency band now covers half of the previous frequency band. The number of
repetitions of this algorithm will determine the level of the wavelet transform
method. This method uses Daubechies wavelets of order 4 in order to detect
voltage interruption, voltage sag/swell and transient disturbances in the
distributed generation power system.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 43


2.5 Artificial Intelligence to Fault Diagnosis in Power
Grids[8]
2.5.1 introduction[8]
Power system fault diagnosis (FD) is to quickly and accurately deal with fault
occurring sections, fault components and fault properties according to
theoretical analysis and operating personnel experiences based on relevant
information generated by various types of protection devices . Due to the large
scale and increasing complexity of modern power grids, traditional FD methods
may lack the ability to deal with the big data collected by SCADA systems in
the case of faults. The intelligent behaviour of humans to analyze and deal with
problems has made up for the deficiencies of traditional methods. In comparison,
AI-based FD approaches may have better accuracy, higher speed, and greater
expandability. Therefore, AI technology has been the focus of future research in
the field of power network FD.
In the past two decades or so, there have been many researches carried out on
the use of AI technology in power grid FD in China and abroad. some good
research results have been achieved but few have found practical applications
due to complex mode and characteristics of power system faults under different
conditions. However, with rapid development and significant breakthrough in
AI technologies over recent years, typified by Google’s AlphaGo, there have
been increased interest in applying AI to power systems, especially FD. They
can be categorized into the following major methods: expert system, artificial
neural network, optimization technology, Petri network, fuzzy set theory, rough
set theory, information fusion technology and multi-agent technology.
This paper reviews the current widely used artificial intelligence methods for
power grid fault diagnosis, including expert systems, artificial neural networks,
Petri nets, etc. It gives a brief introduction, and analyzes the characteristics and
deficiencies of these methods. Key issues and future development trends that
need to be addressed in the future development of the field are discussed.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 44


2.5.2 Overall Framework of AI Approaches in Power Grid Fault
Diagnosis[8]
The overall framework of the grid fault diagnosis application based on artificial
intelligence is shown in Fig. 1. It includes 4 layers: high performance computing
architecture, data acquisition, business scenario and algorithm application. The
High-Performance Computing Architecture (HPC) is mainly composed of
computing devices, storage devices, and network devices. The computing cluster
is a computer cluster formed by a hybrid architecture such as CPU, GPU and
TPU. It can provide powerful computing power for various algorithms to solve
the problem of excessive training time under massive data and multi-level
network parameters. The storage device provides storage support for the massive
operational data of the grid.
The data acquisition layer mainly provides real-time remote signal data,
protection action data, recorded transient data of each station, and grid cross-
data support for the upper layer.
The algorithm application layer provides a unified algorithm engine support for
upper-layer applications by encapsulating various algorithms, including expert
systems, artificial neural networks, and Bayesian networks. The business
scenario is mainly to analyze and judge the fault information of the switch
quantity and the electric quantity, use various artificial intelligence methods to
diagnose the fault, locate the fault, and determine the fault component and fault
type.

2.5.3 AI Methods for Fault Identification[8]


This chapter reviews major AI methods applied to power network FD, including
their advantages and disadvantages.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 45


Fig. 1: Overall framework based on artificial intelligence fault diagnosis
system

2.5.3.1 Expert System[8]


The expert system is the earliest AI technology applied to fault diagnosis in the
power grid. The expert system uses rules to represent the expert's subjective
experience, protection and circuit breaker action logic to form a knowledge base.
When a fault occurs, the fault information is compared with the rules in the
knowledge base to infer the faulty component
The ES has strong reasoning ability for deterministic information and can make
a scientific explanation for the fault development process. However, there are
some shortcomings: when faced with large-scale or complex power grids,
knowledge acquisition is difficult, knowledge base construction is difficult, and
diagnosis speed is slow; when the grid topology changes, the knowledge base is
difficult to update and maintain; and there is poor learning ability and fault
tolerance. In , the fuzzy set theory is combined with the ES. By fuzzy inference
of voltage and current values and protection and circuit breaker information, the
problem of poor fault tolerance of ES can be effectively improved. Literature
uses pattern matching to analyze the relationship between grid structure changes,
protection action behaviours and different faults during grid faults, which
enhances the completeness and interpretability of FD to a certain extent. The
logic process is simple and easy.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 46


combined with artificial neural network to pre-process uncertain information,
simplify knowledge rules, and reduce the interference of abnormal fault
information, in order to effectively dealing with certain uncertainties such as
information, protection and circuit breakers. The literature proposed a new FD
method combining the expert system and the information theory. The traditional
ES based on rule matching judgment is transformed into a numerical calculation
method to solve the ambiguity of information and accuracy of reasoning. The
literature proposed an association rule mining method based on feature mining,
which combined with the ES, greatly improved the speed and accuracy of
reasoning; the literature introduced Knowledge grid technology, using
knowledge map to represent the traditional expert rule base, and updating the
knowledge node online by modifying the knowledge node of the knowledge
map, greatly improves the online maintenance and search efficiency of the ES.
considering the limitations of the traditional expert system, a model-based
diagnosis method was proposed,
which obtained the candidate diagnosis offline, confirmed the candidate
diagnosis online, and reduced the diagnosis time, the finite state automaton is
used to model the ES fault reasoning. Combined with the characteristics of finite
state automata and production reasoning, a simple reasoning model is
constructed to improve the reasoning ability for real-time diagnosis of complex
faults. In recent years, many experts and scholars have combined ES with other
artificial intelligence technologies to make up for the shortcomings of the ES in
practical applications and achieved good results. However, in practical
applications, maintenance of the ES knowledge base under many different
operating conditions is extremely cumbersome and difficult to meet the needs of
large-scale power system online FD. Therefore, in the future, it is necessary to
further seek integration with other AI technologies to study and construct a grid
FD system based on ES with self-maintenance and self-learning ability

2.5.3.2 Artificial Neural Network[8]


Artificial neural network (ANN) is a complex network system that simulates the
biological nervous system for information parallel processing, through input and
output relationships, and is widely interconnected by a large number of simple

Dr. Babasaheb Ambedkar Technological University, Lonere Page 47


neurons. The ANN takes the alarm information as the input, and the diagnosis
result as the output, analyzes and infers a large number of training samples, and
obtains the general law, so that the unknown or unpredictable fault information
can be analyzed and judged
The ANN method has the advantages of strong self-learning ability, strong fault
tolerance, strong generalization ability, good robustness, fast reasoning and
parallel distributed processing However, it has the following disadvantages:
training requires a large number of representative samples, and it is relatively
difficult to obtain a complete sample, especially for complex faults; the
diagnosis process is opaque and lacks the ability to interpret the results.
Literature uses two types of neural networks (NN) modelling to estimate the
fault area. Literature introduced the concept of quantum mechanics when
applying NN for FD. For large-scale power systems, the literature proposed a
combination of radial basis function neural network (RBFNN) and fuzzy
integration. Firstly, the large grid is divided into several subnets according to the
overlapping network partitioning method, which reduces the difficulty of model
construction, and uses fuzzy integral association to merge adjacent subnets. The
diagnosis output is obtained, and the FD of the internal and adjacent sub-network
tie lines is realized. The literature uses a single component as the cell object to
establish a cellular universal neural network (NN), and automatically generates
a model using a 3D matrix, using a fuzzy sagittal shape. The figure describes the
logical reasoning relationship between components and protection and circuit
breakers, which can effectively adapt to changes in network topology and has
good fault tolerance and portability. In order to avoid falling into local minimum
values, the literature combines genetic algorithm with back propagation neural
network (BPNN) to effectively diagnose the fault location of the power grid and
improve fault tolerance. a multi-output attenuation radial basis
function (MDRBF) ANN diagnostic model is constructed. The diagnostic
structural parameters can be approximated without any training, thus improving
the construction speed of the diagnostic model. In the improvement of network
structure and algorithm efficiency, the literature improved the extreme learning
machine for grid FD. The training of the diagnostic model does not need to adjust
the input weight, so that the network model has better generalization ability. In
recent years, ANN have achieved good results in the application to the power
Dr. Babasaheb Ambedkar Technological University, Lonere Page 48
grid, but the interpretation of diagnostic results and the application of large-scale
power grids are still insufficient. Future research should focus on selecting
valuable training samples to enhance the ability to explain FD and make it
suitable for large-scale grid fault diagnosis.

2.5.3.3 Bayesian Network[8]


Bayesian network (BN) is a model for expressing and inferring uncertainty
knowledge based on graph theory and probability theory. It uses network flow
graph to express knowledge and probability theory to deal with the uncertainty
generated by different knowledge components, through conditional probability
reasoning to achieve grid fault diagnosis
The BN diagnostic model is clear and intuitive, easy to find causal relationships
between network node data, and can make good diagnostic decisions for
uncertain and incomplete information based on the inherent logic relationship
between component failure, protection action and circuit breaker trip, a FD BN
model is established and a network model is automatically generated. Proposed
a prior probability calculation method for event sampling, which can accurately
express the statistical relationship between component failure and protection
refusal and maloperation, and is suitable for dealing with complex faults.
with the state information of the monitoring equipment, using BN to achieve the
advance prediction of the fault situation. based on component modelling, by
setting the prior probability of each node, the fault information passes through
the BN, and the probability of failure of each component is obtained, thereby
identifying the faulty component. divides the grid fault into three categories,
constructs a BN model with variable structure considering the credibility,
effectively simplifies the Bayesian model and introduces credibility degree
assessment to improve the accuracy of the diagnosis. Under the condition of lack
of information or uncertainty, combines the fault recording system to locate
faulty components quickly and accurately, and improve the efficiency of grid
fault handling. BN and dynamic causal reasoning chains to establish a
corresponding model for each chain of faulty elements, reducing the impact of
uncertain information on diagnostic results.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 49


In order to ensure the accuracy of the diagnosis, the BN must have sufficient
prior knowledge, but in most cases, the statistical samples of large grid faults are
lacking, which restricts the application of the method and it is difficult to model
the complex network when the grid structure is changing or evolving fast under
different operating conditions. It is therefore necessary for future research to
focus on how to implement automatic modelling of BN and how to combine
them with information fusion theory

2.5.3.4 Optimization Methods[8]


The diagnosis method based on optimization techniques is a mathematical
model-based solution method. The basic idea is to model the FD as integer
programming problem considering topology of the power grid, the configuration
of the protection, and the protection logic of the switch. Algorithms are used to
find the optimal solution
according to the protection action principle, the expected state function of the
protection and circuit breaker during the fault process is obtained, and an
analytical model for characterizing the electrical equipment state of the system
is established. comprehensively considers the secondary system configuration
and diagnostic information characteristics, the particularity of the breaker
failure, redefines the expected function of each device, and improves the grid
fault diagnosis optimization model; based on this comprehensive model,
consider different contribution factors for different protection and circuit
breakers, also considered the reclosing factor,
added the biogeography optimization method to achieve the efficient solution.
introduced the genetic algorithm-table search (GATS) method to improve the
sensitivity and dependence of fault information proposed a new method based
on unconstrained binary programming model and optimized immune algorithm
to improve efficiency and precision of power system FD proposed a new power
system fault interval estimation method, using the Hebb learning rule to obtain
the objective function required for the fault region and then using the continuous
genetic algorithm (CGA) optimization method to estimate the fault area. In order
to adapt to the uncertainty of grid faults in time and space dimensions, proposed

Dr. Babasaheb Ambedkar Technological University, Lonere Page 50


a grid FD method based on variable length chromosome genetic algorithm. In
order to reduce the influence of human subjective factors on the diagnosis results
a multi-objective optimization method to transform the grid FD problem into a
multi-objective optimization problem and solved the model by using the Pareto
optimal solution method.
The mathematical and theoretical basis of the FD model based on the
optimization technique are rigorous, thus avoiding complicated logical
reasoning to a certain extent, being easy to program and realize, and being able
to give a plurality of possible diagnosis results that are globally optimal or
locally optimal under the condition that the information on which the diagnosis
is based is incomplete However, this method has a number of drawbacks: the
objective function and its parameter determination are difficult; the algorithm
may lose the optimal solution due to random factors in the optimization process;
multiple iterations lead to slower speed. Therefore, how to establish a more
reasonable mathematical analytical model to improve the fault tolerance and
accuracy of diagnosis, and how to design a more efficient solution to improve
the efficiency of the model remains to be further studied.

2.5.3.5 Fuzzy Set Theory[8]


The fuzzy set theory (FST) breaks through the classic set with 0 and 1 to
represent the clear concept of one or the other, and uses the concept of fuzzy
membership to describe inaccurate and uncertain objects , and uses approximate
reasoning rules to effectively express expert knowledge assumes that the
information is accurate only between the faulty equipment and the protection of
the corresponding action, and the uncertainty relationship between the protection
and the circuit breaker is described by fuzzy membership. the fuzzy correlation
three-dimensional matrix is used to represent FD knowledge and adaptive FD.
The influence of circuit breaker trip on topology changes and protection action
logic is considered. For the uncertain information, the literature combined with
the rough set theory,
the protection and circuit breaker signals as the condition attribute set of fault
classification, establish the decision table, and then realize automatically
simplified and reduced search of the decision table, and extract the diagnostic

Dr. Babasaheb Ambedkar Technological University, Lonere Page 51


rules from it. To achieve fault diagnosis, used fuzzy sets and fuzzy reasoning to
describe the uncertainty of power system information and the inaccuracy of FD
model and combined with the characteristics of distributed distribution of power
system information to successfully solve the distributed problem.
It is applied to solve the FD of power system. uses hybrid causal networks and
fuzzy logic to quickly diagnose faulty parts of power systems estimates the
faulty part of distribution substations by means of fuzzy matrix inference and
matrix transformation a fuzzy directed graph model for online fault diagnosis,
which establishes a fuzzy directed graph based on the causal logical relationship
between protection and circuit breaker action after fault occurrence, and through
fuzzy reasoning. The algorithm obtains the action probability of each path of the
directed graph and locates the fault. The FD method based on FST has strong
fault tolerance and can deal with the uncertainty of information, but it suffers
from a number of shortcomings: there is no clear standard for the setting of
membership function; poor maintainability. The fuzzy knowledge base and
membership degree change correspondingly with the change of power grid
structure; large-scale power grid modelling is difficult. In recent years, experts
and scholars often combine FST with other artificial intelligence methods (such
as ES, NN, etc.) to analyze the impact of uncertainty information on diagnostic
systems, improve diagnostic accuracy, and enhance system fault tolerance.

2.5.3.6 Rough Set Theory[8]


Rough Set Theory (RST) is a mathematical tool for studying incomplete data
and dealing with uncertainty knowledge In grid fault diagnosis, rough set theory
uses the redundancy of alarm information for data. The relationship is mined,
the hidden knowledge is found from the fault sample set, and the corresponding
diagnostic rules are derived
The diagnosis method based on rough set theory has strong fault tolerance, can
reduce knowledge, and can make diagnosis reasoning when the alarm
information is incomplete and uncertain. the power system fault diagnosis
problem as a problem classified according to the alarm mode. Through the
attribute reduction in the rough set theory, the power grid fault diagnosis rule is
extracted and the inherent redundancy of the alarm information is revealed.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 52


the attribute reduction algorithm based on rough set theory and binary logic
operation and the improved value reduction algorithm are used to form a
comprehensive fault diagnosis knowledge model using the reduction of decision
tables. proposes to obtain enough information from the fault information system,
and then use rough set theory for fault diagnosis. Considering the limitations of
AI technology in fault diagnosis of large power grids, the uses distributed ideas
to segment large power grids, reduces the size of decision tables, and avoids the
problem of "combined explosions". A joint rule mining algorithm based on
rough set theory is proposed to reduce the complexity of rule extraction.
the discretization of continuous attributes, attribute reduction and value
reduction are transformed into the simplification problem of identifiable matrix,
which improves the execution efficiency of the algorithm, and the continuous
signal such as voltage and current with the signal of protection and circuit
breakers. Participating in the fault diagnosis based on rough set reduces the
influence of missing key information and improves the accuracy of fault
identification.
the knowledge reduction of rough set theory and the ability to process uncertain
information, hierarchically mines fault alarm information, and performs attribute
optimization and diagnostic rule extraction on decision tables in multiple grid
regions. Although the fault diagnosis method based on rough set theory has
achieved some good results, it suffers from a number of drawbacks: for the
missing or erroneous key information, the diagnosis result will be affected; when
multiple failures occur, the decision table may suffer from dimensional problem
of “combined explosion”. Therefore, the use of rough set theory for knowledge
reduction to reduce the impact of key information, and integration with other
artificial intelligence to improve the shortcomings of rough sets in current
applications is the future development trend.

2.5.3.7 Petri Net[8]


Petri Net (PN) is a general mathematical model. The whole model can be
represented graphically, and its static structure and dynamic behavior are
analyzed by matrix operation. It is suitable for discrete event dynamic system

Dr. Babasaheb Ambedkar Technological University, Lonere Page 53


modelling and analysis. A fault is a dynamic system with discrete events, so a
Petri net can be used to construct a FD model PN's power grid FD method has
the advantages of simple and fast reasoning process and graphical representation
of diagnostic model expression optimized the FPN structure and reasoning
process, and improved the
efficiency of FD. hierarchically models the different directions of components,
which can improve the adaptability of the model to network topology changes
and reduce the dimension of the inference matrix. established a complete
knowledge representation method, and provided sufficient FD knowledge to the
PN model for logical reasoning, so that its accuracy and diagnostic efficiency
meet the actual needs.
In a dual-network dual protection configuration of 750kV substation is proposed,
and a FD method for substation fuzzy Petri net based on redundancy protection
is proposed. In, a power system FD method based on improved dynamic
adaptive FPN and anti-BP algorithm is proposed. A complementary arc element
group is introduced in the dynamic adaptive FPN, and the BP algorithm in the
NN is used in the model. The parameters are trained to improve the fault
tolerance and accuracy of the FPN model. In view of the uncertainty of
information, the fully considers factors such as protection initiation, protection
of exports and continuous electrical quantity, and proposes a multi-factor
hierarchical PN model.
In an FPN FD model with timing constraints is proposed, the forward-reverse
timing reasoning is introduced into the PN FD process, and the abnormal alarm
information is corrected and its initial confidence is reset, improving the
accuracy of troubleshooting results. Many experts and scholars have
concentrated on the improvement of the PN model in complex power grids and
achieved good results. However, it still has a number of problems: the expansion
of the system network topology is easy to cause state explosion when modelling;
The accuracy of the diagnostic results is low; the fault tolerance is low when the
fault information is uncertain. In future research, consideration should be given
to how to implement automatic PN modeling based on the characteristics of FD,
and combine it with other AI to improve its own deficiencies.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 54


2.5.3.8 Information Fusion Technology[8]
Information fusion technology is actually a comprehensive technology of multi-
source information, which can intelligently analyze and synthesize various data
source information, and obtain a unified result that conforms to the measured
object and its properties, thereby obtaining, compared with the source, the
decision-making and evaluation with higher perfectness and accuracy.
Currently, the use of information fusion technology for power grid FD is
developing rapidly and its advantages are increasingly prominent. In, an
improved optimization model is established for the remote signal based on the
cover set theory, using wavelet technology. The failure degree is extracted from
the fault recording information, and finally the evidence theory data fusion is
performed. uses fuzzy integral theory to fuse the primary diagnosis results of
NN to obtain the diagnosis results, and achieves the improvement of diagnostic
speed in addition to reducing the difficulty of network sample acquisition. Based
on the multi-agent system framework, the establishes an information fusion
model for large-scale power grids. Through the fusion of different FD
techniques,
the various fault information of each region can be analyzed, and the FD task
can be completed in a short time. uses the data information of SCADA and
RPMS, uses the step-by-step FD method to analyze the data, and uses the fault
information group technology to realize the data sharing of the subsystem, which
improves the pertinence and efficiency of analysis and data acquisition. In , the
data of network topology, SCADA and RPAM are used to obtain the vertical
and horizontal original signal sequences, instantaneous amplitude sequences and
energy sequences for the current signals of each line by Hilbert-Huang transform
(HHT)
Using the grey relational theory, the corresponding six grey correlation degrees
are calculated, which are weighted and fused to obtain the comprehensive grey
correlation degree of each suspected line, and the fault criterion is given. The
power grid FD method using multi-source information fusion technology
comprehensively diagnoses the switch quantity and electric quantity of different
data sources, which can greatly improve the real-time, fault tolerance and
accuracy of the diagnosis system, and effectively avoid the uncertainty due to

Dr. Babasaheb Ambedkar Technological University, Lonere Page 55


fault information. However, the diagnostic results are not very explainable, and
the multi-source information is not fully utilized. Therefore, in future research,
information fusion technology should focus on how to select appropriate
information fusion methods, fully integrate multi-source data information, and
enhance the interpretation of diagnostic results, which will raise the level of grid
FD to a new level.

2.5.3.9 Multi-Agent System[8]


Multi-agent System (MAS) is a branch of AI that can transform a target problem
into multiple agents that are logically or physically separated. Each problem can
be solved for each agent, and each agent can coordinate information. The final
result is obtained. The cooperation, coordination between agents and the
autonomy, initiative and responsiveness of agents in MAS provide a modelling
method for grid FD. In a real-time FD method and system structure of power
system based on multi-agent is proposed. According to the characteristics of
real-time power system FD and the Agent, the grid FD system is divided into
seven different core agents. In , the diagnostic system is divided into six modules
by MAS, which combines the FD of switch quantity and electric quantity to
improve the accuracy of diagnosis. combined with RPAM data information to
achieve accurate identification of fault areas, and supplemented by SCADA
system to achieve fault operation status monitoring and accurate fault location.
As a distributed diagnostic system, the MAS diagnostic system is highly
scalable, robust and instructive, adapting to the geographical distribution of the
power grid and the complexity of the diagnostic tasks. However, future research
may focus on establishing the hardware and software problems of the system
and dealing with the problem of collaborative solution, reasoning and learning
algorithm model of each agent.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 56


2.5.4 Trends and Outlooks of AI Application to Power System
Fault Diagnoses [8]
As power systems becomes bigger in size and also complexity, especially the
emergent hybrid HVAC/DC system, the power system faults would become
more complex and impact on larger areas of the system. Traditional fault
diagnosis and protection and control methods would be increasingly challenged.
Application of AI in power system FD would find an increasingly critical role.
It is likely the following areas of researches would become the focus of activities
• Research on fault diagnostic methods combining multiple AI technologies.
Grid FD is a complex, dynamic and multi-level problem. Therefore, a single
artificial intelligent method is good at dealing with one or some problems but
suffers significant draw backs on others. Therefore, it is natural that
integration of different AI technologies, fusing together the strength of each
AI method and utilizing the latest technology in the field would be an
important trend in the development of future power system FD. This is likely
to significantly improve the robustness, accuracy and speed of the power
system FD.
• Research on FD methods based on multi-data source information fusion
technology. At present, most AI technologies used in power grid FD utilize
switching information. In comparison, electrical quantity has a greater
advantage in terms of accuracy and fault tolerance. Therefore, information
fusion techniques could be used to integrate information of different data
sources including switch and electric quantities, making full use of fault
information to produce more accurate and robust fault diagnosis result under
different operating conditions.
• Research on power grid FD methods based on Deep Learning (DL)
techniques. With the significant breakthrough in Deep Learning as typified in
AlphaGo, a number of researchers have started investigating its potential
application in power network FD. Power system produces and collects a large
volume of data through SCADA, RPMS, WAMS (PMU), weather, fault
recorders, expert knowledges, simulations, etc. Big data techniques could be
used to mine the valuable information and extract key attributes of power grid
faults characteristics. These data could be used to effectively train the DL

Dr. Babasaheb Ambedkar Technological University, Lonere Page 57


based FD method, which could produce more accurate and robust FD results
against a wide range of different power grid faults
• Practical research on power grid FD. At present, experts and scholars at home
and abroad have made many achievements in the research of grid FD theory,
but they are still insufficient in practical application. Therefore, in the future
research, how to apply the AI based FD method to actual power system and
demonstrate their effectiveness compared with traditional power system fault
identification and protection methods is an important research topic.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 58


Chapter Three
3.0 Advantages of AI in power system [9]
o Reduction in Human Error
o Takes risks instead of Humans
o Available 24x7
o Helping in Repetitive Jobs
o Digital Assistance
o Faster Decisions
o Daily Application
o New Inventions

3.1 Disadvantages of AI in power system [9]


o High coast of creation
o Making human lazy
o Unemployment
o No Emotions
o Laking out of box thinking

3.2 Current Application of AI In Power Systems


Several problems in power systems cannot be solved by conventional techniques
are based on several requirements which may not feasible all the time. In these
situations, artificial intelligence techniques are the obvious and the only option.
Areas of application of AI in power systems are:
• Operation of power system like unit commitment, hydro-thermal
coordination, economic dispatch, congestion management, maintenance
scheduling, state estimation, load and power flow.
• Planning of power system like generation expansion planning, power system
reliability, transmission expansion planning, reactive power planning.
• Control of power system like voltage control, stability control, power flow
control, load frequency control.
• Control of power plants like fuel cell power plant control, thermal power plant
control.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 59


• Control of network like location, sizing and control of FACTS devices.
• Electricity markets like strategies for bidding, analysis of electricity markets.
• Automation of power system like restoration, management, fault diagnosis,
network security.
• Applications of distribution system like planning and operation of distribution
system, demand side response and demand side management, operation and
control of smart grids, network reconfiguration.
• Applications of distributed generation like distributed generation planning,
solar photovoltaic power plant control, wind turbine plant control and
renewable energy resources.

Dr. Babasaheb Ambedkar Technological University, Lonere Page 60


CONCLUSION

The main feature of power system design and planning is reliability, which was
conventionally evaluated using deterministic methods. Moreover, conventional
techniques don’t fulfil the probabilistic essence of power systems. This leads to
increase in operating and maintenance costs. Plenty of research is performed to
utilize the current interest AI for power system applications. A lot of research is
yet to be performed to perceive full advantages of this upcoming technology for
improving the efficiency of electricity market investment, distributed control and
monitoring, efficient system analysis, particularly power systems which use
renewable energy resources for operation. The field of artificial intelligence
gives the ability to the machines to think analytically, using concepts. The main
feature of power system design and planning is reliability, which was
conventionally evaluated using deterministic methods. Moreover, conventional
techniques don’t fulfil the probabilistic essence of power systems. This paper is
based on the concept of artificial intelligence, areas of artificial intelligence and
the artificial intelligence techniques used in the field of Power Systems A lot of
research is yet to be performed to perceive full advantages of this upcoming
technology for improving the efficiency of electricity market investment,
distributed control and monitoring, efficient system analysis, particularly power
systems which use renewable energy resources for operation

Dr. Babasaheb Ambedkar Technological University, Lonere Page 61


REFERENCES
• Research Papers
1) international Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-5, Issue-1, May 2016
Artificial Intelligence in Power Station Arun Kumar D R, Murali Mohan B M

2) IOSR Journal of Computer Engineering (IOSR-JCE)


e-ISSN: 2278-0661, p-ISSN: 2278-8727 PP 00-00 Artificial Intelligence in
Power Systems R.Pasupathi Nath, V.Nishanth Balaji

3) 2009 International Conference on Artificial Intelligence and Computational


Intelligence YANG QI-PING, LI MENG-QUN Shanghai University of
Electric Power Shanghai, China

4) Yang Qiping, Xue Wude, Shanghai Institute of Electric Power, China, and
Lan Zhida, East China Electric Power Test &. Research Institute, China

5) Zahir J. Paracha1, Akhtar Kalam1, and Rubbiya Ali 2 1 School of


Engineering and Science, Victoria University, Melbourne, Australia.

6) A 0 Ekwe and J F Macqueen the National Grid Company plc, UK

7) Cosmin Darab, Radu Tarnovan, Antoniu Turcu, Corina Martineac


Department of Electric Power System Technical University of Cluj-Napoca
Cluj, Romania

8) Erxuan Chai1, (Peter)Pingliang Zeng1, Sicong Ma2, Hao Xing1, Bing


Zhao2
1. Hangzhou Dianzi University, Hangzhou 310018

Dr. Babasaheb Ambedkar Technological University, Lonere Page 62


• Reference Websites
9) https://towardsdatascience.com/advantages-and-disadvantages-of-artificial-
intelligence-182a5ef6588c

10) https://circuitglobe.com/power-system.html

Dr. Babasaheb Ambedkar Technological University, Lonere Page 63

You might also like