You are on page 1of 32

1

Evolutionary Algorithm Assisted Reduced Order


Modelling

Keshav Sarraf
(10EE35020)
Under the guidance of
Professor J Pal

Department of Electrical Engineering


Indian Institute of Technology, Kharagpur
West Bengal, India

Declaration
I certify that

The work contained in this report is original and has been done by me under the
guidance of my supervisor(s).
The work has not been submitted to any other Institute for any degree or diploma.
I have followed the guidelines provided by the Institute in preparing the report.
I have conformed to the norms and guidelines given in the Ethical Code of Conduct of
the Institute.
Whenever I have used materials (data, theoretical analysis, figures, and text) from
other sources, I have given due credit to them by citing them in the text of the report
and giving their details in the references. Further, I have taken permission from the
copyright owners of the sources, whenever necessary.

.
Keshav Sarraf
10EE35020

CERTIFICATE
This is to certify that the report entitled, Evolutionary Algorithm Assisted Reduced Order
Modelling submitted to the Indian Institute of Technology Kharagpur, India, for the award of
the degree of Bachelor of Technology by Mr. Keshav Sarraf is a record of bonafide research
work carried out by him under my supervision and guidance. The thesis has reached the
standard fulfilling the requirements of the regulations related to the degree. The results
embodied in the thesis have not been submitted to any other University or Institute for the
award of any degree or diploma.

Dated:

Professor J Pal
Department of Electrical Engineering
Indian Institute of Technology

Acknowledgement
It is a matter of great privilege for me to be able to express my deep sense of gratitude to my
project supervisor Professor J Pal for his unwavering support, insightful suggestions,
encouragement and the fruitful discussions throughout the duration of my project. He has
always made himself available for any sort of discussion and support. I have learned a great
deal in the areas of Evolutionary Algorithm based optimisation techniques and Reduced Order
Modelling over the past year through his uninhibited help and motivation.
I am also grateful to all faculty members of the Department for their help, suggestions and
comments during the presentations and throughout the tenure of the work.
Finally, I would like to express my thanks and love to my parents for their support and
encouragement.

Keshav Sarraf
Department of Electrical Engineering
Indian Institute of Technology, Kharagpur

Contents
Abstract: .................................................................................................................................................. 6
Objectives: .............................................................................................................................................. 6
Genetic Algorithms: ................................................................................................................................ 7
Overview of optimization using GA......................................................................................................... 7
Reduced Order Modelling of SISO System: ............................................................................................. 8
Results: .................................................................................................................................................. 10
Reduced Order Modelling of SISO Systems (Contd.): ........................................................................... 12
Results: .................................................................................................................................................. 14
Parallel Processing in Genetic Algorithms:............................................................................................ 16
Extension to MIMO systems: ................................................................................................................ 16
Results: .................................................................................................................................................. 19
Teaching Learning Based Optimisation ................................................................................................. 26
Introduction .......................................................................................................................................... 26
Algorithm: ............................................................................................................................................. 26
Implementation of TLBO in ROM: ......................................................................................................... 27
Results (ROM of SISO systems): ............................................................................................................ 28
Results (ROM of MIMO): ....................................................................................................................... 29
Conclusions: .......................................................................................................................................... 31
Future Work .......................................................................................................................................... 31
References............................................................................................................................................. 32

Abstract:
Industrial processes are very complex in nature. Accurate mathematical analysis of these
processes yield models which are of very high order. With todays advanced computation
technology, we are capable of handling fairly complex systems but there is always a need of
reduced order modelling to shrink the amount of time and storage required for studying and
simulating these complex processes without compromising the reliability of the results. This
project aims at developing a technique to reduce the order of complex plants using
Evolutionary Algorithms like GA (Genetic Algorithms) and TLBO (Teaching Learning Based
Optimization).

Objectives:
Objectives of this project are as follows:

To develop a technique for order reduction of high-order SISO linear systems using
Genetic Algorithms.
To extend the developed technique to MIMO systems.
To check feasibility of TLBO for model order reduction

Genetic Algorithms:
Genetic Algorithms is an optimization technique that involves probabilistic global search
methods which emulate the process of natural evolution. Genetic Algorithms have been found
to be capable of delivering good performance in complicated domains like optimization of
multi-variable systems. Thus it is expected that using Genetic Algorithms for order reduction
of systems should be practically feasible and should result in optimum solutions which match
the desired specifications.

Overview of optimization using GA


While using GA for optimization, the following steps are needed to be followed.

Parameters to be optimized are chosen.


These parameters are then encoded (Encoded parameters are called Genes)
Genes are combined to form a Chromosome (A chromosome represents an
individual person in a population).
Initial population of individual solutions is generated.
Fitness of each individual in the population is evaluated.
Based on the fitness, parents are selected to determine the next generation.
Next generation of the population is then populated.
Again fitness is evaluated for each individual and the cycle goes on till the output
saturates or the number of iterations are completed.

Initial Poputation

Evaluate fitness of each individual

Select parents based on fitness

Create new population

Reduced Order Modelling of SISO System:


The preliminary approach involved direct determination of the reduced order model by optimizing
the value of coefficients of the system transfer function using GA.
For exampleIf the original system is represented by:
s5 + 1014 s4 + 14069 s3 + 69140 s2 + 140100 s + 100000
-----------------------------------------------------------------------s6 + 222 s5 + 14541 s4 + 248420 s3 + 1.454e06 s2 + 2.22e06 s + 1000000

G(s) =

Transfer Function 1

Then a 3rd order reduced system is taken as:


(a2) s2 + (a1) s + (b0)/10
------------------------------s3 + (b2) s2 + (b1) s + (b0)

R(s) =

Transfer Function 2

Where a2, a1, b2, b1 and b0 represent the parameters that have to be optimised.

1. Creation of initial Population:


o

Parameter Selection: Coefficients of the reduced order transfer function are chosen
as parameters. To achieve similar steady state gain as the original system, the constant
terms in the numerator and denominator are taken in the same ratio as the original
transfer function (can be observed in Transfer Function 1 and Transfer Function 2).
Encoding: All the parameters (genes) are then encoded into binary strings of equal
length. For each solution, these binary strings (genes) are concatenated to form a
chromosome. Each chromosome contains all the parameters of a single solution.
Population Generation: A population of solutions (chromosomes) is created keeping
in consideration that each parameter should lie within its viable range. This range is
fixed initially and can be varied from o.

2. Fitness evaluation of each individual:


Fitness of each individual is evaluated by calculating weighted sum of squared error at
various closely spaced instances of the step response. Error at a time instance is
calculated by taking the difference of the step response of the original system and step
response of the reduced system at that time instant.
= 1

= 2

= 1 (() ()) + 2 (() ())2


=0

= 1

Lower the value of J, better is the fitness of the solution. The value of the variables w1,
w2, t1 and t2 are chosen by hit and trial to achieve best matching reduced order
model. To reduce complexity, sum of the weights is taken to be 1.

3. Parent Selection:

o
o
o
o

Parent selection for mating pool is done by using a combination of rank selection and
roulette wheel selection, according to which:
Population is sorted based on the decreasing value of fitness.
Weights are allocated linearly such that the worst solution gets the weight 1 while the
best solution gets the weight equal to the population size.
Based on these weights, an area is allocated to each solution on the roulette wheel.
When the roulette wheel is spun, a solution having better fitness has a greater
probability to be picked.

Chart1
1, 10%
2, 20%

If we have a population of 4 chromosomes, then the roulette


wheel would be divided like Chart 1.

4, 40%

Then a roulette wheel is The


spunbest
on this
pie chart
to randomly
Better the
solution
has been
given a select
weighta4parent.
and it occupies
3, 30% fitness of an individual, more
is the probability
that
roulette
wouldbefore,
select the
that
40% (4/(1+2+3+4))of
the
area
and aswheel
mentioned
worst solution has been given a weight 1 and it occupies
Best individual.
Second Best
o
The
next
parent
is
chosen
similarly,
is made sure that both the parents are not
only
10% ofbut
the itarea.
Thrid Best
Worst
same.

4. Creation of new population:


o

Crossover: Genetic information of the child is created by copying some part of the
chromosomal sequence from one parent and the remaining part from the other
parent. The boundary for crossover is chosen at random for each child.
For example Parent1
= 01010101
Parent2
= 00001111
Boundary
=5
Child
= 01010111
It is made sure that each solution thus obtained from crossover follows the property
that every parameter of the solution lies within its own range that was fixed while
creating the initial population. If any of the parameters is not within its own range,
then that child is rejected.
Elitism: Few best solutions from each generation are carried over to the next
generation without any change. This is done to preserve the best solution throughout
the iteration.
Mutation: In every new generation, few randomly selected solutions are mutated i.e.
their genetic information is altered randomly. This can result in both constructive and
destructive solutions and thus it has to be carried out only on a small set of individuals.
If all the solutions are mutated then our optimisation process will turn into a random
search.

Benchmarks for result testing:


Matlab has inbuilt model reduction functions which are fairly good. 1 such function was used to
compare the results obtained using Genetic Algorithms.

DC Matching model reduction- modred() with balanced reduction and forced DC matching.
This reduction technique produces a reduced order model that matches the DC gain of the
original system.

10

Results:

Figure 1: step response and bode plot of original system and reduced system

s5 + 1014 s4 + 14069 s3 + 69140 s2 + 140100 s + 100000


-----------------------------------------------------------------------6
s + 222 s5 + 14541 s4 + 248420 s3 + 14541006 s2 + 2.22e06 s + 1000000
Transfer Function 3 G(s)

Population Size: 100


Bit Length per Variable: 16
Number of Variables: 5
Cross Over: random, Mutation: 10%
GA parameters used to reduce transfer function 3

10.38 s2 + 14.12 s + 59.69


------------------------------s3 + 131.7 s2 + 619.2 s + 596.9
Transfer Function 4: results obtained from GA

11

Figure 2: step response and bode plot of original system and reduced system

-7.56 s^12 - 495.3 s^11 - 1.404e04 s^10 - 2.244e05 s^9 - 2.185e06 s^8 - 1.288e07 s^7 - 4.09e07 s^6 - 2.274e07 s^5 + 3.401e08 s^4 + 1.347e09 s^3 +
2.249e09 s^2 + 1.729e09 s + 4.682e08
----------------------------------------------------------------------------------------------------------------s^13 + 66.9 s^12 + 2079 s^11 + 4.246e04 s^10 + 6.087e05 s^9 + 6.152e06 s^8 + 4.373e07 s^7 + 2.184e08 s^6 + 7.613e08 s^5 + 1.824e09 s^4 +
2.937e09 s^3 + 3.053e09 s^2 + 1.872e09 s + 5.12e08
Transfer Function 5 G(s)

-4.148s3 + 13.01s2 + 20.2s + 6.538


-------------------------------------------------s4 + 10.17s3 + 21.24s2 + 21.54s + 7.148
Transfer Function 6: results obtained from GA

Population Size: 100


Bit Length per Variable: 16
Number of Variables: 7
Cross Over: random, Mutation: 10%
GA parameters used to reduce transfer function 5

12

Reduced Order Modelling of SISO Systems (Contd.):


It was observed that the reduced model obtained from optimization of transfer function coefficients
did not perform very satisfactory and scope of better results is present. Pal proposed an elegant
method to reduce the order of linear systems that is computationally simple [1]. Inspired from that
work, an indirect approach was implemented. Instead of optimizing the coefficients of the reduced
order model, new constant parameters -i (i = 1, 2, 3r) were introduced. Optimization of these
constants resulted in better step response.
If the original system is represented by a transfer function G(s) and the reduced order system is
represented by the transfer function R(s).

()|= = ()|=

(i = 1, 2, 3 r)

Taking the same test case as before (Transfer Function 3), 5 coefficients of R(s) are needed to be
determined for which at least 5 equations are needed. If suitable values of i (i ranges from 1 to 5)
can be determined then using the above equation, all the coefficients of the transfer function R(s)
could be determined. An approach similar to the method presented in the previous section was used
to determine the value of these i (i = 1-5) using GA.

Results:
-0.01412 s3 + 5.415 s2 + 30.95 s + 58.57
-----------------------------------------s3 + 65.45 s2 + 793.2 s + 585.7

4.495 s2 + 25.56 s + 46.52


--------------------------------s3 + 55.34 s2 + 645.1 s + 465.2

Transfer Function 7: result obtained from modred ()

Transfer Function 8: Result obtained from GA

s5 + 1014 s4 + 14069 s3 + 69140 s2 + 140100 s + 100000


-----------------------------------------------------------------------s6 + 222 s5 + 14541 s4 + 248420 s3 + 14541006 s2 + 2.22e06 s + 1000000
Transfer Function 9: G(s)

Figure 3: Step Response of Original (solid), GA Reduced (dotted) and modred Reduced (dashed) systems

13

Figure 4: Enlarged Step response. Line (Original), dashed (Reference) and dotted (GA obtained ROM)

Figure 5: Error in Step Response of the GA reduced system and modred Reduced systems, line (Reference), dotted (GA ROM)

Figure 6: Bode Plot of Original, GA Reduced and modred Reduced Systems.


modred reduced system has non-minimum phase

14

Results:
-0.09653 s^4 - 3.756 s^3 - 180.2 s^2 + 447.1 s + 417.1
-----------------------------------------------------s^4 + 11.62 s^3 + 304.2 s^2 + 491.4 s + 456.1
Transfer Function 10: result obtained from modred ()

-6.671 s^3 - 156.6 s^2 + 407.6 s + 410.3


--------------------------------------------s^4 + 14.68 s^3 + 297.1 s^2 + 452.9 s + 448.7
Transfer Function 11: Result obtained from GA

-7.56 s^12 - 495.3 s^11 - 1.404e04 s^10 - 2.244e05 s^9 - 2.185e06 s^8 - 1.288e07 s^7 - 4.09e07 s^6 - 2.274e07 s^5 + 3.401e08 s^4 +
1.347e09 s^3 + 2.249e09 s^2 + 1.729e09 s + 4.682e08
----------------------------------------------------------------------------------------------------------------s^13 + 66.9 s^12 + 2079 s^11 + 4.246e04 s^10 + 6.087e05 s^9 + 6.152e06 s^8 + 4.373e07 s^7 + 2.184e08 s^6 + 7.613e08 s^5 +
1.824e09 s^4 + 2.937e09 s^3 + 3.053e09 s^2 + 1.872e09 s + 5.12e08
Transfer Function 12: G(s)

Figure 7: Step Response of the original (Solid), GA Reduced (Dotted) and modred reduced (Dashed) systems.

Figure 8: Error in step response of the GA reduced (dotted) and modred reduced system (solid)

15

Figure 9: Bode Plot of the Original (Solid), GA reduced (dotted) and modred reduced (Dashed) system

Population Size: 100


Bit Length per Variable: 16
Number of Variables: 5
Cross Over: random, Mutation: 10%
GA parameters used to reduce transfer function 9

Population Size: 100


Bit Length per Variable: 16
Number of Variables: 7
Cross Over: random, Mutation: 10%
GA parameters used to reduce transfer function 12

16

Parallel Processing in Genetic Algorithms:


Matlab supports parallel processing on standalone systems with multiple cores. Parallel computation
was implemented on a small section of the algorithm - the fitness evaluation. Since the fitness of an
individual is independent of the fitness of other individuals, its evaluation can be parallelised. In the
present implementation, a pool of 4 local workers is created using parallel computing toolbox and
roughly 65% reduction in code runtime has been observed.

Extension to MIMO systems:


The original high order system is described by the following equations in state space form.
= +
=
The objective is to find a reduced order model represented by the following equations
= +
=
Yr(t) is a close approximation of Y(t) for a step input. Ar, Br and Cr are constant matrices with
appropriate dimensions.

17
Procedure:
1. Hankel Singular Values provide a measure of energy for each state in a system. Depending
upon the energy contribution, order of the ROM is chosen. As shown below, out of 7 states
present in the system 3 states have major contribution in the system energy so we can choose
the order of the reduced model to be at least 3.

Figure 10: Hankel Singular Values of the Original System

2. Characteristic polynomial of the original system is found by the relation


() = | |
Using dominant pole retention, a new characteristic polynomial may be found. The poles
which are closer to the origin are retained and they determine the value of the reduced
polynomial. This process requires human inspection and various trials are needed before
arriving at an acceptable solution.
() = 0 + 1 + 2 2 + + 1 1 +
3. Ar may be formed from r(s) by
0
0
.
=
0

( 0

1
0
.
0
1

0
1
.
0
2

1
1 )

4. Alternatively, Ar may be determined from Balanced Reduction. Using this technique gives
better results and requires no manual inspection.
5. Numerical value of the original system transfer function is found at s = i (i varies from 1 to r)
and stored in a 3d array named T. These are found using the relation
() = [( )1 ]|=
6. Similar to the SISO case, the value of T found for the original system is matched to the value
of T found for the reduced system.
[( )1 ]|= = [( )1 ]|=
7. For a 2 input - 2 output system, Br and Cr are assumed to be of the form

18
1
1
= (
.
1

12
22
)
.
2

11 12 . 1
= (
)
11 12 . 1
8. Since the first Column of Br matrix is taken to be all ones, the first column of
[( )1 ]|=
Is known. (Ar was previously determined)
9. Using the relation mentioned in step 5, the value of Cr can be determined, provided all the
values of i are known.
10. Once Cr is known, using relation 5, the value of Br can be determined.
11. To determine the optimum values of i, GA is used. Optimization is done to minimize the error
in sum of all the step responses of the MIMO system.

19

Results:
The model of a synchronous machine connected to an infinite busbar [2] is considered in this
example. The system has 2 inputs, two outputs and seven states.
Original Model:

Hankel Singular Values of the system are:

Figure 11: Hankel Singular Values of the original system

20

On observation of the relative magnitudes of the singular values, it is decided that the
reduced model should be of 3rd order.
The eigenvalues of A matrix are:
o -0.20
o -0.47 j9.35
o -37.48
o -46.34
o -13.55 j376.33
Retaining the first 3 dominant eigenvalues (ones which are closer to the origin), () is
determined to be:
() = 3 + 1.1346 2 + 87.9027 + 17.9080
This is transformed into Ar matrix and subsequently Cr and Br are determined.

Reduced Model

Reference Model

Reference Model is determined using balanced reduction. Matlab command used are:
sysOld = ss(A,B,C,D);
[hb,g] = balreal(sysOld);
hmdc = modred(hb,4:7,'MatchDC');
hmdc is a structure that contains the Ar, Br and Cr Matrices

21

Figure 12: Step Response of the MIMO system, Original (Solid), GA Reduced (Dotted) and modred Reduced (Dashed)

Figure 13: Grouped Step Response of the Original system (Solid), GA Reduced system (Dotted) and modred Reduced system
(Dashed)

22

Figure 14: Bode Plot of the Original System (Solid), GA reduced system (Dotted) and modred reduced system (Dashed)

A linearized model of a nuclear reactor [3] is considered in this example. The system has 4
inputs, 4 outputs and 11 states.

23

Figure 15: Hankel Singular Values of the original system

Eigenvalues of A:
-0.0006
-0.0020
-0.0082 j0.0079
-0.0127
-0.1923
-1.0030
-2.8313
-3.6275
-6.9349

Reduced Model

For a 3rd order reduced model, eigenvalues


chosen are:
-0.0006
-0.0082 - j0.0079
-0.0082 +j 0.0079

Reference Model

24

Figure 16: Step Response of the Original System (Solid), GA reduced system (Dotted) and modred reduced system (Dashed)

Figure 17: Combined step response of the Original System (Solid), GA reduced system (Dotted) and modred reduced system
(Dashed)

25

Figure 18: Bode Plot of the Original System (Solid), GA reduced system (Dotted) and modred reduced system (Dashed)

26

Teaching Learning Based Optimisation


Introduction:
TLBO is a teaching-learning process inspired algorithm based on the effect of influence of a teacher
on the output of learners in a class. Teacher and learners are the two vital components of the
algorithm. Two basic modes of the learning are:
Through the teacher (known as teacher phase). A teacher is considered as a highly learned
person who trains learners so that they can have better results.
Through interacting with the other learners (known as learner phase).
TLBO is population based optimization algorithm in which a group of learners is considered as a
population and different design variables are considered as different subjects offered to the learners.
Result of the learners is analogous to the fitness value used in GA.

Algorithm:
The working of TLBO is divided into two parts, Teacher phase and Learner phase.
Teacher phase
This phase of the algorithm simulates the learning of the students through a teacher. During this
phase a teacher conveys knowledge among the learners and puts efforts to increase the mean result
of the class.
At any sequential teaching-learning cycle, let M be the mean result of the learners. The best
solution among the pool will act as a Teacher for the running iteration and attempt to increases the
knowledge level of the whole class.
During teaching phase each solution is updated by a factor called difference mean (DM). If the
updated solution has a better fitness then only it is accepted, otherwise the update is rejected.
= ( )
r is a random number between [0,1] whose value changes at every iteration.
TF can assume only one of the 2 values {1, 2}. Its value is chosen randomly at every iteration.
= +
X denotes the updated value of the solution X. Each learner is updated by the same amount and the
update is kept only if it gives better results.
Learner Phase:
This phase of the algorithm simulates the learning of the students through interaction among
themselves. The students can also gain knowledge by discussing and interacting with the other
students. A learner will learn new information if the other learner has more knowledge than him or
her. The learning phenomenon of this phase is expressed below.
Randomly two distinct learners P and Q are selected and the learner P is updated as below
X P = X P + r (X P X Q ),

If fitness(XP ) > fitness(XQ )

X P = X P + r (X Q X P ),

If fitness(X P ) < fitness(XQ )

27
r is again a random number between [0,1]. Similar to the teaching phase, if the XP has better
fitness than XP then only it is accepted otherwise it is rejected.
Termination:
The algorithm terminates after fixed number of iterations, set by the user.
Flow:

Implementation of TLBO in ROM:


Implementation of TLBO for Reduced Order Modelling is similar to the case of GA where in both SISO
and MIMO case the parameter being optimised are i
For SISO case,

()|= = ()|=

(i = 1, 2, 3 r)

And for MIMO case,

[( )1 ]|= = [( )1 ]|=

28

Results (ROM of SISO systems):

Figure 19: Step Response of the original system (Solid), Reduced System(Dotted) and modred reduced system (Dashed) 2

Figure 20: Bode Plot of the original system (Solid), Reduced System (Dotted) and modred reduced system (Dashed)

-0.01412 s3 + 5.415 s2 + 30.95 s + 58.57


-----------------------------------------s3 + 65.45 s2 + 793.2 s + 585.7

4.922 s2 + 25.97 s + 48.07


--------------------------------s3 + 62.54 s2 + 678.8 s + 480.7

Transfer Function 13: result obtained from


modred ()

Transfer Function 14: Result obtained


from GA

s5 + 1014 s4 + 14069 s3 + 69140 s2 + 140100 s + 100000


-----------------------------------------------------------------------s6 + 222 s5 + 14541 s4 + 248420 s3 + 14541006 s2 + 2.22e06 s + 1000000

Transfer Function 15: G(s)

29

Results (ROM of MIMO):


The model of a synchronous machine connected to an infinite busbar [] is considered in this
example. The system has 2 inputs, two outputs and seven states.
Original Model:

Reduced Model

Reference Model

30

Figure 21: Step Response of the original system (Solid), Reduced System (Dotted) and modred reduced system (Dashed)

Figure 223: Grouped Step Response of the original system, Reduced System and modred reduced system

Figure 234: Bode Plot of the original system (Solid), Reduced System (Dotted) and modred reduced system (Dashed)

31

Conclusions:

First Method (direct optimization of coefficients):


o This technique performs poorly with respect to the balanced reduction technique.
Second Method (optimization of i and using it to find the coefficients)
o This technique performs almost similar to the balanced reduction function in time
domain, but on close observation it can be seen that it doesnt outperform the
reference (balanced reduction)
Reference technique (Balanced Reduction) outputs a non-minimum phase system
TLBO has much less complexity as compared to GA but provides equally good results.
Runtime (s)

GA
42.218

Parallel - GA
17.390

TLBO
36.1563

Future Work

GA and TLBO are few of the many optimisation techniques available in the
literature. Analysis of other optimisation techniques could lead to faster and
efficient implementations.
The technique developed for model order reduction for MIMO systems is
computationally expensive and further research could lead to simple and efficient
procedures.
TLBO is a fairly new optimisation algorithm that was introduced in recently (2012)
and further analysis of the algorithm could lead to even better results.

32

References
[1] Pal J (1986), An algorithmic method for simplification of linear dynamic scalar systems, Int.
J. Control, Vol 43, No. 1, 257-269
[2] Pal J, Ghosh M.K, Sarvesh B (1995), A new method for model order reduction, J. Instn.
Electronics and Telecom. Engrs., Vol 41, No. 5&6
[3] Hickin, J., & Sinha, N. K. (1980). Model reduction for linear multivariable systems.
Automatic Control, IEEE Transactions on, 25(6), 1121-1127.
[4] Al-Baiyat, S. A., & Bettayeb, M. (1994). Appliation of recent model reduction techniques
to nuclear reactors. International journal of modelling & simulation, 14(2), 65-69.
[5] Rao, R. V., Savsani, V. J., & Balic, J. (2012). Teachinglearning-based optimization
algorithm for unconstrained and constrained real-parameter optimization problems.
Engineering Optimization, 44(12), 1447-1462.
[6] Rao, R. V., & Patel, V. (2013). An improved teaching-learning-based optimization
algorithm for solving unconstrained optimization problems. Scientia Iranica, 20(3), 710720.
[7] Sinha, A. K., & Pal, J. (1990). Simulation based reduced order modelling using a clustering
technique. Computers & Electrical Engineering, 16(3), 159-169.
[8] Rommes, J., & Martins, N. (2006). Efficient computation of multivariable transfer
function dominant poles using subspace acceleration. Power Systems, IEEE Transactions
on, 21(4), 1471-1483.

You might also like