Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Model Order Reduction by Mixedmathematical Methods: Sumit Mondal, Pratibha Tripathi

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

International Journal of Computational Engineering Research||Vol, 03||Issue, 5||

www.ijceronline.com ||May ||2013|| Page 90


Model Order Reduction By MixedMathematical Methods

1,
Sumit Mondal ,
2,
Pratibha Tripathi
1,
M. Tech,Control & Instrumentation Branch,Electrical & Electronics Engineering Dept.
Sam Higginbottom Institute of Agriculture, Technology & Sciences
Allahabad, India
2,
Assistant Professor,Control & Instrumentation Branch,Electrical & Electronics Engineering Dept.
Sam Higginbottom Institute of Agriculture, Technology & Sciences
Allahabad, India








I. INTRODUCTION
The approximation of high-order complex systems to low-order simple models has been the concern of
many researchers during the last two decades. The low-order simple model can replace the original one to
facilitate the design and online implementation of the feedback controllers and digital filters. The ease with
which the analysis can be performed depends mainly upon the complexity of the model. Simple lower order
models are very desirable in the design of control systems, as the analysis of higher order systems is usually
cumbersome and mathematically intensive. The purpose of obtaining reduced model is to provide a good
approximant for the higher order systems, maintaining the characteristics of the latter as much as possible.
Nowadays, systems have become complex and the interrelationship of many controlled variables need to be
considered in the design process.
Various model reduction methods have been proposed. Two approaches have attracted much attention
]b ]g ]Y b fY kcf sYb ]cb]b HYgY fY ccfYg VbWY fY]n ]cb O b cptimal Hankel norm
approximation [2]. During the past three decades, several more model order reduction techniques have been
developed [6-9]. Each of them has their own merits and applications. In recent times, mixed mathematical
techniques [10-11] have gained much importance in model order reduction of large scale Single-Input Single-
Output (SISO) systems. Extensions of these SISO techniques for MIMO systems are not trivial. Shamash [3]
has proposed a multivariable system reduction using Pad Approximation and dominant eigen values. This
method assumes that the dominant pole of the higher order system are known, and hence suffer from the
drawback of its applicability to systems with no dominant poles, or where the dominant poles are difficult to
identify. Pal [4] has developed a system reduction methodology using the continued fraction approach and
Routh Hurwitz Array, in which the initial transient response of the reduced order model might not match with
the higher order system, as only the first few time moments are considered depending upon the order of the
reduced model. The viability and limitations of similar methods have been discussed by Shamash [5].
Recently, mixed methods are getting greater attention in the model order reduction of MIMO systems.
In these methods, the common denominator of the transfer function of the reduced order model is fixed by using
a stability preserving algebraic method, while the numerators are obtained using one of the available model
reduction methods.

II. PROPOSED METHOD
In this proposed method the advantages of mixed method of model order reduction is used for Single-
Input Single-Output (SISO), where the numerator is reduced by Pad Approximation [17] and second order
reduced denominator is derived by utilizing the basic characteristics [19] of higher order original system. These
WfW Yf]g ]Wg fY rbsdY b rf fY rYbW c cgW] ]cbg
n
sd]b f ]c gY ]b ]sY H
s
), peak
ABSTRACT
In this paper, a mixed method mathematical technique is proposed for finding reduced order model of
Single-Input Single-Output (SISO) system. The model reduction of large scale systems is achieved by
using Pad approximation and utilizing the characteristics of the higher order system which is
maintained in the reduced model. The denominator polynomial of the reduced model is determined by
the characteristic equation which retains the basic characteristics of the original system, and the
coefficients of the numerator polynomial are obtained by using the Pad approximation technique.

I ndex Terms: Integral Square Error, Pad Approximation, Single-Input Single-Output

Model Order Reduction By Mixed...
www.ijceronline.com ||May ||2013|| Page 91
amplitude (P
a
) and peak time (T
p
). the reduced second order approximant maintains the characteristics of the
original system. This method tries to minimize the Integral Square Error (ISE) and the reduced second order
model behaviour to the input signal almost matches with the original system behaviour.

Pad Approximation:
An asymptotic expansion or a Taylor expansion can often be accelerated quite dramatically by being re-arranged
into a ratio of two such expansions.

A Pad approximation

Gs =
a
k
s
k M
k=0
b
k
s
k N
k=0
(i)

(normalized by b
0
= 1) generalizes the Taylor expansion with the same total number of coecients:
T
M+N
(s) = c
k
s
k M+N
k=0
(ii)

(the two expansions being the same in the special case of M = 0). From a truncated Taylor expansion (ii), one
determines the corresponding Pad coecients by requiring that if (i) is Taylor expanded; the result shall match
all the terms given in (ii).
Such that:
Gs =
a
k
s
k M
k=0
b
k
s
k N
k=0
= c
0
+ c
1
s + c
2
s
2


The coefficients of the power series can be calculated as:

c
0
= a
0

c
k
= 1/b
0
[a
k
- b
j
c
kj
k
j=1
]; k > 0
a
k
= 0 k > n-1

Hence, Pad Approximants can be given as:

a
0
= b
0
c
0

a
1
= b
0
c
1
+ b
1
c
0

.
.
. a
k-1
= b
0
c
k-1
+ b
1
c
k-2
V
k-1
c
1
+ b
k
c
0


HY fYrWY scYg fbgYf rbW ]cb Wb VY cfsY by using Pad Approximants in the numerator and the
denominator is derived by the basic characteristics of higher order systems as:

Gs =
a
0
+ a
1
s
s
2
+2
n
s +
n
2


III. NUMERICAL EXAMPLE

Consider a fourth order system:

Gs =
s
3
+7s
2
+24s + 24
s
4
+10s
3
+ 35s
2
+50s +24

Characteristics:
Rise Time: 2.2609
Settling Time: 3.9312
Settling Min: 0.9062
Settling Max: 1.0000
Overshoot: 0
Undershoot: 0
Peak: 1.0000
Peak Time: 10.4907

Model Order Reduction By Mixed...
www.ijceronline.com ||May ||2013|| Page 92
The unit step response can be shown as:

Reducing the denominator by utilising the characteristics of the system, like damping ratio rbsdY
b rf fY rYbW c cgW] ]cbg
n
) etc.

.cf b dYf]c]W cf scg dYf]c]W g g Ys l brsVYf c cgW] ]cbg VYcfY Y g g Ys gY Yg l 1

Since,
n
l (H
s
Therefore,
n
= 4/ (0.99*3.93) = 1.0281

Reduced denominator:

D
2
(s) = s
2

n
g
n
2
Therefore, D
2
(s) = s
2
+ 2.0356s + 1.0569

Now, reducing numerator by Pad Approximation:

b
0
= 1.0569
b
1
= 2.0356
c
0
= 1
c
1
= - 1.08333
Now, reduced model representation:

Gs =
a
0
+ a
1
s
s
2
+2.0356s + 1.0569


a
0
= b
0
c
0
= 1.0569
a
1
= b
0
c
1
+ b
1
c
0
= 0.891

Hence, final reduced model:

Gs =
1.0569 + 0.891s
s
2
+2.0356s + 1.0569


Its characteristics can be given as:
Rise Time: 2.3589
Settling Time: 4.1169
Settling Min: 0.9021
Settling Max: 1.0000
Overshoot: 0
Undershoot: 0
Peak: 1.0000
Peak Time: 10.3072



0 1 2 3 4 5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Step Response
Time (sec)
A
m
p
l i t
u
d
e
Model Order Reduction By Mixed...
www.ijceronline.com ||May ||2013|| Page 93
The unit time step response can be shown as:


Integral Square Error, ISE = 5.1018x10
-4


IV. CONCLUSION
In this work a mixed mathematical technique has been proposed for model reduction. It can be clearly
seen from the step response that the reduced model is in close approximant with the original model and tries to
minimize the error. If the original model is stable, the reduced model must also be stable. The proposed
technique is evaluated for Linear Time Invariant Single-Input Single-Output (SISO) system, and the result
proves to be better than the past proposed methods. This method utilises the basic characteristics of the higher
order system and forms second order approximant that resembles the second order characteristics equation. The
method is computationally simple and reduced model stability is assured if the original system is stable.

REFERENCES
[1] ccfY Df]bW]d csdcbYb b g]g ]b ]bYf C g Ysg. cb fcV]] CVgYf|V]] b cY FYrW ]cb IEEE
Transactions, Automatic Control, January 1981, vol. 20, issue 1, pp. 17-31.
[2] c|Yf Cd ]s bY-Bcfs ddfc ]s ]cb c ]bYf r ]|f]VY C g Ysg International Journal of Control,1984,
vol. 39, issue 6, pp. 1115-1193.
[3] Csg r ]|f]VY C g Ys |] c Y cg b DY ddfc ]s ]cb r cs ]W cb fc IEEE Transactions,
Automatic Control, 1975, vol. 20, page: 815-817.
[4] D C g YsFYrW ]cb V ] Y Y c IEEE Transactions, Automatic Control, 1980, vol. 25, issue 5, pp. 973-976.
[5] Csg HY J]V]] c b ]W Y cg cf HY FYrW ]cb c r ]|f]VY C g Ysg IEEE Proceedings, 1981, vol.
69, issue 9, pp. 1163-1164.
[6] HB rWg cb ]brY-fraction cf] s cf ]gY cY FYrW ]cb 9YW fcb]W Y Yfg , |c ]ggrY dd(((445.
[7] C fg HY 8Yg]b c FYrWY CfYf C g Ysg International Journal of Control, 1980, vol. 31, issue 4, pp.677690.
[8] kb Cb rYf ]f Wcb ]brY fW ]cb Y dbg]cb sY c cf Y g]sd]]W ]cb c fY g g Ys bs]Wg International
Journal of Control, 1983, vol. 37, issue 3, pp.599614.
[9] DC r sb . bbYfY D cbYf cb f]Vr ]cbg c Y scY fYrW ]cb dfcVYs IEEE Transactions, Automatic
Control, 1982, vol. 27, issue 2, pp.454455.
[10] Dfsf C rYfYY F Dfg C g Ys FYrW ]cb rg]b .W cf 8]|]g]cb cf] s b 9]Yb CdYW frs b g]g
International Journal of Applied Mathematics Modelling, 2007, vol. 31, pp. 2542-2552.
[11] T. Yb b b K b cY FYrW ]cb rg]b C V]] 9 r ]cb Y c b Y cb ]brY .fW ]cb Y c
International Journal of Control, 1980, vol. 32, issue 1, pp. 81-94.
[12] F Dfg C D Cfs b ] sdfc|Y DY ddfc ]snts for multivariable systems using stability equation
sY c J. Inst. Eng. India IE(I), 2003, vol. 84, pp. 161-165.
[13] F Dfg C D Cfs b ] ]bYf scY fYrW ]cb rg]b Y |b Yg c ]]c| Wf] Yf]cb b W cf ]|]g]cb
J. Inst. Eng. India IE(I), 2003, vol. 84, pp. 7-10.
[14] F Dfg D Db r ]|f]VY g g Ys fYrW ]cb rg]b scY sY cg b DY dY ddfc ]s ]cb J. Inst. Eng.
India pt EL, 1998, vol. 79, pp. 84-90.
[15] F Dfg DY dY scY cfYf fYrW ]cb cf sr ]|f]VY g g Ys rg]b Fcr ddfc ]s ]cb Computers and Electrical
Engineering, 2000, vol. 26, pp. 445-459.

You might also like