Documente Academic
Documente Profesional
Documente Cultură
I. INTRODUCTION
Increased use of non-linear loads and the proliferation of
power electronic based equipment have led to many problems
in the power distribution network. Solid state switching
devices such as thyristors and other switches are used in
Uninterruptible Power Supplies (UPSs), variable speed drives
etc. Because of their inherent non-linearity, they draw reactive
power and harmonic current components from the utility. This
could lead to low system efficiency and poor power factor,
apart from harmonic pollution of the system. Many power
quality problems usually occur in the distribution system [1].
Harmonics and non-linear loads from a single customer can
affect the quality of power received by other customers. Low
power factor loads can cause power losses at the distribution
feeder. The utility does not have any control over the loads.
The conventional approach to this solution has been the use of
passive power filters. Though they are a low-cost solution,
there are disadvantages to using them, the main risk being that
of resonance with the system impedance. They are also bulky
and overcompensation can lead to a lower power factor.
Active power filters can overcome these problems. The
Harish Suryanarayana is a graduate student from the Indian Institute of
Technology-Madras, Chennai, INDIA. (E-mail : harish.sury@gmail.com)
Dr. Mahesh Kumar is faculty at the Department of Electrical Engineering IIT
Madras. (E-mail: maheshk@ieee.org)
v + ( vsb vsc )
*
i fa = ila isa = ila sa
Plavg + Ploss
v + ( vsc vsa )
*
i fb = ilb isb = ilb sb
Plavg + Ploss
v + ( vsa vsb )
*
i fc = ilc isc = ilc sc
Plavg + Ploss
where
j = a ,b ,c
(1)
2
vsj and tan( ) 3
figure. The DSTATCOM has 12 IGBT switches with antiparallel diodes(not shown in the figure for the sake of clarity),
a dc storage capacitor, three isolation transformers and three
interface inductors. In modeling the system, the inductance
and resistance of the isolation transformers can be included in
Lf and Rf, the interface inductance and resistance. Since we
assume the source voltages to be balanced, sinusoidal and
stiff, Ls and Rs, the source impedances, are assumed to be
negligible. Hysteresis band current control has been used for
tracking the reference currents. As seen in Fig. 1, the
DSTATCOM injects the reference filter currents generated
using the symmetrical component theory at the Point of
Common Coupling (PCC). For good current injection and thus
compensation, maintaining the dc capacitor value close to the
reference value is very important.
Ls
Rs
isa
PCC
isb
ilb
vsc
isc
ilc
Rf
Lf
N
A phas e
H-Bridge
S1a
Za
Zb
Zc
Unbalanced
load
(2)
Ploss depends
DC Link
S3a
S1b
S3b S1c
S3c
if
Cdc
S4a
ila
v sb
S2a
S4b
S2b S4c
i Cd c
L nl
S2c
n'
vdc
i0
R nl
Nonlinear load
Fig. 1. A Three phase, four wire compensated system using H-bridge VSI
based DSTATCOM.
B. Fuzzification
The fuzzification interface modifies the inputs to a form in
which they can be used by the inference mechanism. It takes
in the crisp input signals and assigns a membership value to
the membership function under whose range the input signal
falls. Typical input membership functions are triangular
trapezoidal or exponential. Seven triangular membership
functions have been chosen: NL (Negative Large), NM
(Negative Medium), NS (Negative Small), Z (Zero), PS
(Positive Small), PM (Positive Medium) and PL (Positive
Large) for both error (err) and change in error (derr ) . The
input membership functions are shown in Fig. 3. The tuning of
the input membership function is done based on the
requirement of the process. Each membership function has a
membership value belonging to [0 1]. It can be observed that
for any value of error or change in error, either one or two
membership functions will be active for each.
NL
-15
N
M
-10
NS
PS PM
-5
10
PS
PM
10
PL
15
err(t) in Volts
(a)
Input
Fuzzification
Inference
Mechanism
Defuzzification
Output
Rule Base
NL
NM
NS
PL
(3)
(4)
In (3) and (4) above, e(i) is the error and e (i ) is the change in
error in the ith iteration. vdcref is the reference dc link voltage
and vdc (i ) is the dc link voltage in the ith iteration. The outputs
of the fuzzy supervisor are chosen as the change in Kp value
and the change in Ki value.
K p = K p ref + Kp
K i = K i ref + Ki
(5)
(6)
K p ref and K i ref are the steady state values determined by the
-15
-10
-5
15
derr(t)in Volts
(b)
Fig. 3(a) Membership functions for error input. (b) Membership functions for
change in error input
C. Inference Mechanism
The two main functions of the inference mechanism are:
a)
Based on the active membership functions in error
and the change in error inputs, the rules which apply
for the current situation are determined.
b)
Once the rules which are on are determined, the
certainty of the control action is ascertained from the
membership values. This is known as premise
quantification.
Thus at the end of this process, we shall have a set of rules
each with a certain certainty of being valid. The database
containing these rules is present in the rule base from which
the control action is obtained. The rule base will be discussed
in the next section. An example of a rule is given in (7). The
terms PL and PM are the membership functions for error and
for change in error respectively.
IF
"error"
is
"change in error" is
THEN
"Kp"
is
"Ki"
is
PL (positive large)
PM (positive medium)
NM
NS
PS
PM
PL
(7)
L ( Large Kp )
SKi ( Small Ki )
(a)
derr
NL
NM
NS
PS
PM
PL
NL
SKi
SKi
SKi
NM
SKi
SKi
SKi
NS
LKi
LKi
LKi
LKi
LKi
LKi
LKi
LKi
LKi
PS
LKi
LKi
LKi
PM
SKi
SKi
SKi
PL
SKi
SKi
SKi
err
Volts
NL
err < 0
NM
NS
Z
time(s)
PS
PM
PL
(b)
err > 0
derr>0
derr<0
derr>0
derr<0
error. The rules in the rule base are designed based on which
part of the graph the waveform is in. The important points
involved in the design of the rule base are the following:
a)
If the error is large and the change in error shows the
dc link waveform deviating away from the reference,
then increase Kp.
b)
If the waveform is approaching the reference value,
then increase the Ki value to reduce overshoot and
improve settling time.
Keeping these aspects in mind, two rule base matrices have
been developed for Kp and Ki. Fig. 5(a) gives the rule base
matrix for Kp and Fig. 5(b) gives the rule base matrix for Ki.
The output membership functions for the proportional gain are
LKi, SKi and Z and the output membership functions for
integral gain are L, M, S and Z. These matrices provide rules
such as the example seen in (7) for all possible combinations
of the membership functions for error and change in error.
Thus, using information from the rule base, the rule and its
certainty is determined by the inference mechanism. The
method to convert the fuzzy result to crisp control action is
called defuzzification. This is explained in the next section.
derr
err
NL
NL
NM
NS
PS
PM
PL
Fig. 5(a) Rule base matrix for change in Kp. (b) Rule base matrix for change
in Kp
E. Defuzzification
The inference mechanism provides us with a set of rules each
with a premise. The defuzzification mechanism considers these
rules and their respective premise values, combines their effect
and comes up with a crisp, numerical output.Thus, the fuzzy
control action is transformed to a non fuzzy control action.
The center of gravity method has been used in this work for
this. If we use this method, the resultant crisp output is
sensitive to all of the active fuzzy outputs of the inference
mechanism. Fig. 6(a) and Fig. 6(b) show the output
membership functions chosen for Kp and Ki.
premise
SKi
50
(a)
LKi
100
Ki
premise
50
100 150
Kp
(b)
Fig. 6(a) Output membership function for Kp (b) Output membership function
Ki.
phas e-a
Non-linear load
-400
Rla = 50 , Lla = 20 mH
Rlb = 35 , Llb = 40 mH
Rla = 70 , Lla = 20 mH
Three-phase full wave rectifier drawing
a dc current of 5 A
Interface inductor
Lf
V dc ref = 500 V
= 20 mH,
0 .0 1
0 .0 2
0 .0 3
0 .0 4
0 .0 5
0 .0 6
Time in s e c o n d s
0 .0 7
0 .0 8
0 .0 9
0 .1
(a)
No n lin e a r Unb a la nc ed L o a d Cu r r e n ts
15
Rf = 5
Current in Amperes
220V (phase-rms), 50 Hz
C dc = 2200 F
-100
phas e-a
phas e-b
phas e-c
10
Values
DC capacitor
-300
TABLE I
SIMULATION PARAMETERS
Unbalanced load
100
-200
Supply voltage
phas e-c
200
System Parameters
phas e-b
300
Voltage in volts
B a la n c e d S in u s o id a l So u r c e V o lta g e s
400
-5
-10
-15
0
0 .0 1
0 .0 2
0 .0 3
0 .0 4
0 .0 5
0 .0 6
Time in s e c o n d s
0 .0 7
0 .0 8
0 .0 9
0 .1
(b)
Fig. 7(a). Balanced sinusoidal source voltages (b) Nonlinear unbalanced load
currents.
Re f e r en c e a n d A c tu a l Filte r Cu r r e n ts in Ph a s e A
5
Referenc e c ompens ator c urrent
Hysteresis band
h = 0.6 A
K p = 40, K i = 20
Current in Amperes
K i = 55
3
2
1
0
-1
-2
-3
-4
0 .1
0 .10 2
0 .1 0 4
0 .1 0 6
0 .1 0 8
0 .11
0 .1 1 2
Time in s e c o n d s
0 .11 4
0 .1 1 6
0 .1 1 8
0 .1 2
(a)
A c tu a l S o u r c e Cu r r e n t W a v e f o r ms
25
phas e-a
phas e-b
phas e-c
20
15
10
Current in Amperes
K p = 110,
ac tual c urrent
5
0
-5
-10
-15
-20
-25
0 .0 1
0 .0 1 5
0 .0 2
0 .0 2 5
0 .0 3
0 .0 3 5
Time in s e c o n d s
0 .0 4
0 .0 4 5
0 .0 5
(b)
Fig. 8(a). Phase-a reference and actual compensator currents (b) Balanced
and sinusoidal source currents.
[5]
[6]
[7]
[9]
Fuzzy supervised PI
Normally tuned PI
[10]
520
DC Link Voltage in
Volts
[4]
[8]
[3]
510
[11]
500
[12]
490
480
[13]
470
460
0
0.1
0.2
0.3
Time in
seconds
0.4
0.5
0.6
[14]
[15]
[16]
[17]
VI. CONCLUSIONS
A fuzzy logic supervisor to the dc link PI controller of the
DSTATCOM has been proposed. The supervisor varies the
gain of the PI controller during the transient period in a way
that improves performance. The system has been modeled and
simulated in the MATLAB technical environment. The
performance of the dc link and compensation were observed
with and without the fuzzy supervisor. Simulation results
show a 50% reduction in voltage deviation of the dc link
voltage with faster settling time. Instantaneous symmetrical
component theory has been used for load compensation. Good
compensation has been observed as source current THDs for
each phase is 1.63%, 1.77% and 1.58% while the load THDs
are 12.37%, 10.5% and 14.54% respectively. Thus, through
simulation studies, the implementation of a fuzzy supervisor
for DC link voltage control in a DSTATCOM using
instantaneous symmetrical component theory for load
compensation has been demonstrated.
VII. REFERENCES
[1]
[2]
VIII. BIOGRAPHIES
Harish Suryanarayana is a graduate student in the
department of Electrical Engineering at the Indian
Institute of Technology Madras, Chennai, India.
His reseach interests are Power Electronics, Power
Quality and Drives.