Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out

new PowerPoint Presentation - KSU Faculty Member websites

VIEWS: 65 PAGES: 49

									Artificial Intelligence
  Applications to
 Digital Protection

            By
Hossam Eldin Abdallah Talaat
                 Scope of the Study

           AI Applications to Digital Protection

Transmission Line        Rotating Machines       System fault Diagn.
           Fault 6.Winding
1. Fault Classification       Winding Protection
                                                   Alarm
                                                   10. Alarm Processing
                        Incipient Fault
      Direction Protection Processing Est.    Faulted Section
 Classification Incipient Fault Detect 11. Faulted Section
       Distance Detection
2. Direction Discriminat   7.
Discrimination
3. Distance Relaying                             Estimation
                          DifferentialDistribution
       Relaying
          Series Relaying Out-of-Step
4. Series Compensated l.
                               Transformer        System Protect.
                                                    12. Distribution
5. AC/DC Transmission                         Relay Out of Step Protect.
                         Transformer 13. Setting&
 CompensatedDifferential Relaying Protection
         AC/DC             8.
                                                Coordination
            Line Fault Diagnosis 14. Relay Setting& Coor
 Transmission              9. Fault Diagnosis


        Systems
                  Methodology
    For each protection area:
   Motivation of applying AI (Problems).
   Detailed description of a selected application.
   Other AI Applications: differences & additional
    features.
   Summary of attributes of all AI applications:
    (ref#, functions, AI technique, Input features,
    pre-processing & drawbacks).
   Discussion.
   Functional Requirements of Power System Protection

                        Objectives of Power
                         System Protection



      Selectivity
   Selectivity                Reliability
                                Reliability               Speed
                                                       Speed
     Max service             Max service          Min fault time &
                Simplicity
              Simplicity
 continuity with min continuity with min Economy
                                               Economy damage
                                                 equipment
system disconnection system disconnection
          Min equipment and              Max performance at
                 circuitry                    min cost
                         Security
                      Security           Dependability
                                      Dependability
                  Ability to avoid Ability to perform
                    unnecessary         correctly when
                     operation              needed
       Development in Power System Relaying
                                                          AI-Based Relays
                                                            (Intelligent)


             Performance                   Microprocessor- Communication
                                            Based Relays       Facility
                                              (Digital)
                                                             AI-based
                                                             Methods
                                Static       Digital Proc.
                                Relays        Algorithms     Digital ICs
  Electromechanical Relays
                                             Digital ICs   (mP,DSP,ADC,
                               Electronic                    neuro-IC
                                          (mP,DSP,ADC,)
                                Circuits                     fuzzy-IC)

1900             years       1960     1975             2000
Characteristics of Digital Relaying

   Self-diagnosis: improving reliability.
   Programmability: multi-function, multi-
    characteristic, complex algorithms.
   Communication capability: enabling
    integration of protection & control.
   Low cost: expecting lower prices.
   Concept: no significant change (smart
    copy of conventional relays).
  Shortcomings of Conventional Protection Systems

Protection Area                Dependability   Security Speed selectivity
TL fault classification
                                   X            X        XX        -
Distance Relaying
                                   X            X        XX       X
Machine Winding Relaying
                                   XX            -        X        -
Transformer differ. relaying
                                   X            XX        -        -
Transformer fault diagnosis
                                   XX            -        -        -
HIF detection
                                   XX            -        -        -
Relay setting& coordination
                                    -            -        -       XX

Key: “-” no problem, “X” some problems, “XX” big problems
    Motivation for AI-Based Protection


 Enabling the introduction of new relaying
  concepts capable to design smarter, faster,
  and more reliable digital relays.
 Examples of new concepts: integrated
  protection schemes, adaptive protection &
  predictive protection.
Classification of AI Techniques
                Artificial Intelligence
                  (AI) Techniques


      Symbolic Knowledge          Computational
        Representation             Knowledge
                                  Representation



   Exact         Approximate         Artificial
 Reasoning        Reasoning           Neural
                                     Network
                                      (ANN)

Expert System     Fuzzy Logic
    (ES)             (FL)
    Characteristics of Expert Systems

     Adavantages                    Drawbacks
• Availability              o Lack of information
• Comprehensiveness         o Brittleness (noise)
• Generalization            o Expertise-based shortcomings
• Explanation               o Expert-based shortcomings
• User friendly interface
     Characteristics of Artificial
     Neural Networks (ANN)
          Advantages                    Drawbacks
• Powerful pattern classification. o Network design
 Optimization capabilities.        using trial & error
• MLP (Back-propagation): Classifiaction and Nonlinear (no.
Mapping                             of layers, no. of
• Fast response.
                                    neurons in hidden
 Kohonen (Self-organizing Map): Feature Extraction
• Fault tolerant (noise).
                                    layer, learning rate,
 Excellent generalization.
•
  Hopfield (Recurrent): Optimization
                                    etc.
• Trend prediction.                 o Generation of large
• Good reliability.                 training set.
 Steps of Designing an AI-Based Protective Scheme
System model,
parameters &                           Training target
                                                          Classifier output
  operating
  conditions                                                 (training)
                                 Training Set        Pattern
                                                    Classifier
 Simulation     Samples of 3-ph
                      Anti-    Filtered Samples               Training error
Environment     Voltages & Currents Feature
                     aliasing
  “EMTP”                                                   Classifier
                     & other        Extraction
                                                          parameters
                     Filters
                                                             Classifier output
                                                                  (testing)
 Fault type,                      Testing Set         Pattern
 location &                                          Classifier
  duration            Testing target
                                   Performance Evaluation
                                                                  Testing error
      Modules of Intelligent Transmission Line Relaying


                     Features
  V            Data             Transmission Line
            Processing          Fault Identification
  I



 Fault                        Fault Type                       Arcing
Detection                    Classification                   Detection

              Direction                       Faulted Phase                Fault
            Discrimination                      selection                 Location




                                 Decision Making                    Trip Signal
Application 1
Transmission Line Fault Classification
     Motivation
  Conventional schemes: cannot adapt to
   changing operating condtions, affected by
   noise& depend on DSP methods (at least 1-
   cycle).
  Single-pole tripping/autorecloser SPAR
   requires the knowledge of faulted phase (on
   detecting SLG Single-pole tripping is
   initiated, on detecting arcing fault recloser is
   initiated).
      Transmission Line Relaying Scheme
                                             Fault Type    5-7 ms
                                                           RG
 1/4 cycle    VR,VS,VT                                               K
   each
(5 samples)                                                RST       N
               IR,IS,IT
                                                                     O
                          ANN1
 45000                30-20-15-11          Control Logic             W
 training                                                            L
                      Enabling Signals
 patterns                                                 25 ms      E
                                    VR                               D
                                                          Arcing     G
                                     ANN2
                                    20-15-10-1
                                                          fault      E   Decision
                                                           phase-R
                         One
                        cycle       VS                               B
                        each                              Arcing     A
                                     ANN3
                         (20        20-15-10-1
                                                          fault      S
                      samples)                             phase-S
                                                                     E
                                    VT
                                                          Arcing
                                     ANN4
                                    20-15-10-1
                                                          fault
                                                           phase-T
    Detailed Topology of ANN1
            VR(k)
                                         Output Layer
             IR(k)
                                              (11 )
            VS(k)                                     RG
            IS(k)                                     SG
            VT(k                                      TG
  Input     ) T(k)
            I                                         RS
                .                                     ST
 voltage        .                                     TR
                .
&current        .                                     RSG
                .                                     STG
samples         .
                                                      TRG
                                                      RST
                                                      Normal
                                   Hidden Layer 2
           VT(k-4)                     (15 )
                           Hidden Layer 1
           IT(k-4)
                 Input Layer (20 )
                     (30 )
Other AI Applications
   Fuzzy & fuzzy-neuro classifiers used for fault type
    classification (1-cycle).
   Pre-processing: 1- Changes in V&I,
    2- FFT to obtain fundamental V&I,
    3- Energy contained in 6 high freq. bands obtained
    from FFT of 3-ph voltage.
   Measures from two line ends.
   Implementation of a prototype for ANN-based
    adaptive SPAR relay using transputer system
    (T800).
Application 2:
Distance Relaying
Motivation


 Changing the fault condition, particularly in
  the presence of DC offset in current
  waveform, as well as network changes lead
  to problems of underreach or overreach.
 Conventional schemes suffer from their
  slow response.
AI Applications in Distance Relaying

 Using ANN schemes with samples of V&I
  measured locally, while training ANN with
  faults inside and outside the protection zone.
 Same approach but after pre-processing to get
  fundamental of V&I through half cycle DFT
  filter.
 Combining conventional with AI: using ANN
  to estimate line impedance based on V&I
  samples so as to improve the speed of
  differential equation based algorithm.
AI Applications in Distance Relaying

 Pattern Recognition is used to establish the
  operating characteristics of zone-I. The
  impedance plane is partitioned into 2 parts:
  normal and fault. Pre-classified records are
  used for training.
 Application of adaptive distance relay using
  ANN,where the tripping impedance is
  adapted under varying operating conditions.
  Local measurements of V&I are used to
  estimate the power system condition.
Application 3:
Machine Winding Protection
Motivation
 If the generator is grounded by
  high impedance, detection of
  ground faults is not easy (fault
  current < relay setting).
 Conventional algorithms suffer
  from poor reliability and low speed
  (1-cycle).
ANN-Based Generator Winding Fault Detection
                                                                     A
                                                                     B
                                                                     C
               Ia1 Ib1 Ic1                          Ia2 Ib2 Ic2

                             Sampling
 Ra
      Ia1(n) I (n)           Ic1(n) Ia2(n) Ib2(n)         Ic2(n)
              b1
               Current Manipulator                                   Iad(n) = Ia1(n)- Ia2(n)
      Iad(n)           Iaa(n) Ibd(n) Iba(n) Icd(n)          Ica(n)   Iaa(n) = ( Ia2(n) + Ia1(n) )/2
                      DFT Filtering
               In1     In2       In3     In4        In5     In6




                L-G              L-L            L-L-L
               ANN1             ANN2            ANN3

                     Output            Output          Output
Application 4:
Transformer Differential Relaying
Motivation

 Conventional differential relays may fail in
  discriminating between internal faults and
  other conditions (inrush current, over-
  excitation of core, CT saturation, CT ratio
  mismatch, external faults,..).
 Detection of 2nd and 5th harmonics is not
  sufficient (harmonics may be generated
  during internal faults).
Multi-Criteria Differential Relay
based on Self-Organizing Fuzzy Logic

 One differential relay per phase.
 12 criteria are used and integrated by FL.
 Examples of criteria: (ID=differential current)
    Sign   Definition          Criterion Statement
    q1         ID1      q1> highest expected inrush current
    q3      ID2/ID1 q < 10-15%
                     3
    q4         ID1      q4 > current for over-excitation
    q6      ID5/ID1 q6 < 30%
     Fuzzy Logic Based Multi-Criteria
     Differential Transformer Relay
                         Fuzzification
       v                                           Weighting     Ruling-out the
                 q1                      m1         Factors      hypothesis of
            M                                 W1                 inrush current
            E
      CB         q2
            A                        m3 m2 W2
       i1   S                              Non-Inrush
CT          U                 1.0
                 q3                      m3
            R                                 W3
            I
                q4
            N            hypothesis of stationary     w2 w1    q3
            G                0.0
                      overexcitation of a transformer
                                         0.08      0.12       MIN
                                                            0.16
       i2       q7                 core
CT                                                                              d
                      hypothesis of an external S.C.
            U         combined with CTs saturation w3                               Trip
            N q10                                                          d>D
      CB
            I          hypothesis of an external fault         w4
            T         combined with ratios mismatch
                q12                                             Tripping    D
                                                               threshold
Other AI Applications
   ANN approaches with training using inrush current,
    external & internal faults.
   Input features: 3-ph current samples expressed as
    differential and retraining OR apply FFT to get
    fundamental, 2nd & 5th harmonics.
   A prototype is implemented using DSP card with
    the objective of reconstruction of distorted CT
    secondary current due to saturation. Tested on 50
    MVA plant transformer with time response 5-10
    ms.
APPLICATION 5:
Transformer Fault Diagnosis
Motivation

 Conventional methods, e.g., Dissolved Gas
  Analysis (DGA), suffers from imprecision
  & incompleteness.
 IEC/IEEE code for DGA relates the fault
  type to the ratios of gases; e.g.,
   IF (C2H2/C2H4 =0.1-3) AND (CH4/H2 < 0.1) AND
    (C2H4/C2H6 < 1) THEN (the fault is High energy partial
    discharges)
Transformer Fault Diagnosis using
GA-based Fuzzy Classification
     Genetic Algorithm   Data Base of
           (GA)          Dissolved Gas
        Optimizer        Test Records


          Set up          Transformer
        Membership       Fault Diagnosis
        Functions &         System
        Fuzzy Rules


         IEC/IEEE           Diagnosis
        Transformer          Results
       DGA Criterion
Each subspace is described by a fuzzy if-then
rule based on the patterns of training set.
                            S
                                    CH4/H2
                        M                           L
                L
                                                    M


                                                     S
                                                 C2H2/C2H4


    C2H4/C2H6

                    S           M            L
Application 6:
Alarm Processing
 The enormous no of signals and alarms after
  a fault occurrence complicates the fault
  diagnosis process.
 ES versus ANN:
 ES is better for: large power systems and
  explanatory facility.
 ANN is better for: noisy inputs, low cost
  and fast response.
 Some practical implementations of ES:
  Wisconsin, Taiwan and Portugal.
Hardware Implementation
 Fuzzy Processors:
 Siemens SAE81C99: 256/128 I/O, 16384
  rules, 10 M fuzzy logic instruction per sec.
 Siemens SAE81C991: 4096/1024 I/O,
  131072 rules, 10 M FL instruction per sec.
 Neuro-Processors:
 Analog or Digital implementation but not
  yet commercialized.
 Example: 1000 neuron, 1M synapses,
  1.37M connection per sec.
Hardware Implementation
 Advanced Communication Systems:
 Synchronized sampling can be obtained at
  0.2-0.5ms using Global Positioning
  Systems (GPS) satellite.
CONCLUSIONS
 Expert Systems of system fault diagnosis and
  relay coordination has been practically
  Implemented.
 Some prototypes of ANN-based relays have
  been implemented and tested using laboratory
  setups.
 Major problem facing the practical application
  of AI-based relays is the generation of training
  patterns from comprehensive computer
  simulation.
Application 7:
Relays Setting & Coordination
Motivation

 Setting and coordination of relays in
  complex power networks requires computer
  aids especially for meshed networks.
 The problem is non-algorithmic, i.e.,
  application of expert system ES is needed.
Expert System for Setting& Coordination of Distance Relays

      Formation of         Loop                Break     Control Rules
    Primary/ Backup     Enumeration            Points
       Pairs Rules         Rules               Rules
                                                          Inference
       Relative           Set of           Setting and                         Facts
                                                           Engine
    Sequence Vector   Sequential Pairs    Coordination
         Rules            Rules              Rules
                                                           Agenda


2                                1        11                22        5
         14                 13
                                          23                24
    15                           12                                       21




    16                           19
3                                     4
         17                 18            20
           loop 1 23 22
Rule 3: Primary/Backup Pairs
If         loop (R1) 11
         Relay 2 24 is located on line (L1) at bus (B1),
           Line (L1) is 18 16 14
   AND loop 3 11 21connected between bus (B1) & bus (B2),
           Relay 23 21 located
   AND loop 4 (R2) is18 16 14 on line (L2) at bus (B2);
   AND .Line (L2) is not line (L1),
           .
THEN Relay (R1) acts as a buckup to relay (R2)
           loop 11 19 11 21
           loop 12 19 23
Rule 9: Zone-2 Overlap21
If         break-points buckup to
         Relay (R1) is a23 11 17 12 relay (R2),
           Zone-2 setting for 15 12
   AND break-points 23 11 relay (R1) is (X12),
           Zone-1 setting for 13 12
   AND break-points 23 11 relay (R2) is (X21),
           Relay (R1) 23 11 17 on
   AND chosen-B.P. is located12 line (L1)
   AND Line (L1) has a reactance equal (Xp),
           RSV 23 11 17 12 15 13 24 22 14 16 21
   AND (X12-Xp) > (X21),
   AND SSP 23 21 23 22
           Time delays of zone-2 of (R1) and zone-2 of (R2) are equal;
           SSP 11 24 11 delay of zone-2 for relay (R1) by one grading time
THEN Increase time 21
           .
unit (0.2 s)
           SSP 21 23 21 11 21 13 21 16
   Structure of Rule-Based Expert System

      Definition: Expert System
Knowledge                                   Data
  Baseis a computer program
                  Inference Engine          Base
 (Rules)                                   (facts)
      that uses knowledge and
      inference procedures to
          Explanation
                               Knowledge
                               Acquisition
            Facility
      solve problems that are    Facility

      ordinarily solved through
                   User Interface
      human expertise
    Classification of ANN Models
                    ANN Models



         Feedback                      Feed
                                     Forward


Constructed    Trained      Linear         Nonlinear


                                 Unsupervised     Supervised
 Hopfield      Adaptive
(recurrent)    Resonance           Kohonen          MLP
                                    (Self-         (Back-
                                  Organizing     Propagation
                                    Map)
            Main Components of Fuzzy Logic Reasoning

                               Inference methods:
                              Max-Min composition,
                              Max-Average comp., ..
             X1 is 20% BIG&
             80% MEDIUM                                                 Output
  Input                                                                 Decision
variables
            Fuzzification              Fuzzy                 Defuzzification
                                     Inference




            Membership           Fuzzy If-Then Rules          Defuzzification
             functions                                          methods:
                              If X1 is BIG and X2 is SMALL
                                                              Center of area
                                       Then Y is ON,
                                                              Center of sums
                                If X1 is BIG and X2 is BIG
                                                             Mean of Maxima,..
                                       Then Y is OFF.
                            ..
    M
V
    E       V
    A
    S   V
    U
    R
    I
    N
    G

    U
    N
    I
    T
c- Fuzzy Rule-based Classification
   A5
             R15   R25   R35   R45   R55

   A4        R14   R24   R34   R44   R54



   A3        R13   R23   R33   R43   R53



   A2        R12   R22   R32   R42   R52


  A1         R11   R21   R31   R41   R51


        A1         A2    A3    A4      A5

								
To top