of x

Neural Networks in Power Electronics

Published on December 2016 | Categories: Documents | Downloads: 11 | Comments: 0
44 views

neural network application to power electronics devices

Comments

Content




ARTIFICIAL INTELLIGENCE TECHNIQUES IN POWER
ELECTRONICS AND MOTOR DRIVES

Bimal K. Bose



Abstract:
Artificial intelligence (AI) techniques, such as expert system (ES), fuzzy logic (FL), artificial neural network
(ANN), and genetic algorithm (GA) have recently brought a new and advancing frontier in power electronics and
motor drives area, which is already a complex and interdisciplinary technology. The goal of AI is to plant human
intelligence in a computer so that a computer can think intelligently like a human being. Computational
intelligence has been debated over a long time. There is no denying the fact that AI techniques can solve complex
problems which are difficult to solve by traditional methods. Currently, AI technology is advancing at a fast rate,
and its applications in power electronics and motor drives are growing fast, as evident by large number of
publications in IEEE journals. Recent advancement of powerful DSPs and FPGAs is making implementation of
fuzzy and neural systems economical with improvement of performance, compact and more competitive.
Evidently, the future impact of this technology on power electronics and motor drives is very significant. The
frontier of AI is bringing a new challenge to the traditional engineers specialized in power electronics and motor
drives.
The tutorial presentation will discuss comprehensively the principles of AI and its applications in power electronics
and motor drives. In the beginning, the importance of AI will be reviewed, which will be followed by brief
discussion on principles of different AI techniques. However, the presentation will mainly focus on fuzzy logic and
neural network (main focus) applications in the control and estimation of power electronic systems, illustrating
some application examples. Fuzzy logic example applications will include robust motor speed control, online
efficiency optimization of ac drive, and optimal control of modern wind generation system. The ANN application
examples will include space vector PWM wave synthesis for 2-level and multi-level converters, delayless filtering
of inverter output waves, waveform generation for converters, model referencing adaptive control (MRAC) of ac
drives, drift-free flux estimation of drives (approaching zero speed), and neuro-fuzzy control of drives. Time
permitting, computer-aided design examples of fuzzy and neural systems will be discussed. Finally, in conclusion,
the future prognosis of the technology will be reviewed.

Contacts:
Bimal K. Bose University of Tennesse, USA [email protected]

ARTIFICIAL INTELLIGENCE TECHNIQUES IN
POWER ELECTRONICS AND MOTOR DRIVES
Dr. Bimal K. Bose, Life Fellow, IEEE
Department of Electrical Engineering and Computer Science
The University of Tennessee
Knoxville, TN 37934, USA
Tel: (865)974-8398, Fax: (865)974-5483
E-mail: [email protected]
Home Page: web.eecs.utk.edu/~bose
1
September 18, 2011
1:00 PM - 3:00 PM, 3:30 PM - 5:00 PM
(Tutorial)
2
SELECTED REFERENCES
• B. K. Bose, Modern Power Electronics and AC Drives, Prentice-Hall, Upper Saddle River, NJ, 2002.
• B. K. Bose, Power Electronics and Motor Drives - Advances and Trends, Elsevier, Burlington, MA, 2006.
• L. H. Tsoukalas and R. E. Uhrig, Fuzzy and Neural Approaches in Engineering, Wiley, NY, 1997.
• S. Haykin, Neural Networks, Macmillan, NY, 1994.
• Math Works Inc., Fuzzy Logic Toolbox User’s Guide, Jan. 1998.
• I. Miki, N. Nagai, S.Nishigama, and T. Yamada, “Vector control of induction motor with fuzzy PI controller”, IEEE IAS
Annu. Meet. Conf. Rec., pp. 342-346, 1991.
• G. C. D. Sousa, B. K. Bose, and J. G. Cleland, “Fuzzy logic based on-line efficiency optimization control of an
indirect vector controlled induction motor drive”, IEEE Trans. Ind. Electron., vol. 42, pp. 192-198, Apr. 1995.
• M. G. Simoes, B. K. Bose, and R. J. Spiegel, “Design and performance evaluation of a fuzzy logic based variable
speed wind generation system”, IEEE Trans. Ind. Appl., vol. 33, pp. 956-965, July/Aug. 1997.
• Math Works Inc., Neural Network Toolbox User’s Guide, 2001.
• B. K. Bose, “Neural network applications in power electronics and motor drives – an introduction and perspective” ,
IEEE Trans. Ind. Electron., vol. 54, pp. 14-33, Feb. 2007.
• J. Zhao and B. K. Bose, “Neural network based waveform processing and delayless filtering in power electronics
and ac drives”, IEEE Trans. Ind. Electron., vol. 51, pp. 981-991, October 2004.
• M. G. Simoes and B. K. Bose, “Neural network based estimation of feedback signals for a vector controlled
induction motor drive”, IEEE Trans. Ind. Appl., vol. 31, pp. 620-629, May/June 1995.
• J.O.P. Pinto, B. K. Bose, L. E. B. da Silva, and M. P. Kazmierkowski, “A neural network based space vector PWM
controller for voltage-fed inverter induction motor drive”, IEEE Trans. Ind. Appl., vol. 36, pp. 1628-1636, Nov./Dec.
2000.
• C. Wang, B. K. Bose, V. Oleschuk, S. Mondal, and J. Pinto, “Neural network based SVM of a 3-level inverter covering
overmodulation region and performance evaluation on induction motor drives”, IEEE IECON Conf. Rec., pp. 1-6,
2003.
• L. E. B. da Silva, B. K. Bose, and J.O.P. Pinto, “Recurrent neural network based implementation of a programmable
cascaded low pass filter used in stator flux synthesis of vector controlled induction motor drive”, IEEE Trans. Ind.
Electron., vol. 46, pp. 662-665, June 1999.
• J.O.P. Pinto, B.K.Bose, and L.E.B. da Silva, “A stator flux oriented vector controlled induction motor drive with
space vector PWM and flux vector synthesis by neural networks”, IEEE Trans. Ind. Appl., vol. 37, pp. 1308-1318,
Sept./Oct. 2001.
• B. K. Bose, “Power electronics and motor drives –recent progress and perspective”, IEEE Trans. Ind. Electron., vol.
56, pp. 581-588, Feb. 2009.
WHAT IS ARTIFICIAL INTELLIGENCE (AI)?
• HUMAN BRAIN WITH BIOLOGICAL NEURAL NETWORK HAS
NATURAL INTELLIGENCE – ABILITY TO LEARN, REASON AND
COMPREHEND
• GOAL OF AI – PLANTING HUMAN INTELLIGENCE IN COMPUTER SO
THAT COMPUTER CAN THINK INTELLIGENTLY LIKE HUMAN BEING
• CAN A COMPUTER REALLY THINK AND TAKE INTELLIGENT
DECISION?
• COMPUTER INTELLIGENCE – FAR INFERIOR TO NATURAL
INTELLIGENCE. HOWEVER, IT HELPS TO SOLVE COMPLEX PROBLEMS
• AI TECHNIQUES ARE USED EXTENSIVELY IN
- INDUSTRIAL PROCESS CONTROL
- MEDICINE
- GEOLOGY
- INFORMATION MANAGEMENT
- MILITARY SYSTEM
- SPACE TECHNOLOGY ETC.
3
ARTIFICIAL INTELLIGENCE (AI) CLASSIFICATION
• EXPERT SYSTEM (ES)
• FUZZY LOGIC (FL)
• ARTIFICIAL NEURAL NETWORK (ANN)
OR NEURAL NETWORK (NNW)
• GENETIC ALGORITHMS (GA)
4
FEATURES OF EXPERT SYSTEM
• INTELLIGENT COMPUTER PROGRAM BASED ON BOOLEAN LOGIC
• PLANTS HUMAN EXPERTISE IN A CERTAIN DOMAIN IN COMPUTER TO
REPLACE THE HUMAN EXPERT
• A SET OF IF ……. THEN ….. RULES USING “YES” – “NO” OR 1-0 LOGIC
• DEFINED AS CLASSICAL AI – FORERUNNER OF ALL AI TECHNIQUES
- DEVELOPED IN 1970’S AND APPLIED EXTENSIVELY IN 1980’S
• A KNOWLEDGE ENGINEER ACQUIRES THE KNOWLEDGE FROM THE
TECHNICAL EXPERT AND PLANTS IN KNOWLEDGE BASE
• APPLICATIONS IN POWER ELECTRONICS:
- ADAPTIVE P-I TUNING CONTROL
- FAULT DIAGNOSTICS
- AUTOMATED DRIVE TEST AND PERFORMANCE EVALUATION
- SELETION OF VENDOR PRODUCTS FOR A SPECIFIC APPLICATION
- AUTOMATED DESIGN AND SIMULATION OF CONVERTER-FED
DRIVE
ETC.
5
WHAT IS INTELLIGENT CONTROL AND ESTIMATION?
• CONTROL AND ESTIMATION BASED ON ARTIFICIAL
INTELLIGENCE (AI) TECHNIQUES
• OFTEN DEFINED AS:
- LEARNING CONTROL
- SELF-ORGANIZING CONTROL
- SELF-ADAPTIVE , ADAPTIVE OR ROBUST
CONTROL
• GOOD CANDIDATE WHERE:
- MATHEMATICAL PLANT MODEL MAY NOT
EXIST
- MODEL IS ILL-DEFINED
- PLANT PARAMETER VARIATION PROBLEM
- COMPLEX NONLINEAR CONTROL SYSTEM
• NEEDS POWERFUL DSP OR ASIC CHIP FOR
IMPLEMENTATION
6
FEATURES OF FUZZY LOGIC(FL)


• BOOLEAN OR CRISP LOGIC: 1(Yes), 0 (No)
FUZZY LOGIC : MULTI-VALUED (0 TO 1)

• EMULATION OF FUZZY HUMAN THINKING

ES EXAMPLE:
IF: SPEED < 1000 RPM
THEN: CURRENT > 50 A

FL EXAMPLE:
IF: SPEED OF THE DRIVE MOTOR IS LOW
THEN: CURRENT SHOULD BE LARGE

• FUZZY VARIABLES (SPEED, CURRENT) AND LINGUISTIC FUZZY SETS (LOW,
LARGE) ARE REPRESENTED BY MEMBERSHIP FUNCTIONS

• INVENTED BY LOTFI ZADEH OF UNIVERSITY OF CALIFORNIA, BERKELEY IN
1965

7
8
9
10
FUZZIFICATION
DEFUZZIFICATION
IMPLICATION
INPUT VARIABLES
OUTPUT
VARIABLE
MFs
11
12
LINEAR COMBINATION
OF INPUTS
13
FUZZY OUTPUT
14
Example 1: Robust Control of Induction Motor
TL
MAMDANI METHOD OF FUZZY INFERENCE SYSTEM
USING TRIANGULAR MEMBERSHIP FUNCTIONS
DEFUZZIFICATION
FUZZIFICATION
IMPLICATION
15
16
FUZZY SET MEMBERSHIP FUNCTIONS OF FUZZY SPEED CONTROLLER VARIABLES
17
18
19
Example 2: IM Efficiency Optimization
20
21
• ADAPTIVE STEP SIZE FOR FAST CONVERGENCE
• TRANSIENT RESPONSE OPTIMIZATION
• EFFICIENCY OPTIMIZATION AT STEADY STATE
FUZZY EFFICIENCY OPTIMIZER BLOCK DIAGRAM
22
23
TRANSITION BETWEEN EFFICIENCY OPTIMIZATION MODE
AND TRANSIENT OPTIMIZATION MODE
24
EFFICIENCY OPTIMIZATION PEREFORMANCE
WITH PULSATING TORQUE COMPENSATION
(ωr = CONSTANT)
25
26
Example 3: Wind Generation System Control
27
WIND TURBINE
IG
OPTIMUM SPEED
SEARCH
OPTIMUM
FLUX
SEARCH
ROBUST
SPEED
CONTROL
28
WHY FUZZY CONTROL IN WIND GENERATION SYSTEM?
29
30
MATLAB FUZZY LOGIC TOOLBOX ENVIRONMENT
31
INDIRECT VECTOR-CONTROLLED DRIVE WITH FUZZY iqs AND ids CONTROL
Example 4: IM Vector Drive
32
FUZZY CONTROLLER MFs FOR ids LOOP
33
FUZZY CONTROLLER RULE TABLE FOR ids LOOP
34
SIMULINK SIMULATION
OF THE DRIVE
SIMULINK SIMULATION OF
“VECTOR CONTROLLER”
SHOWING FUZZY
Ids AND iqs CONTROL
35
CONTROL SURFACE OF
FUZZY ids CONTROLLER
RESPONSE OF ids LOOP WITH
FUZZY FUZZY AND PI CONTROLS
36
RESPONSE OF iqs LOOP WITH FUZZY AND PI CONTROL
37
CONCLUSION ON FUZZY LOGIC
• FUZZY CONTROL POSSIBLY IS THE BEST CONTROL FOR ROBUST PERFORMANCE
IN NONLINEAR FEEDBACK SYSTEM WITH PARAMETER VARIATION AND LOAD
TORQUE DISTURBANCE
• FUZZY CONTROL IS EASY TO DESIGN BY TRIAL AND ERROR APPROACH BASED ON
EXPERIENCE ON SYSTEM RESPONSE – NO MATHEMATICAL MODEL MAY BE
NEEDED
• NEURAL NETWORK BASED ANFIS (ADAPTIVE NEUROFUZZY SYSTEM INFERENCE
SYSTEM) IS GETTING MORE ACCEPTANCE
• BASICALLY NONLINEAR MULTI-DIMENSIONAL INPUT-OUTPUT STATIC MAPPING
• ONLINE FAULT DIAGNOSTICS AND FAULT-TOLERANT CONTROL – EXCELLENT
AREA OF FL APPLICATION
• CURRENTLY, APPLICATIONS ARE LIMITED IN POWER ELECTRONICS AND DRIVES
AREA
FUZZY SPEED CONTROLLER DESIGN DEMO
USING MATLAB/FUZZY LOGIC TOOLBOX
(USER’S MANUAL 2-29 )
(Make an initial design of MFs and Rule Table before using Toolbox)
1. ACCESS MATLAB FUZZY LOGIC TOOLBOX
>>fuzzy
FIS (Fuzzy Inference System) Editor arrives
* Select “mamdani” from FILE
* Add input variable from Edit
* Name input1 as E
* Name input2 as CE
* Name output1 as dU
* From File Export to workspace as SPEED
2. ENTER MFs OF INPUT AND OUTPUT VARIABLES
* Highlight E
* From Edit select MFs (Or doubleclick on E)
* Select triamf (triangular MFs)
* Select 7
* Select range and display [-1 1]
* Assign names of MFs and parameter values
* Select CE and dU and repeat the above steps
38
3. ENTER RULE BASE
* From Edit select RULES (or doubleclick on SPEED)
- Rule Box appears
* Select MFS from E, CE and dU with AND to construct Rule Table
- save
4. RULE AND SURFACE VIEWER
* From VIEW select RULES
* Move vertical lines to select E and CE (or give input values)
- Generates defuzzified dU by centroid principle
* From VIEW select SURFACE
- generates 3-dimensional control surface with E, CE and dU
* Save the FIS on a disk
5. SPEED can be linked in a Simulink program for system testing
>>Simulink
* Click Fuzzy Logic Toolbox
* Open new model
* Drag “fuzzylogic controller block and load SPEED
* Generate Simulink model for testing
39
NEURAL NETWORK FEATURES
(ANN or NNW)
• INPUT-OUTPUT NONLINEAR SIGNAL MAPPING CHARACTERISTICS
PATTERN RECOGNITION
PATTERN CLASSIFICATION
ASSOCIATIVE MEMORY
• KNOWLEDGE IS ACQUIRED BY LEARNING ( OR TRAINING)
THROUGH EXAMPLE INPUT-OUTPUT DATA SETS
• HIGH SPEED PARALLEL COMPUTATION – WITH FAULT-TOLERANCE
AND NOISE FILTERING PROPERTY
• TYPICAL ENGINEERING APPLICATIONS:
- CONTROL AND ESTIMATION IN POWER ELECTRONIC SYSTEM
- GENERAL INDUSTRIAL PROCESS CONTROL
- ROBOT VISION
- ON-LINE DIAGNOSTICS, ETC. ETC.
40
BIOLOGICAL NEURON
MODEL OF ARTIFICIAL
NEURON
41
42
HARD LIMITER
SIMPLE EXAMPLE OF NNW APPLICATION
Y = A Sin X for –Pi <X<+Pi
NEURON WITH HYPERBOLIC TAN TF
Example 1:
SAWTOOTH
WAVE
43
44
SOME MODELS OF NEURAL NETWORK
• PERCEPTRON
• ADALINE AND MADALINE
• BACKPROPAGATION (BP) NETWORK
• RADIAL BASIS FUNCTION NETWORK(RBFN)
• MODULAR NEURAL NETWORK (MNN)
• LEARNING VECTOR QUANTIZATION (LVQ) NETWORK
• FUZZY NEURAL NETWORK (FNN)
• KOHONEN’S SELF-ORGANIZING FEATURE MAP (SOFM)
• ADAPTIVE RESONANCE THEORY (ART) NETWORK
• REAL TIME RECURRENT NETWORK
• ELMAN NETWORK
• HOPFIELD NETWORK
• BOLTZMANN MACHINE
•RECIRCULATION NETWORK
• BRAIN-STATE-IN-A-BOX (BSB)
• BI-DIRECTIONAL ASSOCIATIVE MEMORY (BAM) NETWORK
45
1 1 1,1 2 1,2 1 1 2
1 n pW p W b p p = + + = − + +
TRAIN BY ANN TOOLBOX
USING EXAMPLE DATA
SOLVE PATTERN
CLASSIFICATION PROBLEMS
46
FEATURES OF BACKPROPAGATION NETWORK


• LOGICAL OR CONTINUOUS INPUT AND OUTPUT SIGNALS WHICH CAN BE
UNIPOLAR OR BIPOLAR

• SYNAPTIC WEIGHTS CONSTITUTE DISTRIBUTED INTELLIGENCE – SIMILAR
TO HUMAN MEMORY OR INTELLIGENCE

• NONLINEAR INPUT-OUTPUT MAPPING OR PATTERN RECOGNITION
PROPERTY

• FAST PARALLEL COMPUTATION BY ASIC CHIP – INSTEAD OF SLOW
SEQUENTIAL COMPUTATION BY DSP

• FAULT TOLERANCE PROPERTY

• NOISE IMMUNITY PROPERTY

• REQUIRES SUPERVISED TRAINING BY EXAMPLE DATA SETS
- SIMILAR TO SUPERVISED ALPHABET TRAINING OF A
CHILD

• BACKPROPAGATION TRAINING ALGORITHM BY A COMPUTER PROGRAM

47
48
MAPPING OF INPUT LETTER “A” BY FIVE-BIT BINARY CODE
Example 2: Optical Character Recognition – Data Compression
49
(a) INVERSE MAPPING OF THE LETTER “A”
(b) AUTOASSOCIATIVE MAPPING OF “A”
(a)
(b)
FLOWCHART FOR BACKPROPAGATION TRAINING
50
51
52
MINIMIZATION OF SQUARE ERROR BY GRADIENT DESCENT METHOD
(a) STRUCTURE OF REAL TIME RECURRENT NETWORK
(b) BLOCK DIAGRAM FOR DYNAMIC BACKPROPAGATION TRAINING
53
ANN IDENTIFICATION OF DYNAMIC PLANT MODEL
TIME-DELAYED NNW (TDNNW) WITH TAPPED DELAY LINE
(DYNAMICAL NNW)
Y(k)
54
TRAINING OF INVERSE DYNAMIC
MODEL OF A PLANT
INVERSE MODEL-BASED ADAPTIVE
CONTROL
INVERSE MODEL CANCELS
THE FORWARD MODEL
55
INVERSE DYNAMICS BASED ADAPTIVE CONTROL
MODEL REFERENCING ADAPTIVE CONTROL (MRAC) OF A PLANT
DIRECT MRAC
INDIRECT MRAC
GENERATE ANN FORWARD
MODEL OF PLANT
REPLACE PLANT BY ANN
MODEL
CONVENIENT FOR BP TRAINING)
TRACKS REFERENCE
MODEL RESPONSE
56
ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM (ANFIS)
(IMPLEMENTS FUZZY CONTROL WITH NNW)
DESIGN A FUZZY SYSTEM
WITH ANN – DESIRED
RESPONSE GIVEN
USE SUGENO METHOD
-ZERO OR FIRST ORDER
TRIANGULAR INPUT MFs
SINGLETON MFs
SUGENO FIS – ZERO ORDER
F= (W1f1+W2f2)/(W1+W2)
57
Rule 1
Rule 2
ANFIS TRAINING BY ZERO ORDER SUGENO METHOD
1. SUGENO FIS:
Rule 1: IF X is A1 AND Y IS B1 THEN Z=f1
Rule 2: IF X IS A2 AND Y IS B2 THEN Z=f2
OUTPUT F = (W1f1+W2f2)/(W1+W2) = W1/(W1+W2). F1 W2/(W1+W2).f2
2. ANFIS: (Five layers of computation)
Layer 1 output:
Layer 2 output:
Layer 3 output: Generate:
Layer 4 output: Generate:
Layer 5 output: Generate:
FIRST TRAIN f1 and f2 BY BP
THEN TRAIN a and b of TRIANGULAR MFs
1( ), 2( ) 1( ), 2( )
,
A X A X B Y B Y
Generateµ µ µ µ
1 1( ). 1( )
1 2( ). 2( )
A X B Y
A X B Y
Generate
W
W
µ µ
µ µ
=
=
1 1 1 2
2 2 1 2
/( )
/( )
W W W W
W W W W
= +
= +
1 1 1 1
2 2 2 2
.
.
W f W f
W f W f
=
=
1 1 2 2
F W f W f = +
58
INPUT
SIGNAL
OUTPUT
SIGNALS
3-PHASE
OUTPUT
WAVES
SYMBOLIC
ANN
NNW TRAINING EXAMPLE FOR 3-PHASE SINE WAVE GENERATION
59
Example 2:
60
61
WEIGHT MATRICES OF 1-5-3 NNW AFTER TRAIINING
62
SELECTED HARMONIC ELIMINATION (SHE) PWM IMPLEMENTATION BY A NNW
NOTCH ANGLE LOOK-UP TABLE
NNW
Example 3:
63
FEEDBACK SIGNAL ESTIMATION EQUATIONS FOR IM VECTOR DRIVE
Example 4:
64
....(1)
....(2)
( / ) ..(3)
( / ) ..(4)
s
ds
s
qs
s s s
dm ds ls ds
s s s
qm qs ls qs
s s
dr r m dm lr
s s
qr r m dm lr
L i
L i
L L L i
L L L i
ψ ψ
ψ ψ
ψ ψ
ψ ψ
= −
= −
= −
= −
2 2
ˆ ....(5)
cos .........(6)
ˆ
sin ..........(7)
ˆ
3
( )( )...(8)
2 2
r
s s
dr qr
s
dr
e
r
s
qr
e
r
s s s s
e dr qs qr ds
P
T i i
ψ ψ ψ
ψ
θ
ψ
ψ
θ
ψ
ψ ψ
= +
=
=
= −
65
NEURAL NETWORK BASED FEEDBACK SIGNAL ESTIMATION
OF DIRECT VECTOR DRIVE
66
Example 5:
67
fc = 10 kHz
fc = 1.0 kHz
Example 6:
VOLTS/HZ SPEED CONTROL WITH NNW BASED SVM OF 2-LEVEL INVERTER
Example 7:
68
PWM WAVES IN SECTOR- A WITHIN Ts
TURN-ON TIME EQUATIONS OF PHASE A
NEURAL NETWORK BASED SVM OF 2-LEVEL INVERTER
PHASE-A
PHASE-B
PHASE-C
69
TURN-ON TIME PLOT OF PHASE - A IN UM AND OM REGIONS
70
NNW TOPOLOGY FOR SVM OF A 2-LEVEL INVERTER
V*
f(V*) – V* RELATION IN UM AND OM REGIONS
71
3000V


ω
e
*
θ
e
*


THREE-LEVEL DIODE-CLAMPED INVERTER INDUCTION MOTOR DRIVE WITH ANN BASED
SVM
72
SVM
Example 8:
TURN-ON TIME PLOTS FOR U-PHASE FOR BOTH UNDERMODULATION AND
OVERMODULATION REGIONS
(a) TUP-ON FOR P-STATE, (b) TUN-ON FOR N-STATE
(a)
(b)
PWM WAVES
73
PRECOMPUTED CURVES
FOR NNW TRAININIG
CONVERTER
VOLTAGE VECTOR
COMMAND
LOGIC CIRCUIT
AND
TIMER
FEEDFORWARD NEURAL NETWORK BASED SVM OF 3-LEVEL CONVERTER
(2-9-9-9-6)
SWITSCHING STATES
OF CONVERTER
74

*
V
*
e
θ
network
neural
Artificial
ON UP
T
− 1 UP
T
2 UP
T
UP
P
COUNTER
UP/DOWN
ON UN
T
− UN
P
VP
P
ON VP
T

ON VN
T

ON WP
T

ON WN
T

VN
P
WP
P
WN
P
Limiter
A
B



FIVE-LAYER NNW WITH THE INTERFACE LOGIC AND UP/DOWN COUNTER
75
4
3
2
0
1
A
B
C
I
4
I
b
I
c
I
3
I
2
I
1
I
0
C
1
C
2
C
3
C
4
I
a
V
C1
V
C2
V
C3
V
C4
+
-
+
-
+
-
+
-
V
c0
V
b0
V
a0
S
xA
S
xB
S
xC
V
dc

SIMPLIFIED REPRESENTATION OF FIVE-LEVEL INVERTER
76
Example 9:
000
111 222
333
444
100
211
200
322
433
311 422
300
411
401
001
112 223
334
022
113 244
033
144
044
410
310
110 443 010
121 232
343 021
132 243
032
143
043
420
410
440 340 240 140
041
040
042
320
431
220
331 442 231 342
020
131 242
031
142
330
441
230
341
130
241
030
141
402
403
404
304
104 204 004
014
024
034
400
023
134
012
123
011
122 233
344
101
212 323
434 201
423
301
412
302
413
303
414
013
124
003
114
002
113 224
102 202
313 424
203
314
103
214
triangle 1
234
324 213
312
210
120
m = 0.53
A
B
C
D
F
E
321 432 421 221 332

(a)
2 5
4
1
3 6
7
8
10
9
11
12
13
14
15
16
t
a
t
c
t
b
t
b
t
a
t
c
t
a
t
c
t
b
t
b
t
a
t
b
t
c
t
c
t
a
m = 0.53

(b)
( . 1).16
r
hex
T SectorNo T = − + ….. (1)
(a) SWITCHING STATES OF FIVE-LEVEL INVERTER
(b) SECTOR-A TRIANGLES SHOWING THE SWITCHING STATES (t
a
, t
b
AND t
c
)
77














4 4 4 4
( ). ( ). ( ).
a c
A a A c A hex b A hex b hex
T K T t K T t K T t = + + ….. (1)

3 3 3 3
( ). ( ). ( ).
a c
A a A c hex b hex b hex A
T K T t K T t K A T t = + + ….. (2)

2
2 2 2
( ). ( ). ( ).
c A a c
A a A hex b hex b hex A
T K T t K T t K T t = + + ……. (3)

1
1 1 1
( ). ( ). ( ).
b a c A A a A c hex hex b hex
T K T t K T t K A T t = + + …… (4)


PULSE WIDTH SYNTHESIS FOR PHASE VOLTAGE A IN A SAMPLING INTERVAL T
s

Phase Direct Sequence Reverse Sequence
A 0 1 1 1 2 2 2 3 3 3 4 4 4 4 4 4 3 3 3 2 2 2 1 1 1 0
B 0 0 1 1 1 2 2 2 3 3 3 4 4 4 4 3 3 3 2 2 2 1 1 1 0 0
C 0 0 0 1 1 1 2 2 2 3 3 3 4 4 3 3 3 2 2 2 1 1 1 0 0 0
Duty cycles



Level 4
t
b
/
1
0

t
a
/
8

t
c
/
8

t
b
/
1
0

t
a
/
8

t
c
/
8

t
b
/
1
0

t
a
/
8

t
c
/
8

t
b
/
1
0

t
a
/
8

t
c
/
8

t
b
/
1
0

t
b
/
1
0

t
c
/
8

t
a
/
8

t
b
/
1
0

t
c
/
8

t
a
/
8

t
b
/
1
0

t
c
/
8

t
a
/
8

t
b
/
1
0

t
c
/
8

t
a
/
8

t
b
/
1
0

Level 3 T4A
Level 2 T
3A

Level 1 T
2A

Level 0 T
1A


TS
78
(for triangle 1 in previous figure)
79
COEFFICIENTS
INPUT
OUTPUT
5 . 9 4 5 . 9 5 5 . 9 6 5 . 9 7 5 . 9 8 5 . 9 9 6
- 3 0 0
- 2 2 5
- 1 5 0
- 7 5
0
7 5
1 5 0
2 2 5
3 0 0
T i m e S
V

(a)
5 . 9 4 5 . 9 5 5 . 9 6 5 . 9 7 5 . 9 8 5 . 9 9 6
- 8
- 6
- 4
- 2
0
2
4
6
8
T i m e S
A

(b)
0 2 0 4 0 6 0 8 0 1 0 0
0
2 0
4 0
6 0
8 0
1 0 0
1 2 0
1 4 0
1 6 0
1 8 0
2 0 0

(c)
SYSTEM PERFORMANCE AT MODULATION FACTOR
m′ ′′ ′ = 0.53 (31,8 Hz)






80
LINE VOLTAGE
WAVE
LINE CURRENT
WAVE
LINE VOLTAGE
SPECTRUM
(a) STRUCTURE OF REAL TIME RECURRENT NETWORK
(b) BLOCK DIAGRAM FOR DYNAMIC BACKPROPAGATION TRAINING
81
82
Stator Flux Estimation by Recurrent Neural Neural Network
Example 10:
83
HYBRID ANN TOPOLOGY FOR STATOR FLUX ESTIMATION (0.01 Hz – 200 Hz)
84
85
Example 11:
86
NNW BASED INVERSE DYNAMICS CONTROL OF A MULTIPLE DEGREE
OF FREEDOM ROBOTIC MANIPULATOR
Example 12:
.
..
F-1
F
87
CONCLUSION
• AI TECHNIQUES ARE EXPANDING THE FRONTIER OF POWER ELECTRONICS
- NEW CHALLENGE TO POWER ELECTRONICS ENGINEERS
• AMONG ALL THE AI TECHNIQUES, NEURAL NETWORKS WILL HAVE MAXIMUM IMPACT ON
POWER ELECTRONICS
• CURRENTLY MOST APPLICATIONS USE BACKPROPAGATION TYPE FEEDFORWARD NETWORK
• MANY OTHER FEEDFORWARD AND RECURRENT ANN TOPOLOGIES REQUIRE EXPLORATION
• NONAVAILABILITY OF LARGE ANN ASIC CHIP IS A PROBLEM – MOST APPLICATIONS NOW USE
FPGAs AND DSPs
• ANN CAN INTERPOLATE TRAINING DATA – EXTRAPOLATION IS VERY LIMITED
(SIMILAR TO HUMAN BEING)
• ADAPTIVE ANNs NEED FAST ONLINE TRAINING
• POWERFUL INTELLIGENT CONTROL AND ESTIMATION TECHNIQUES CAN BE DEVELOPED
USING HYBRID AI (NEURO-FUZZY, NEURO-GENETIC, NEURO-FUZZY-GENETIC, FUZZY-GENETIC)
• EXPECTED TO HAVE WIDESPREAD APPLICATIONS IN POWER ELECTRONICS AND MOTOR
DRIVES IN FUTURE
NEURAL NETWORK DEMO PROGRAM
FOR
3- DIMENSIONAL SINE LOOK-UP TABLE
(USE MATLAB/NEURAL NETWORK TOOLBOX)
1. BASIC EQUATIONS: INPUT 0<X<2pi
OUTPUT Y1 = sinX
Y2 = sin(X-2pi/3)
Y3= sin(X+2pi/3)
2. CREATE TRAINING DATA IN MATLAB WORKSPACE
>> X=(0:0.1:6.3); Increment of 0.1 rad
>>Y=[sin(X);sin(X-2*pi/3);sin(X+2*pi/3)]; Target data matrix
>>plot(X,Y) Checks X-Y plot of training data
Fig. 1. 3-PHASE SINE WAVES
88
3. ACCESS ANN GRAPHICAL USER INTERFACE (GUI)
>>nntool
4. SELECT THE NETWORK AND ASSIGN PARAMETERS
* Import X as Input
* Import Y as Target
* Click “New Network”
Select: Network name –“SINET”
Network Type – “Feed-forward backprop”
Input ranges – Get from input X
Training function – TRAINLM (Backpropagation LM
algorithm)
Adaption learning function- LEARNGDM (Gradient descent
with momentum)
Performance function – SSE (Sum squared error)
Number of layers – 2 (3 layers with input layer)
Properties for –Layer 1 (hidden layer) Number of neurons – 5
Transfer function – TRANSIG (hyperbolic tan)
Layer 2 (output layer) No. of neurons -3
Transfer function – PURELIN
(bipolar linear)
*Click – Create
(Page -3-23)
89
X
Y
SINET_outputs
SINET_errors
SINET
GUI OF NNW TOOLBOX
90
5. VIEW THE CREATED NETWORK TOPOLOGY
Click –View
6. TRAIN THE NETWORK WITH ACCEPTABLE SSE
Highlight SINET
Click – Train
Training data – select X as Input and Y as Target
Check Training parameters – select epoch -100
Initialize Weights
Training data: Input –X
Target – Y
Click – Train
Fig. 3. VIEW OF NEW NETWORK
91
* Iterate Training with different epochs and hidden layers
until the actual SSE is near the Goal SSE
*Check Weight vectors
7. SIMULATE THE NETWORK TO VERIFY ITS PERFORMANCE
*Export – all parameters
>> X (shows all the values of X)
>> Y (shows all the values of Y)
>> SINET_outputs (shows the last value of Y)
>> SINET_errors (shows all the errors)
To Simulate at X = 3.142:
Enter New Data: data1=3.142
Simulate
Select input as data1
Simulate
Export - all
>> data1 3.142
>>SINET_outputs Check value
Change X and check new values of SINET outputs
92
TRAINING ERROR(SSE) WITH EPOCHS
93
THANK YOU VERY MUCH

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close