CHAPTER 1

INTRODUCTION

Poly House is a climate regulated environment to increase quality & production of crop.
Automatic Poly house Remote System is Arduino based system to achieve the specific
requirements which are to reduce human effort, remotely access the factor and to control the
factor effecting for the growing of plants which may be manually or automatically.

Water is a very precious resource and must be properly utilized. Agriculture is one of those
areas which consume a lot of water. Irrigation is a time consuming process and must be done
on a timely basis. The aim of the project is to develop an automatic system which measures
the moisture of the soil, temperature and humidity of Poly house and automatically turns on or
off the system and send the status of it to the user. The project requires very less human
involvement once installed. The circuit is based on microcontroller, temperature and humidity
sensor and also a soil moisture sensor. A properly configured soil moisture sensor can save up
to 60 percent of water used in irrigation. The designed system can be used in turf grass or with
small garden plants. Temperature and Humidity sensor allow us to know the two main aspects
for the growing of the plants. As we now that different type of plants needs different
environments so this become very useful for the plants as we can maintain the environments
according to it.

Status factor become very important for us. We know the factor

1
problem and characterize the minimal reduction degree for which disturbance decoupling of a
full order plant can be maintained in the reduction procedure. The obvious method of dealing
with a process of uncomfortably high order is to approximate the process by a lower order
process which is an adequate representation of the original. This is done routinely, often
unconsciously, by every control engineer with a job to accomplish: countless control system
designs are based on second- order approximations to whatever order the process happens to
be, and these control systems work, usually quite well. There are other situations, however, for
which a second-order model is not adequate, and where a systematic, analytical approach to
development of an approximation would be desirable.
The Routh frequency-domain reduction (approximation) method[3,8], coupled with powerful
dominant pole-selection criteria, is introduced for the purpose of application to high-degree
transfer functions of multiple-input multiple-output (MIMO)[9] linear time-invariant systems
to obtain corresponding adequate reduced-order models. The proposed method has been
applied successfully to a 9th-order matrix transfer function of a multi-input multi-output time-
invariant linear model of a practical power system. The Routh approximation method for
model reduction is modified, enabling a reduced-order model to be derived whose numerator
order can be selected independently of the order of its denominator. Hence, the proposed
method overcomes the hitherto existing problem that the numerator order of the reduced
model, derived while applying the Routh approximation, is fixed at one less than its
denominator order.The requirement that an approximant to a stable system must be stable
suggests the Routh Approximation Method described in this project. The basic idea underlying
in this method is to develop the well Routh Table for the original system, and then to
construct the approximant in such a manner that the coefficients of its Routh table agree, up to
a given order, with those of the original. By this construction it is obvious that any Routh
approximant of a stable system is stable. The Routh approximation not only preserves stability,
but has other interesting and useful properties that are described in this report.

A MATLAB- based toolbox for model order reduction with a graphical user interface has
options for twenty order reduction methods such as pade approximation, causer continued-

2
fraction expansion, routh stability criterion, differentiation method, and a number of different
mixed methods of frequency techniques.
Recent developments in modeling of complex physical and technical process lead to a
growing interest in model order formulation of large scale dynamical systems. Model order
formulation is the process of deriving the lower order model from the higher order model.
Model order formulation approximates the complex the system by simple one. Modeling is the
mathematical description of dynamic characters of a physical system. For achieving the higher
accuracy while modeling the complex systems, it can be noted than the system order is also
increased. Higher order model are difficult to handle because it has higher computational
complexities and implementation difficulties. The exact analysis of higher order system is both
tedious, costly and it is too complicated to be used in real problems. The use of mor makes it
easier to implement in analysis, design and simulation of the the controllers, compensators,
state variable controllers and observers for the stabilization of the output response of the given
systems.
The main aim of the formulation is to find the best possible approximation of the output of the
original system. Several methods have been proposed for reduction of linear systems in
different fields like control engineering, micro-systems and applied mathematics. During the
past four decades, numerous impressive varieties of new techniques have been developed for
obtaining lower order models from higher order linear system. The first approach is based
upon the classical mathematical concept such as the pade approximation, continued fraction
method, and the time moment Matching method. These approaches have been recognized to
be a powerful method to obtain a reduced order models, but these methods have the
disadvantage that the reduced order model may be unstable although the original system is
stable. Second group of reduction techniques are based on some criteria of stability (such as
the routh stability criterion, Hurwitz polynomial, Mihailov criterion , etc.). The absolute
stability of these methods achieved only by the cost of series loss of accuracy.

3
 CHAPTER 2
COMPLEX SYSTEMS

The power system components, such as generators and FACTS[6] devices, are the basic cells
to onstitute the complicated power systems. When performing the simulation, the stability
analysis, or area controller design, the global model of the power system is needed to be
formed through establishing the local component models. However, the global model is very
complicated since the power system is composed of numerous components with controllers.
Without simplification, the global model cannot meet the actual requirements. In order to
obtain the reduced global model, one can simplify the component models at local level
(component model reduction), or simplify the system model as a whole at global level (system
model reduction).
In order to simplify the approximate linearization models of power systems[3], one can use the
linear model reduction methods (MRM) Ref[3,4] to reduce their orders, such as the balanced
truncation method and the Krylov subspace method. In large disturbance cases the
approximate linearization model becomes inaccurate, which causes above linear MRM not
applicable under these situations. For the reduction methods of nonlinear models, Ref.
[3]modifies the computing algorithm of Gramian controllability and observability matrices in
the balanced truncation method, so as to apply this method to the nonlinear model reduction.
Ref. [3] presents a method to simplify the nonlinear structure via system immersion while
preserving the inputoutput maps. A kind of piecewise approximation method is applied to
simplify the nonlinear system model.
These simplification methods are utilized to handle the nonlinear ordinary differential model
of the power system components, but it is difficult for them to deal with models with nonlinear
differentialalgebraic (DA) form.
As for the power system control, there have been many literatures reporting the application of
nonlinear control theory to power systems, such as Refs. [710]. If the local components adopt
the nonlinear controllers, the reduction problem of component complex models with nonlinear
controllers will be encountered during the process of modeling the complicated power system.
However, at present, the relevant research on this problem has not been reported. Compared
4
with the existing linear or nonlinear model reduction problems, the component complex model
reduction has its particularities: (1) it has the DA structure because a component serves as a
subsystem of the whole power system; and (2) it includes the complicated nonlinear controller.
In the field of power system component modeling and control, a kind of component structural
models with DA subsystem form proposed and discusses the decentralized nonlinear controller
design method for models with this form .This focuses on the reduction method of the
component complex models with DA structure. Firstly, in order to simplify the differential
dynamic equation, this paper uses the linear dynamic equations to describe the original
nonlinear dynamics completely or partially (corresponding to the complete or partial
linearization, if completely linearized, all the nonlinear dynamics can be described by the
linear dynamic both models should be equivalent to each other. The reduced model should
have simpler structure than the original model.equations; otherwise, only those linearized
dynamics can be described by them). Based on the fact that the component complex model has
the linear dynamic equations in the external inputoutput relation after the nonlinearities in the
nonlinear controller and the component offset each other. Secondly, the original state variables
in the algebraic connection equations are substituted by the new defined ones in the above
linear dynamic equations, since the new state variables have already represented the original
ones equivalently, so the equivalent algebraic connection equations of the reduced model are
obtained. Through the above two steps, the reduced model of the complex component with DA
structure is obtained. By the proposed MRM, the paper selects two power system components
(generator and SVC) as examples to derive their reduced models. In order to validate the
method, simulation tests are performed on a typical two-area four-machine power system. The
results show that the derived reduced components to satisfy the demands of theoretical
analysis, simulation and controller design for large-scale systems [15]. In the power system,
the acceptable reduced model of a component subsystem should meet the following
requirements: The reduced model should have the similar dynamic response to the original
model. The connection equations in the reduced model should have the same features to the
ones in the original model (the same connection variable type and number, the same
connection equation number), and the connection equations in both models should be

5
equivalent to each other. The reduced model should have simpler structure than the original
model.

2.1 Problem Associated With Complex System

The nonlinear control configuration of a local component is shown in Fig. 1. The component
model includes differential dynamic equations and algebraic connection equations Ref.[1].
Here, the differential equations are used to describe the dynamic behaviors of the component;
the connection equations are used to express the interrelation between the component and the
external power system. The algebraic variables t of the connection equations are the impact of
the external system on the component, and the impact of the component on the external
system. The nonlinear controller consists of a nonlinear compensator and a linear closed-loop
controller. The compensator is used to offset the non linearities of the component, and the
closed-loop controller guarantees the whole system stable or the controlled variables traceable.
The component subsystem structured model consisting of differential dynamic equations and
algebraic connection equations generally can be written as

where x are the state variables; u are the input variables; x are the intermediate variables; t are
the connection variables; f and g are the vector functions.
The nonlinear controller can be expressed as

where x are the state variables of the controller; y and yr are the output variables and the
x
reference values respectively; f and are the vector functions. It is notable that there is no
dynamic equation if the controller is static.
Through merging the component subsystem model and the nonlinear controller model the
component complex model is written as

6
The variables and equations in the complex model have the clear physical meaning and
specific function. However, this model is not applicable under many situations because it has
complicated structure (too many equations and variables, strong nonlinearity and coupling).
Therefore, it is necessary to simplify the complex model to benefit the research.

7
2.2 Reduction Of Complex Model
For convenience, this paper assumes that the nonlinear controllers are ideal so that the
constraints of the controllers are not taken into account. According to the nonlinear control
methods Ref.[17] (such as the feedback linearization method and the inverse system method),
it is known that the nonlinear compensator designed by these methods can completely (or
partially) offset the nonlinearities of the component dynamics and linearize the nonlinear state
equations. Based on the linearized dynamic model, one can design the linear closed-loop
controller to realize the system stabilization or tracking. In this paper, the linearized dynamic
model and the linear closed-loop controller are combined to replace all the (or part of)
nonlinear dynamic equations in the component complex model. Thus the reduced dynamic
equations are obtained. It is worth noting that all the nonlinear state equations of the
component complex model will be replaced by the linear ones if the nonlinear compensator
completely offsets the nonlinearities of the component dynamics. Otherwise, the hidden
dynamic equations, which are the original nonlinear state equations that are not linearized, will
be included in the reduced dynamic equations.
This paper adopts the inverse system method to design the nonlinear compensator. After the
designed compensator offsets the nonlinearities of the component, assume that the linearized
dynamic model is expressed as

(2.1)

where z1 are the new defined state variables; l are the new defined input variables; the
constant matrices A1 and B1 have appropriate dimensions. Based on the dynamic model (2.1),
suppose that the designed linear closed-loop controller is expressed as

(2.2)

8
where z2 are the introduced state variables of the linear closed-loop controller; yr are the
reference values of the real output; A21, A22, C21, C22, D1 and D2 are the matrices with
appropriate dimensions. By substituting (2) into (1), the linear dynamic equations consisting of
the linearized dynamic model and linear closed-loop controller obtained as below,

(2.3)

According to the linearization extent between the nonlinear compensator and the component,
the MRM will be analyzed for the component complex model in two cases. Without hidden
dynamics If the nonlinear compensator offsets all the nonlinearities of the component
dynamics, there will be no hidden dynamic in the linearized dynamic model (2.1), and thus all
the state equations in (2.1) are replaced by the linear state Eq. (2.3). Furthermore, the algebraic
equations of the nonlinear controller, the third equations in (2.1), are eliminated because the
functions of the controller are integrated into the (2.3). For the algebraic connection equations,
their basic property is not changed, because the connection relation between the component
and external power system stays the same whether or not the nonlinear controller is included.
Nevertheless, because the linear state equations have replaced the original state equations in
the final reduced model, the state variables x in original connection equations are needed to be
substituted by the new defined variables z1 equivalently. Here, assuming that the
transformation from the variables z1 to the variables x is

(2.4)

By substituting (4) into the algebraic connection equations in (2.1), the equivalent connection
equations rewrite as below

(2.5)

9
Combining the linear state Eq. (2.3) and the connection Eq. (2.5) to
describe the component complex model (1) equivalently, the simplified form of model (2.1)
can be obtained as below;

(2.6)

Including hidden dynamics If the nonlinear compensator cannot offset all the nonlinearities of
the component, namely, the linear dynamic model (2.2) still includes some hidden dynamics,
and then the dynamic equations in (2.1) should be equivalent to the linear state Eq. (2.4) and
the hidden dynamic equations.
Here, state variables x decompose into x =(xl,xh), where xl and xh denote the linearized state
variables and the hidden state variables respectively. The hidden dynamic equations are
supposed as
(2.7)

Considering that the original state equations corresponding to xl are replaced by the linear state
Eq. (2.2) corresponding to zl, suppose the transformation from zl to xl as follows:

(2.8)

Then, the original state variables can be expressed as,

(2.9)

Here, the new state variables introduced by the linear closed-loop controller are x = z 2. Thus
by substituting (2.8) and output equations y= h(x,) into the algebraic equations of nonlinear
controller (2.2), (2.1) gains

(2.10)

10
 By substituting (2.9) and (2.10) into
(2.7) to eliminate the control input,
merging the same variables, and defining z = (z1,z2), the hidden dynamic equations can be
written concisely as

(2.11)

Thus the state equations which are used to replace the state equations of the complex model
consist of the linear state Eq. (2.4) and the hidden dynamic Eq. (2.11), namely

(2.12)

Similarly, the algebraic equations of the nonlinear controller in (2.3) are eliminated, and the
linearized state variables xl in the algebraic connection equations need to be substituted by the
new defined state variables zl. Substituting (2.12) into the algebraic connection equations in
(2.3), one gains
(2.13)

Here, combining the above state Eq. (2.13), and algebraic connection Eq. (2.12) to replace the
component complex model (2.3) equivalently, the reduced form of (2.3) obtained from (2.13);

(2.14)

According to the proposed MRM Ref.[1], the reduced structure of the component complex
model without hidden dynamics, the hidden dynamic equations are removed in the structure.
Compared with the complex model, the reduced model has simpler expression, for the reasons
that in the reduced model the new state equations consisting of the linear state equations and
the hidden dynamic equations replace the complicated nonlinear state equations, and the

11
algebraic equations of the nonlinear controller are eliminated. Furthermore, the reduced model
has the same DA subsystem form to the component model without the nonlinear controller,
which may bring some convenience for the simulation modeling and the global controller
design of power systems.
The main processes to obtain the reduced form of the component complex model are
summarized as follows:
Step 1: Find out the dynamics whose nonlinearities are offset by the nonlinear compensator
and gain the corresponding linearized state equations, and then formulate the final linear state
equations by combining these linearized state equations and those of the linear closed-loop
controller.
Step 2: Analyze whether or not the hidden dynamics exist. In the situation without the hidden
dynamics, only the gained linear state equations at Step 1 are used to replace all the state
equations of the component complex model. If the hidden dynamics exist, the gained linear
state equations and the hidden state equations are united to replace all the state equations of
the complex model.
Step 3: Eliminate the algebraic equations of the nonlinear controller, and use the new defined
state variables to substitute the original state variables in the algebraic connection equations so
as to gain the equivalent connection equations.
Step 4: Combine the reduced state equations and the equivalent connection equations to obtain
the reduced form of the component complex model.

12
2.3 Model reduction cases
Reduction for generator complex model Ref.[10] Establish generator complex model. To
establish the generator complex model which is suitable for excitation controller and valve
controller design, this paper ignores some factors of the generator, including the damping
coefficient D, the resistance Ra of stator windings, the electromagnetic transient process of
stator windings, and the dynamic process of excitation system. The state equations of the
generator is
expressed as:

(2.15)

The first two equations denote the dynamic behaviors of rotor angle and rotating speed x, the
following two equations represent q and d axis transient EMF of the generator Ref.[5]
respectively, and the last equation means the valve control of the generator. In the public
coordinate system, the algebraic connection equations (illustrated in Fig. 1) are;

(2.16)

13
Combining the differential state Eq. (2.15) and the algebraic connection Eq. (2.16) we gain the
structured model of the generator subsystem, where, the state variables are

input variables are the intermediate variables are

and the connection variables are

Here, denote the impact of the generator on the external system, and

denote the impact of the external system on the generator. Here, if the terminal
voltage and rotor angle are selected as the controlled variables, then the output equations are

Based on the selected output equations and the structured model of the generator subsystem,
firstly, we can design the nonlinear compensator by the inverse system method to realize the
linearization, and then devise the linear closed-loop controller by the static feedback method.
Finally, synthesize the nonlinear controller as follows

14
 Fig.2.1 The configuration of SVC

Combining the differential state Eq. (2.15), the algebraic connection Eq. (2.16), and the
nonlinear controller (2.16), one gains the generator complex model as

15
Eq. (2.16) is composed of three parts, the first five equations represent the generators
dynamic behaviors, the following two equations represent the connection relation between the
generator and the external system and the last two ones function as the nonlinear controller.

16
 CHAPTER 3
Model Order Reduction Technique

Many modern mathematical models of real-life processes pose challenges when used in
numerical simulations, due to complexity and large size (dimension). Model order reduction
Ref.[1,3,8,9] aims to lower the computational complexity of such problems, for example, in
simulations of large-scale dynamical systems and control systems. By a reduction of the
model's associated state space dimension or degrees of freedom, an approximation to the
original model is computed. This reduced-order model (ROM) can then be evaluated with
lower accuracy but in significantly less time.
After spatial discretization, the number of degrees of freedom is usually very high. It is
therefore very time consuming to simulate such large-scale systems of ODEs or DAEs.
Developed from well-established mathematical theories and robust numerical algorithms,
MOR has been recognized as very efficient for reducing the simulation time of large-scale
systems. Through model order reduction, a small system with reduced number of equations
(reduced model) is derived. The reduced model is simulated instead, and the solution of the
original PDEs or ODEs Ref.[7] can then be recovered from the solution of the reduced model.
As a result, the simulation time of the original large-scale system can be shortened by several
orders of magnitude. The reduced model as a whole can also replace the original system and
be reused repeatedly during the design process, which can further save much time.
Parametric model order reduction methods are designed for model order reduction of
parametrized systems, where the parameters of the system play an important role in practical
applications such as Power Systems Ref.[2,5,6], Integrated Circuit (IC) design, MEMS design,
etc.. The parameters could be the variables describing voltage, frequency, geometrical
measurements, material properties, the damping of the system or the component flow-rate. The
reduced models are constructed such that all the parameters can be preserved with acceptable
accuracy.
Modern discretization techniques for differential equations yield high dimensional simulation
models, which require high computational effort for determining approximate solutions. This
even gets more problematic, if many of such simulations are required, e.g. for parametrized
17
problems. Such settings can be parameter studies, interactive simulations, parameter
identification problems, statistical investigations, etc. For such problems, efficient techniques
for dimensionality reduction are desirable. In addition to fast algorithms, also error
quantification is crucial. Methods for this can be found and are developed in the fields of
Reduced Basis (RB) techniques for parametrized partial differential equations and Model
Order Reduction (MOR) for parametrized dynamical systems.

3.1 Model Reduction Techniques for Continuous-Time Systems

Although there are so many methods available in the literature for the reduction of high-order
continuous time systems, here we introduce some of thewell-known methods viz.
Pade Approximation method
Moment Matching method
Continuous fraction expansion method
Routh Approximation method

3.2 Higher Order Discrete Time System Reduction Methods

There are numerous methods obtainable for the simplification of high order linear continuous
time systems, where very fewer are obtainable for the reduction of linear discrete time system
some of the well-known methods obtainable in the literature and considered here are:
Continued Fraction method
Stability equation method
Error minimization technique
Bilinear Routh Approximation
Improved Bilinear Routh Approximation method

18
3.3 Brief Description Of Various Techniques

3.3.1 Balancing and truncation

The most famous and commonly used method in SVD-based approaches is the Lyapunov
balancing and truncation known also as truncated balanced realization Ref.[15] (TBR). The
basic idea of this approach is to try to delete the states that do not contribute considerably to
the system input-output behavior. In order to study the contribution of each state, the input-
output response is divided into input-to-state and state-to-output responses. The controllability
gramian, which is defined as the integral of the product of the input-to-state response and its
complex conjugate,

is used as an indicator of the amount of input energy needed to reach a certain state. Whereas
the observability gramian, defined as the integral of the state-to-output response and its
complex conjugate,

is used as an indicator of the effect of any given initial state on the output.
It can be shown that these gramians can be calculated by solving the following two Lyapunov
equations involving matrices in the size of the original model:

However, states that are difficult to control, i. e. requiring a large amount of input energy, may
not be difficult to observe and vice-versa. Thus, there is a need for a certain representation of
the system which makes the states difficult to control also difficult to observe and vice-versa.
This can be achieved by the so-called balancing transformation. Balancing was first
introduced by Mullis and Roberts in the design of digital filters and was later brought to the
field of control and order reduction by Moore in the beginning of the 80s. By balancing, a
state transformation T is found which, after being applied to the system matrices, makes the

19
controllability and observability gramians diagonal, equal, and having the Hankel singular
values (HSV) on their diameter. By definition, the Hankel singular values of an observable,
controllable, and stable system are the positive square roots of the eigenvalues of the product
of the gramians and are input-output invariants. Now, for a balanced stable system, the
contribution of each state to the input-output behavior can be measured by the relative
magnitude of its Hankel singular value. Hence, after balancing the original system and
computing the Hankel singular values, a clear idea about the contribution of each state to the
system I/O behavior can be extracted. The truncation of the least contributing states results in
the reduced order model. One of the most important advantages of balancing and truncation is
preserving the stability of the original model, i.e. if the original high-order model is stable,
then the obtained reduced model will be also stable. Also, a global apriori error bound between
the original and reduced systems exists and depends only on the neglected Hankel singular
values. In different reduction approaches that modify the original TBR were introduced. The
method of balancing is then changed with the aim of achieving better approximations or
preserving system properties like passivity. While TBR is theoretically attractive and yields to
excellent approximate models in practice, its use for model reduction of large scale systems (a
few ten thousands) is costly by growing computational complexity in solving two Lyapunov
equations. Many works tried to extend the application of TBR to higher order systems by
parallel computation and approximate solution of Lyapunov or Sylvester equations, leading to
low rank gramians, and resulting in what is known nowadays as approximate balancing and
truncation.

20
3.3.2 Krylov subspace methods
The second main approach in MOR of linear systems is Krylov-based order reduction Ref.[7]
which is a projection-based method that was initially developed to solve large-scale linear
mathematical problems, and at a later stage applied to the control engineering eld and
subsequently to model order reduction. In this family of methods, the aim is to match some of
the rst coecients of the Taylor series expansion of the transfer functions of the original and
reduced models. When the series expansion is developed about a nite point, we speak about
moment matching and Pade interpolation Ref.[12] however when it is about s = 8, we speak
about Markov parameter matching and partial realization. This is in fact the reason behind
calling this approach moment matching or order reduction by moment matching. Krylov
subspaces Ref.[7] are needed in this approach to calculate the projection matrices that are
shown to form bases of these subspaces when aiming at moment and/or Markov parameter
matching. To this end, numerically ecient and iterative methods known as the Lanczos, the
Arnoldi and the two-sided Arnoldi algorithms are employed. Once calculated, these projection
matrices are applied to the system matrices of the original model resulting in a reduced-order
model matching some of the moments and/or the Markov parameters. When comparing TBR
to Krylov subspaces methods, it can be clearly stated that TBR is superior to Krylov subspaces
when stability of the reduced system and error bound are in question. However, Krylov
subspaces methods are superior to TBR when computational eort and memory requirements
are in question. These methods involve mainly matrixvector multiplications and the reduced
order model is obtained directly by a projection to a lower dimensional subspace calculated
through a completely iterative approach. This makes them suitable for the reduction of very
large-scale systems (hundreds of thousands), far above the orders achieved by balancing and
truncation.

3.3.4 Pade Approximation Method:

In this method Ref.[11], the reduced order denominator polynomial has been obtained using an
a dominant pole based pole-clustering approach for reduced order model. The method uses the
improved clustering approach by deciding the value of the ratio of the residue to real parts of

21
poles, taken in descending order and its corresponding reduced order model was obtained
through a simple mathematical procedure. The model so obtained preserves the stability. The
clustering method proposed in this paper differs from the existing pole clustering technique by
considering the distance of system poles from the first pole in the group clustering process and
it gives better approximation for order reduction.
The Lanczos-Pade Ref.[11] connection can be employed to devise efficient and numerically
well behaved algorithms for model order reduction of large-scale linear dynamical systems.
For the case of single-input single-output systems, one such method is the Pade Via Lanczos
(PVL) algorithm Ref.[11], and the extension to the case of general multi-input multi-output
systems is the Matrix-Pade Via Lanczos (MPVL) algorithm. These methods are optimal in the
sense of Pade approximation of the underlying transfer functions. More precisely, the transfer
function H of the original large-scale system is a p m-matrix-valued rational function of the
form

H(s) = CH(sE A)-1B, s C

where A, E CNN , B CNm, C CNp , N is the state-space dimension, m is the number of

inputs, and p is the number of outputs. The transfer function H n of any reduced-order model of
state-space dimension (n< N) of the original system is a p m-matrix-valued rational function
of the form
H
Hn(s) = Cn (sEn An)1Bn, s C

where An, En Cnn , Bn Cnm, and Cn Cnp . The reduced-order transfer function H n is
called an n-th Pade approximant of H (with respect to the expansion point s0 C) if

Hn(s) = H(s) + O((s s0)q(n))

where q(n) is as large as possible. For any fixed expansion point s 0, the MPVL algorithm
generates reduced-order models of state-space dimen- 2 sion n that are optimal in the sense
that the corresponding transfer functions Hn are n-th Pade approximants of the original
22
transfer function H. Unfortunately, Pade-based order reduction methods such as MPVL do not
preserve all the important properties and structures of the original large-scale system. For
example, any meaningful original system will at least be stable, but Pade-based reduced-order
models are not stable in general. One remedy is to relax and to require instead that Hn is only
an n-th Pade-type approximant of H, i.e.

Hn(s) = H(s) + O(s s0)q(n)

where q(n) < q(n). A simple approach to constructing reduced-order models characterized by
such a Pade-type approximation property is to employ projection onto suitable block Krylov
subspaces. For example, if block Krylov Ref.[9] subspaces with the input matrix B as initial
block are used, the transfer functions H n of the resulting reduced-order models are n-th Pade
type approximants of the original transfer function H. Moreover, we have q(n) n, m with
equality in the generic case. Algorithms characterized by such a Pade-type approximation
property include PRIMA and SPRIM. While the approximation order is best possible in the
general case, in the case of SPRIM, there are certain special cases where the Pade-type
approximation property even holds true with q(n) replaced by 2q(n). The reason is that the
original system exhibits certain symmetries and that these symmetries are preserved in the
reduced-order models. A general framework that explains this higher accuracy of Pade-type
reduced-order models. In particular, the notion of J-Hermitian linear dynamical systems to
show that in the case of J-Hermitian linear dynamical systems, the reduced-order transfer
functions match twice as many Taylor coefficients of the original transfer function as in the
general case. An application to the SPRIM algorithm for order reduction of general RCL
electrical networks is also discussed. The remainder of this article is organized as follows. In
Section 2, we recall some basic facts about linear dynamical systems, Krylov subspace-based
model order reduction, and the resulting Pade-type reduced-order models. The notion of J-
Hermitian time-invariant linear dynamical systems and establish the main result of this paper
on Pade-type reduced-order models in the J-Hermitian case. Main result to obtain the Pade-
type approximation property of the SPRIM algorithm.

23
3.3.5 Routh Approximation Method:
The analysis of high order systems (HOS) Ref.[1] is generally very much complicated and
costly. On other hand it became easy to analysis of lower order system. The reduced models
for the original high order system is achieved by using mathematical optimization procedures
or simplification procedures based on physical considerations. Thus analysis, synthesis and
simulation of reduced low order systems is easier and practicable as compared to its high
order systems. An approach to get reduced order model of a higher order system using
timemoments method is presented by. The reduced model have a problem of stability because
of mathematical approximation in model reduction technique. The reduced model may be
unstable even though the high order system is stable. The instability problem of reduced
models was studied by Hutton, Shamash, Gutman et. al. and Wan. Some method based on
stability criterion and other not based on stability criterion but the reduced model for a stable
high order system (HOS) is always stable. Different methods give different approach some
gives batter result in rise time, some gives batter results in settling time. The combination of
these methods gives batter results. The reduced model using combination of methods is nearly
with its higher order system. The combination of Routh approximation Ref.[3,8,16] and
particle swarm optimization (PSO) is presented in. The concept of preservation of stability is
presented in. The differentiation method for reduction of systems is presented in. The
differentiation method is used to derive reduced order model of single machine infinite bus
power system in. The application of Routh stability algorithm is presented in. The application
of soft computing techniques have been presented in literature in the field of model order
reduction. The concept used is minimization of integral squared error using bat algorithm.
The Routh approximation method is a dominant-mode reduction tech- nique. In the frequency
domain the method incorporates parameters obtained from a Routh stability table analysis in
the reduction process. These parameters, called alpha and beta parameters, are obtained from
the coefficients of the MIMO transfer function.

(3.1)

24
where the ai are scalars and the Bi are I x m matrices. (All symbols are defined in appendix
A. ) Algorithmically, the n alpha parameters are computed from the de-
nominator of equation (3.1)

(3.2)

and

(3.3)

where

j
The coefficients ai can be arranged as entries in a Routh table. This tabular construction

suggests the name Routh approximation. The beta parameters are computed from both the
numerator and the denominator of the given transfer function. Consider the single-input -
single- output (SISO) Ref.[16] transfer function from input q to output p implicitly given in
equation (l), W (s). The beta parameters for this SISO transfer function are

(3.4)

and

25
 (3.5)
where

and br is the (qp)th element of the matrix B,, where r = 1 to n. In this way, beta parameters for
the l x m possible SISO transfer functions of equation (3.1) can be found. Again, considering
the SISO transfer function Wpq(s), the computation of the kth approximant from the first k
alpha and beta parameters is given recursively as

(3.6)

and

(3.7)

(3.8)

where

Note that, for different output-input pairs, different

approximant orders k can be chosen. The kth-order approximant for the (pq)th output-input pair
is written as

26
 (3.9)
Properties: The Routh Apprximation exhibits
several very useful properties. Each is important in the reduction problem and is described
briefly here.
Stability: If the original transfer function is asymptotically stable, the Routh approximant, of
any order, will be asymptotically stable.

Pole-zero locations: The poles and zeros of the approximants approach the poles and zeros of
the original function as the order of the approximation is increased.

3.3.6 Continued Fraction Method

One of the more popular methods for large scale systems order reduction has been the
continued fraction techques which was introduced by Chen and Shieh and extended by many
others .the original technique (which would be handled in later chapters ) is based on a Taylor
Series expansion of the system closed loop transfer function about s=0.
Our objective here is to use the continued fraction technique to obtain a reduced order model
for large single input- single output linear time invariant system which falls under the concept
of aggregation.
The continued fraction and time-moments methods for the reduction in order of linear
continuous-time systems are extended for use in reducing the order of linear discrete-time
systems- Simple numerical examples are given to illustrate the methods
"Recurring fraction" redirects here. It is not to be confused with Repeating fraction.

27
A finite continued fraction, where n is a non-negative integer, a0 is an integer, and ai is a
positive integer, for i=1,,n.
In mathematics, a continued fraction is an expression obtained through an iterative process of
representing a number as the sum of its integer part and the reciprocal of another number, then
writing this other number as the sum of its integer part and another reciprocal, and so on.[1] In a
finite continued fraction (or terminated continued fraction), the iteration/recursion is
terminated after finitely many steps by using an integer in lieu of another continued fraction.
In contrast, an infinite continued fraction is an infinite expression. In either case, all integers in
the sequence, other than the first, must be positive. The integers ai are called the coefficients or
terms of the continued fraction.[2]
Continued fractions have a number of remarkable properties related to the Euclidean algorithm
for integers or real numbers. Every rational number p/q has two closely related expressions as
a finite continued fraction, whose coefficients ai can be determined by applying the Euclidean
algorithm to (p,q). The numerical value of an infinite continued fraction will be irrational; it is
defined from its infinite sequence of integers as the limit of a sequence of values for finite
continued fractions. Each finite continued fraction of the sequence is obtained by using a finite
prefix of the infinite continued fraction's defining sequence of integers. Moreover, every
irrational number is the value of a unique infinite continued fraction, whose coefficients can
be found using the non-terminating version of the Euclidean algorithm applied to the
incommensurable values and 1. This way of expressing real numbers (rational and irrational)
is called their continued fraction representation.
It is generally assumed that the numerator of all of the fractions is 1. If arbitrary values and/or
functions are used in place of one or more of the numerators or the integers in the
denominators, the resulting expression is a generalized continued fraction. When it is

28
necessary to distinguish the first form from generalized continued fractions, the former may be
called a simple or regular continued fraction, or said to be in canonical form.

29
 CHAPTER 4
Model Under Consideration

Both active and reactive power demands are never steady and they continually change with the
rising or falling trend: Steam input to turbo-generators Ref.[2] (or water input to hydro-
generators) must, therefore, be continuously regulated to match the active power demand,
failing which the machine speed will vary with consequent change in frequency which may be
highly undesirable* (maximum permissible change in power frequency is 0 5 Hz). Also the
excitation of generators must be continuously regulated to match the reactive
power demand with reactive generation, otherwise the voltages at various system buses may
go beyond the prescribed limits. In modem large interconnected systems, manual regulation is
not feasible and therefore automatic generation and voltage regulation equipment is installed
on each generator. The controllers are set for a particular operating condition and they take
care of small changes in load demand without prescribed limits. With the passage of time, as
the change in load demand becomes large, the controllers must be reset either manually or
automatically.
For small changes active power is dependent on internal machine angle and is
independent of bus voltage; while bus voltage is dependent on machine
excitation and is independent of machine angle . Change in angle
is caused by momentary change in generator speed. Therefore, load
frequency and excitation voltage controls are non-interactive for small
changes and can be modelled and analysed independently. Furthermore
excitation control voltage is fast acting in which the major time constant
encountered is that of the generator field. Thus, the transients in excitation
voltage control vanish much faster and do not aect the dynamics of
power frequency control.

30
4.1 Turbine model:
Let us now relate the dynamic response of a steam turbine Ref.[2] in terms of changes
in power output to changes in steam valve opening yE. Figure 8.4(a) shows a two stage
steam hirbine with a reheat unit. The dynamic response is largely influenced by two factors, (i)
entrained steam between the inlet steam valve and first stage of the turbine, (ii) the storage
action in the reheater which causes the output of the low pressure stage to lag behind that of
the high pressure stage.

Fig.4.1 Two-stage steam turbine

Fig.4.2 Turbine transfer function model

Thus the turbine transfer function is characterised by two time constants. For ease of analysis
it will be assumed here that the turbine can be modelled to have a single equivalent time
constant. Figure (4.2) shows the transfer function medel of a steam turbine. Typically the tie
constant Tt lies in the range 0.5 to 2.5 s.

31
4.2 Generator Load Model:
The increment in power input to generator-load system Ref.[2] is,
PG - PD
where PG is the incremental turbine power output and P D is the load increment. This
increment in power input to the system is accounted for in two ways:
1. Rate of increase of stored kinetic energy in the generator rotor. At scheduled frequency,
the stored energy is
W0ke = H Pr kWs (kilojoules)
where Pr is the kW rating of the turbo-generator and H is defined as its inertia constant.
The kinetic energy being proportional to square of speed (frequency), the kinetic energy at
a frequency of (f0 + f) is given by
2
f 0+ f
0
Wke = W ke ( f0 )
2 f
= HPr ( 1+
f
0 ) (4.1)

Rate of change of kinetic energy is therefore,

d 2 H Pr d
(W ke )
= f )
dt f 0 dt (

(4.2)

2. As the frequency changes, the motor load changes being sensitive to speed, the rate of

PD / f
change of load with respect to frequency, i.e., can be regarded as nearly

constant for small changes in frequency f and can be expressed as,

PD / f
( ) f = B f (4.3)

where the constant B can be determined empirically. B is positive for predominantly motor
load.
32
Writing the power balance equation, we have
2 H Pr d
PG - PD = f )+B f
f
0
dt (

Dividing throughout by Pr and rearranging, we get

2H d
PG(pu) PD(pu) = f )+B(pu) f
f 0 dt (

(4.4)
Taking the Laplace transform, we can write F(s) as
PG ( s ) P D ( s )
F(s) = 2H
B+ 0 +(s )
f

K ps
= [ PG ( s ) PD ( s ) ] ( 1+ T ps S ) (4.5)

where,
2H
Tps = Bf 0 = power system time constant

1
Kps = B = power system gain

Equation (4.5) can be represented in block diagram form as shown in figure

Fig.4.3 Block diagram representation of generator-load model

33
34
4.3 Two area load frequency control:
An extended power system can be divided into a number of load frequency control areas
interconnected by means of tie lines Ref.[2]. Without loss of generality, we shall consider a
two-area
case connected by a single tie line as illustrated in Fig.4.4
The control objective now is to regulate the frequency of each area and to simultaneously
regulate the tie line power as per inter-area power contracts. As in the case of frequency,
proportional plus integral controller will be installed so as to give zero steady state error in tie
line power flow as compared to the contracted power.

Fig.4.4 Two interconnected control areas

It is conveniently assumed that each control area can be represented by an equivalent turbine,
generator and governer system. Symbols used with suffix 1 refer to area 1 and those with
suffix 2 refer to area 2.
In an isolated control area case, the incremental power (P G - PD) was accounted by the rate
of increase of stored kinetic energy and increase in area load caused by increase in frequency.
Since a tie line transports power in or out of an area, this fact must be accounted for in the
incremental power balance equation of each area.

Power transported out of area 1 is given by,

35
 |V 1|V 2 X 0 0
Ptie,1 = 12 sin( 1 2 )

(4.6)
where,
01 , 02 = power angles of equivalent machines of the two areas.

For incremental changes in and , the incremental tie line power can be expressed
1 2

as

|V 1|V 2 P
0 0
T12 = r 1 X 12 cos( 1 2 ) = synchronizing coefficient

Since incremental power angles are integrals of incremental frequencies, we can write Eq,
(4.6) as

Ptie,1 = 2T12 ( f 1 dt f 2 dt ) (4.7)

f1 f2
where, and are incremental frequency changes of ares 1 and 2 respectively.

Similarly, the incremental tie line power out of area 2 is given by

Ptie,2 = 2T21 ( f 1 dt f 12 dt ) (4.8)

where,
Pr1
|V 2|V 1 P
T21 =

0 0
r 2 X 21 cos( 2 1 ) = ( ) Pr2 T12 = 12T12

(4.9)

36
With reference to equation (4), the incremental power balance equation for area 1 can be
written as
2 H1 d
PG1 - PD1 = f )+B1 f
f 1 dt ( +Ptie,1
0
1 1

(4.10)

It may be noted that all quantities other than frequency are in per unit in Eq. (10). Taking the
Laplace transform of Eq. (10) and reorganising, we get

K ps 1
F1(s) = [PG1(s)-PD1(s)-Ptie,1(s)] 1+T ps 1 s (4.11)

Fig.4.5 Block diagram of an isolated area

Compared to equation (4.5) of the isolated control area case, the only change is the appearance
of the signal Ptie,1(s) as shown in above figure.
Taking the laplace transform of equation (6) the signal Ptie,1(s) is obtained as

2 T 12
Ptie,1(s) = s [F1(s)- F2(s)] (4.12)

37
Combining the basic block diagrams of the two control areas with Pc1(s) and Pc2(s)
generated by integrals of respective ACEs Ref.[5,6] (obtained through signals representing
changes in he line power and local frequency bias) and employing the block diagrams we
easily obtain the composite block diagram of Fig. 4.6.

Fig.4.6 Composite block diagram of two-area load frequency control (feedback loops provided with integral of
respective area control errors)

Fig.4.7 State space model of two-area system

38
Let the step changes in loads PD1 and PD2 be simultaneously applied in control areas 1 and
2. respectively. When steady conditions are reached, the output signals of all integrating
blocks will become constant and in order for this to be so, their input signals must become
zero.
Before presenting the optimal design, we must formulate the state model. This is achieved
below by writing the differential equations describing each individual blockin terms of state
variables.
Comparing Figs. 6 and 7,

x1= f 1 x4= f 2 x8= ACE1 dt

x2= P G1 x5= P G2 x9= ACE2 dt

u1=PC1 u2=PC2
w1=PD1 w2=PD2

For block 1

x1+Tps1 x 1 = Kps1(x2-x7-w1)

or,
1 K ps1 K ps1 K ps1
x =-
1 T ps 1 x1 + T ps 1 x2 - T ps 1 x7 - T ps 1 w1

(4.13)
For block 2

x2+Tt1 x 2 = x3

or,
1 1
x =-
2 T t 1 x2 + T t 1 x3

(4.14)

For block 3

39
 1
x3+Tsg1 x 3 = - R 1 x1 + u1

or,
1 1 1
x
3= - R 1 T sg 1 x1 - T sg 1 x3 + T sg 1 u1

(4.15)

For block 4

x4+Tps2 x 4 = Kps2(x5 + 12x7 - w2)

or,
1 K ps2 12 K ps 2 K ps2
x =-
4 T ps 2 x4 + T ps 2 x5 + T ps2 x7 - T ps 2 w2

(4.16)

For block 5

x5+Tt2 x 5 = x6

or,
1 1
x =-
5 T t 2 x5 + T t 2 x6

(4.17)

For block 6
1
x6+Tsg2 x 6 =- R 2 x4 + u2

or,
1 1 1
x =-
6 R 2 T sg 2 x4 - T sg 2 x6 + T sg 2 u2

(4.18)
40
For block 7
x 7 T 12 x 1 T 12 x 4
=2 -2

For block 8
x 8 b1 x1 x7
= +

For block 9
x 9 b2 x 4 a12 x 7
= -

x =Ax + Bu + Fw

Where,
T
x = [x 1 x 2 . x 9 ] = state vector
T
u = [u 1 u2 ] = control vector
T
w = [w1 w2 ] = disturbance vector

while the matrices A,B and F are defined below:

41
 [ ]
1 Kps 1 Kps
0 0 0 0 0 0
Tps 1 Tps 1 Tps
1 1
0 0 0 0 0 0 0
Tt 1 Tt 1
1 1
0 0 0 0 0 0 0
R 1Tsg 1 Tsg1
1 Kps 2 a 12 Kps 2
0 0 0 0 0 0
A= Tps 2 Tps 2 Tps 2
1 1
0 0 0 0 0 0 0
Tt 2 Tt 2
1 1
0 0 0 0 0 0 0
R 2Tsg 2 Tsg 2
2 T 12 0 0 2 T 12 0 0 0 0 0
b1 0 0 0 0 0 1 0 0
0 0 0 b2 0 0 a 1 2 0 0

[ ]
1
0 0 0 0 0 0 0 0
T Tsg 1
B = 1
0 0 0 0 0 0 0 0
Tsg 2

[ ]
Kps1
0 0 0 0 0 0 0 0
T Tps 1
F = Kps 2
0 0 0 0 0 0 0 0
Tps 2

As the control areas extend over vast geographical regions, there are two ways of obtaining
full state information in each area for control purposes.
1. Transport the state information of the distant area over communication channels. This
is, of course, expensive.
2. Generate all the states locally (including the distant states) by means of observer or
Kalman filter Ref.[2] by processing the local output signal. An observer being itself
42
 a high order system, renders the overall system highly complex and may, in fact, result
in impairing system stability and dynamic response which is meant to be improved
through optimal control.

CHAPTER 5
Present work

The ninth order transfer function is taken as,

G(s)o =

10806 s5 +72.93 s 4 +847.8 s3 +2216 s2 +785.1 s1.309 1012

s 9+ 45.02 s 8+ 697.6 s 7+ 4629 s 6 +1.676 10 4 s5 +3.507 104 s 4 + 4.161 104 s 3 +2.452 104 s 2 +6673 s +664.1

5.1 Programming of the model in MATLAB

clc
clear
% ROUTH APPROXIMATION METHOD ON AGC FOR TWO AREA control model 9 th
order
Ac=[-0.05 5.988 0 0 0 0 0 0 -5.988 0 0;0 -0.1 -1.567 1.667 0 0 0 0 0 0 0;0 0 -3.333 3.333 0 0 0
0 0 0 0;-5.208 0 0 -12.5 0 0 0 0 0 0 0;0 0 0 0 -0.05 5.988 0 0 5.988 0 0;0 0 0 0 0 -0.1 -1.567
1.667 0 0 0;0 0 0 0 0 0 -3.333 3.333 0 0 0;0 0 0 0 -5.208 0 0 -12.5 0 0 0;0.545 0 0 0 -0.545 0 0
0 0 0 0;0.425 0 0 0 0 0 0 0 1 0 0;0 0 0 0 0.425 0 0 0 -1 0 0]

B=[0 0;0 0; 0 0;12.5 0;0 0;0 0; 0 0; 0 12.5; 0 0;0 0;0 0]

43
C=[1 0 0 0 0 0 0 0 0 0 0;0 0 0 0 1 0 0 0 0 0 0;0 0 0 0 0 0 0 0 1 0 0]

D=[0]

n=9;r=6;k=n;
i=1;
for j=1:5;
R(1,j)=den(1,i); %routh array first row from denominator odd coeff.
R(2,j)=den(1,i+1); %routh array first row from denominator even coefficient
i=i+2;
end
% TO COMPLETE ROUTH ARRAY
for i=3:n+1;
for j=1:4;
R(i,j)=[R(i-2,j+1)*R(i-1,1)-R(i-2,1)*R(i-1,j+1)]/R(i-1,1);
end
end
R
'CALCULATION OF ALPHA COEFFICIENT'
for i=1:r;
alpha(i)=R(i,1)/R(i+1,1)
end
'CALCULATION OF BETA COEFFICIENTS FOR g11'
i=1;
for j=1:4;
BT(1,j)=g11(1,i); %routh array first row from numerator g11 odd coeff.
BT(2,j)=g11(1,i+1); %routh array first row from numerator g11 even coefficient
i=i+2;
end
k=0

44
for i=3:n+1;
k=k+1;
beta(1,k)=BT(k,1)/R(k+1,1);
for j=1:3;
BT(i,j)=[BT(i-2,j+1)-beta(1,k)*R(i-1,j+1)];
end
end
BT
'CALCULATION OF BETA COEFFICIENTS FOR g21 OUTPUT'
i=1;
for j=1:4;
BT21(1,j)=g21(1,i); %routh array first row from denominator odd coeff.
BT21(2,j)=g21(1,i+1); %routh array first row from denominator even coefficient
i=i+2;
end
k=0
for i=3:n+1;
k=k+1;
beta21(1,k)=BT21(k,1)/R(k+1,1);
for j=1:3;
BT21(i,j)=[BT21(i-2,j+1)-beta21(1,k)*R(i-1,j+1)];
end
end
BT21
'CALCULATION OF BETA COEFFICIENTS FOR g31 OUTPUT'
i=1;
for j=1:4;
BT31(1,j)=g31(1,i); %routh array first row from denominator odd coeff.
BT31(2,j)=g31(1,i+1); %routh array first row from denominator even coefficient
i=i+2;

45
end
k=0
for i=3:n+1;
k=k+1;
beta31(1,k)=BT31(k,1)/R(k+1,1);
for j=1:3;
BT31(i,j)=[BT31(i-2,j+1)-beta31(1,k)*R(i-1,j+1)];
end
end
BT31
'CALCULATION OF BETA COEFFICIENTS FOR g12 '
i=1;
for j=1:4;
BT12(1,j)=g12(1,i); %routh array first row from denominator odd coeff.
BT12(2,j)=g12(1,i+1); %routh array first row from denominator even coefficient
i=i+2;
end
k=0
for i=3:n+1;
k=k+1;
beta12(1,k)=BT12(k,1)/R(k+1,1);
for j=1:3;
BT12(i,j)=[BT12(i-2,j+1)-beta12(1,k)*R(i-1,j+1)];
end
end
BT12
'CALCULATION OF BETA COEFFICIENTS FOR g22'
i=1;
for j=1:4;
BT22(1,j)=g22(1,i); %routh array first row from denominator odd coeff.

46
 BT22(2,j)=g22(1,i+1); %routh array first row from denominator even coefficient
i=i+2;
end
k=0
for i=3:n+1;
k=k+1;
beta22(1,k)=BT22(k,1)/R(k+1,1);
for j=1:3;
BT22(i,j)=[BT22(i-2,j+1)-beta22(1,k)*R(i-1,j+1)];
end
end
BT22
'CALCULATION OF BETA COEFFICIENTS FOR g32 OUTPUT'
i=1;
for j=1:4;
BT32(1,j)=g32(1,i); %routh array first row from denominator odd coeff.
BT32(2,j)=g32(1,i+1); %routh array first row from denominator even coefficient
i=i+2;
end
k=0
for i=3:n+1;
k=k+1;
beta32(1,k)=BT32(k,1)/R(k+1,1);
for j=1:3;
BT32(i,j)=[BT32(i-2,j+1)-beta32(1,k)*R(i-1,j+1)];
end
end
BT32
'BETA MATRICES'
beta

47
beta12
beta21
beta22
beta31
beta32

B1=[beta(1,1) beta12(1,1);beta21(1,1) beta22(1,1);beta31(1,1) beta32(1,1)]

B2=[beta(1,2) beta12(1,2);beta21(1,2) beta22(1,2);beta31(1,2) beta32(1,2)]

B3=[beta(1,3) beta12(1,3);beta21(1,3) beta22(1,3);beta31(1,3) beta32(1,3)]

B4=[beta(1,4) beta12(1,4);beta21(1,4) beta22(1,4);beta31(1,4) beta32(1,4)]

B5=[beta(1,5) beta12(1,5);beta21(1,5) beta22(1,5);beta31(1,5) beta32(1,5)]

B6=[beta(1,6) beta12(1,6);beta21(1,6) beta22(1,6);beta31(1,6) beta32(1,6)]

a1=alpha(1,1);a2=alpha(1,2);a3=alpha(1,3);a4=alpha(1,4);a5=alpha(1,5);a6=alpha(1,6);
%reduced order=3

A3S=[1 a1+a3 a2*a3 a1*a2*a3]

B3S=[a2*a3*B1 a3*B2 (B1+B2)]

%reduced order=6

A6S=[1 (a2+a4+a6) (a1*a6+a3*a6+a5*a6+a1*a4+a3*a4+a1*a2)

(a4*a5*a6+a2*a5*a6+a2*a3*a6+a2*a3*a4)
(a1*a2*a3*a6+a1*a4*a5*a6+a3*a4*a5*a6+a1*a2*a5*a6+a1*a2*a3*a4) (a2*a3*a4*a5*a6)
(a1*a2*a3*a4*a5*a6)]

48
B6S=[(B2+B4) (a4*B1+a4*B2+a2*B1+B1+B2) a5*a6*(B2+B4)+a3*a6*B2+a3*a4*B2
(a4*a5*a6*B1+a2*a5*a6*B1+a2*a3*a6*B1+a2*a3*a4*B1+a4*a5*a6*B2)
(a3*a4*a5*a6*B2) (a2*a3*a4*a5*a6*B1)]

num6={[0.1926 0.0207 0.6244 0.0250 0.0087 0] [0.1645 0.0207 0.5350 0.0250 0.0087 0];
[0.1645 0.0207 0.535 0.025 0.0087 0] [0.1926 0.0207 0.6244 0.0250 0.0087 0];[0.067 0.0088
0.2183 0.0106 0.0037 0] [-0.067 -0.0088 -0.2183 -0.0106 -0.0037 0]};

red6=tf(num6,A6S);
step(org,red6)
tf(red6)
tf(org)

Step Response
From: In(1) From: In(2)
0.8

0.6

0.4
To: Out(1)

0.2

0
Amplitude

-0.2
0.8

0.6

0.4
To: Out(2)

0.2

-0.2
0 10 20 30 0 10 20 30
Time (sec)

49
5.2 Reduced model(3rd order transfer function)

0.2867 s 2+ 0.2534 s9.6683 1021

G(s)r = 3 2
s + 0.7639 s + 0.2175 s+ 0.2164

50
5.3 Validation of reduced order

Step Response
1.4

1.2

0.8
Am plitude

0.6

0.4

Original
0.2 Redused

0
0 2 4 6 8 10 12 14
Time (sec)

Step Response
4

3.5

2.5
Amplitude

1.5

0.5

0
0 2 4 6 8 10 12
Time (sec)

51
 Step Response

Original
Reduced

Amplitude

0 5 10 15 20 25 30 35 40 45
Time (sec)

Fig.5.1 Graphical validation of reduced order

Measurements

52
 Chapter 6
Result and Conclusions

In this project, we have proposed a Routh Approximation model reduction method for linear
systems. The proposed approach retains the required physical realizability property of the
original full-order system while ensuring that the transfer function of the reduced-order system
matches that of the original system at multiple points (or frequencies) along some tangent
directions. That is, the reduced system can be designed to match specified input-output
responses of the original system at various frequencies. We establish a new result illustrating
that our reduced-order system will typically be asymptotically stable.
It has been concluded that the proposed method is simple, computer oriented and comparable
in quality. This method guarantees stability of the reduced model if the original high-order
system is stable and exactly matches the steady-state value of the original system. The
proposed method is equally applicable to linear multivariable systems and is reported
elsewhere. In the proposed reduction strategy, the system complexity has been decreased from
the original order n towards the dimension of the largest controlled invariant subspace of the
system, which is strictly smaller than n.
This extends the possible reduction strategies towards models for real physical systems where
we cannot feedback the complete state space.

53
 REFERENCES

[1] Orlando FL, A model reduction scheme with preserved optimal performance, 50th
IEEE Conference on Decision and Control and European Control Conference (CDC-
ECC), 2011
[2] D P Kothari and I J Nagrath, Modern Power System Analysis, New Delhi, Tata
McGraw Hill Education Pvt. Ltd., Third Edition, 2009.
[3] Maurice F. Hutton and Bernard Friedland, Routh Approximations for Reducing Order
of Linear, Time-Invariant Systems, EEE Transactions on Automatic Cohmol, vol. ac-
20, no. 3, June 1975
[4] Lucas TN. linear system reduction by impulse energy approximation, IEEE trans auto
contr; 30(8):784-786, 1985.
[5] Dr. B.R. Gupta, Power System Analysis and Design, New Delhi, S. Chand and
Company Pvt. Ltd., Seventh Edition, 2014
[6] Ashfaq Hussain, Electrical Power Systems, New Delhi, CBS Publishers and
Distributors Pvt. Ltd.,Fifth Edition, 2007
[7] B. D. O. Anderson, The realization problem for hidden Markov models, Math.
Control Signals Syst., vol. 12, no. 1, pp. 80122, Apr. 1999.
[8] Hutton MF, Friedland, Routh Approximations for reducing order of linear, time
invarient systems. IEEE Trans auto control; 20:329-337, 1975.
[9] Kumar BK, Singh SN, Srivastava SC. A decentralized nonlinear feedback controller
with prescribed degree of stability for damping power system oscillations. Electric
Power Syst Res; 77(34):20411., 2007;
[10] Moore BC. Principal Component Analysis in linear systems: controllability,
observability and model reduction. IEEE J contr; 26(1):17-32, 1984
[11] Aoki M. Some approximation methods for estimation and control of large scale
systems. IEEE trans auto contr; 23(2): 173-182, 1978.
[12] Lopez Rios A, Messina AR. An optimal modal approximation method for
model reduction of linear power system models. Int J Electr Power Energy
Syst;44:293300 2013.

54
[13] K. Gallivan, A. Vandendorpe, and P. Van Dooren, Model reduction of MIMO
systems via tangential interpolation, Siam J. Matrix Analysis Appl., vol. 26, no. 2, pp.
328349, 2004.
[14] S.S.Lambaand S.Vitthal RAO,Aggregation matrix for the reduced order
continued fraction expansion model of chen and shieh, IEEE transaction on automatic
control,23,81,83,1978.
[15] Joanna Iwaniec Novel Approach to Model Order Reduction by means of the
Balanced Realization Method,Molecular and Quantum Acoustics vol.27,2006.
[16] G.Buskes and M.Cantoni.A step-wise procedure for reduced order
approximation in the v-gap metric, in proc. 47th IEEE conf. on decision and
control,cancun,mexico,2008.
[17] Jianminglian, Wei Zhang, Jianghai Hu and Stanislaw H.Zak Variable neural
adaptive robust output feedback control of uncertain systems American Control
Conference Marriott Swaterfront,Baltimore,md,USA; 2010.

55