Model Reference Adaptive Controller 8.1 Objective

Adaptive controller strategies for FACTS devices in a power system to enhance stability
Chapter 8
Model Reference Adaptive controller
8.1 Objective
The objective of this section of research is to develop a controller which is
adaptive in nature. To adapt means to change a behavior to conform to new
circumstances. In this section a model reference adaptive controller is developed. A
model which can adapt itself to changes in the power system‟s specific parameter is
developed. The reference model chosen here is the Zreference which changes
periodically. TCSC can operate in two modes inductive and capacitive. A MRAS
SIMULINK model is developed both in inductive mode and capacitive mode of
operation. The system is applied with an ANN to train the firing angle of the SCR in the
TCSC. This also presents the results to justify the use of adaptive controller to improve
the stability.
8.2 Basics of Model reference adaptive controller

Introduction: Power systems are subjected to unpredictable disturbances such as
injection or rejection of loads, changes in operating points of machines, and system
faults. The presence of Flexible AC Transmission System (FACTS) devices further
introduces nonlinearity in the system. The traditional phase lead-lag controllers in a
power system are tuned for a certain operating condition, which may not perform as
desired at another operating condition. Adaptive control techniques are desirable under
such circumstances. The adaptive controller adapts to new operating condition in real-
time and yields optimum response over wide operating regions .An adaptive control
algorithm works on real-time identified plant model. The model estimator should be able
to track changes in power system rapidly and smoothly for uniform control action.
Recursive least square estimation (RLS) method is one of the most popular system
identification routine because of its simplicity and numerical stability .However, general
RLS algorithm has divergent behavior due to round-off errors .Furthermore, the noise
PET Research centre, PESCE , Mandya 197

model is very unrealistic in practical situations and in most of the applications; it results
in erroneous parameter estimates for the RLS .
Power system disturbances are in general large disturbances such as change in
load and system faults. The parameters identified in such conditions using normal RLS
algorithm can have rapid deviations and this can cause undesirable control output. To
rectify such problems, ad-hoc solutions have been proposed in the literature. The authors
propose the use of a dynamic limit to the parameters. The use of tracking constrained
coefficient in recursive updating formula of the identified parameters is proposed in some
papers and the use of random-walk term in updating the covariance matrix for short
duration following the large disturbances has been proposed.
One of the way to limit the effects of dramatic process variations is to use an
adaptive control scheme. Model reference adaptive controller is one of the adaptive
controller methods.
In everyday language to adapt means to change a behavior to conform to new
circumstances. Intuitively, an adaptive regulator is a regulator that can modify its
behavior in response to changes in dynamics of the process and disturbances. An adaptive
system is any physical system that has been designed with an adaptive viewpoint.
The model reference adaptive system was originally proposed to solve a problem in
which the specifications are given in terms of a reference model that tells how the process
output ideally should respond to the command signal.
The model reference adaptive system (MRAS) is one of the main approaches to
adaptive control. The basic principle is illustrated in Fig. The desired performance is
expressed in terms of a reference model, which gives the desired response to a command
signal. The system also has an ordinary feedback loop composed of the process and
regulator. The error e is the difference between the outputs of the system and reference
model. The regulator has parameters that are changed based on the error. There are two
loops an inner loop , which provides the ordinary control feedback, and an outer loop,
which adjusts the parameters in the inner loop. The inner loop is assumed to be faster
than the outer loop.
The basic concept of a model reference adaptive control is shown in Figure 8.1.
The response of the nonlinear power system including FACTS devices is modeled by a
known model. The coefficients of the model of the plant are estimated in real-time using
a recursive algorithm. The estimated parameters are used to design the controller to meet
the specific requirement. Once the optimized controller parameters are achieved, the
controller generates appropriate control signal. In this control architecture, the feedback
control loop is called inner-loop while model adjustment loop is referred to as outer-loop.
The inner loop is relatively faster compared to the outer loop.
A(z-1) y(t) = z-nd B(z-1) u(t) + C(z-1) e(t)[171] 8.1
Where, y(t), u(t) and e(t) are system output, system input and noise terms
respectively. A(z-1), B(z-1) and C (z-1) are the polynomials expressed in terms of the
backward shift operator z-1 and are defined as:
A(z-1) = 1+ a1 z-1 + a2 z-2 +…+ ana z-na.
B(z-1) = 1+ b1 z-1 + b2 z-2 +…+ bnb z-nb.
C(z-1) = 1+ c1 z-1 + c2 z-2 +…+ cnc z-nc.
na, nb and nc are the order of the polynomials A(z-1), B(z-1), and C(z-1) respectively. The
variable nd represents the delay term.
Fig. 8.1 Model reference adaptive control schematic
A. System identification
The system parameter estimation is critical step for designing a proper adaptive
control of the system. It is desirable to have smooth identified parameter variation for

large burst disturbances, which are typical type of disturbances for power systems. To
cope with such large disturbances, a robust recursive least square approach is
implemented in this work. This algorithm is robust for cases that arise due to numerical
truncations and large disturbances.
Equation (8.1) is expressed as:
Y(t) = φT (t)W(t-1) + e(t) 8.2
Where W(t-1) = [a1 a2 … ana b1 b2 … c1 c2 … cnc]T is the system parameter vector,
φ(t) = [-y (t-1) –y(t-2) … -y(t-na) u(t-nd) u(t-nd -1) … u(t-nd –nb) e(t-1) … e(t-nc)]T is the
sampled input / output data vector and e(t) is the error term. Then the robust recursive
least square algorithm given in Eq. (8.3) can be utilized to identify the parameter vector
W(t) [8.5]:
8. 3
Where, yr(t) is the reference plant output, is the prediction error and P(t) is
the covariance matrix. The initial conditions are P(0)=cI, and W(0)=0. The tuning
parameters are c>1 and γ. The initial value of c should satisfy the conditions c/ γ2 < 1 and
P(1) ≤ P(0). Furthermore, selection of γ should satisfy the condition ,
where δ is small positive number and ø(t) = ∑φφT.
B. Pole-shift control concept

A pole-shift feedback controller has the transfer function given as:
8.4
Where,
F(Z-1) = 1 +f1z-1+ f2z-2+…….+ fn f z-n f.
G(Z-1) = g0 +g1z-1+ g2z-2+…….+ gn g z-n g.

And nf = nb-1, ng=na-1.The closed loop system configuration of Eq.(8.1) and (8.4) is
shown in Figure 8.2.
Fig. 8.2 Closed-loop system configuration.

Fro Eq.(8.1) and (8.4), characteristics equation of the closed loop control is:
T(z-1) = A(z-1)F(z-1)+B(z-1)G (z-1) 8.5
In pole-shift control, the characteristics equation of Eq. (8.5) takes the form of
-1
A(z ) with the pole shifted by factor of α. The new characteristics equation can be
obtained by replacing z-1 in A(z-1) by α(z-1) as follows:
A(αz-1) = A(z-1)F(z-1)+B(z-1)G (z-1) 8.6
Rearranging the Eq.(8.6) in matrix form:
Or,
M w(α)=L(α) 8.7
The parameters ai and bi are obtained from identifier. The Eq. (8.7) can be solved
for fi and gi for a known value of α. Once the value of fi and gi are obtained, the control
signal can be computed using Eq. (8.4). For fixed value of α, control becomes a special
case of the pole-assignment control.

The value of α can be selected to satisfy some optimized performance indices.

One such performance index is the minimization of the one time-step ahead prediction
error i.e.
Min J (t+1, αt) = E[ (t+1) - yr(t+1)]2 8.8
αt
where, (t+1) is predicted output and yt(t+1) is the reference value. The predicted output
(t+1) can be calculated as:
(t+1) = XT(t)β+b1u(t,αt) 8.9
Where, X(t) = [- u(t-1)… -u(t-nf) –y(t) –y(t-1)… -y(t-ng)]T, and β =[-b2-b3…-bnb a1 a2 …
ana]T.
Minimization of the objective function defined in Eq.(8.8) yields the optimal
value of αt. The value of αt should be kept in the range of [-1/λt < αt <1/ λt] to satisfy the
stability constraints, where λt represents the largest absolute value of the roots of
characteristics equation T(z-1) [2,3]. Furthermore, the control signal also should lie within
the control constraint:
Umin< u(t, αt) < umax.
Where Umin and umax are minimum and ;maximum control signal boundaries.
The general idea behind Model Reference Adaptive Control (MRAC, also know
as an MRAS or Model Reference Adaptive System) is to create a closed loop controller
with parameters that can be updated to change the response of the system. The output of
the system is compared to a desired response from a reference model. The control
parameters are update based on this error. The goal is for the parameters to converge to
ideal values that cause the plant response to match the response of the reference model.
For example, you may be trying to control the position of a robot arm naturally vibrates.
You actually want the robot arm to make quick motions with little or no vibration. Using
MRAC, you could choose a reference model that could respond quickly to a step input
with a short settling time. You could then build a controller that would adapt to make the
robot arm move just like the model.
The Model Reference Adaptive System is one of main approaches to adaptive
control. The basic principle is illustrated in Fig.8.1. The desired performance is expressed
in terms of a reference model which gives the desired response to a command signal. The
system also has an ordinary feedback loop composed of the processor and regulator. The
error e is the difference between the output of the system and reference model. There are
two loops an inner loop which provides the ordinary control feedback and the outer loop
which adjusts the parameters in the inner loop.
MRAS were originally derived for servo problem in deterministic continuous time
system. There are three basic approaches to the analysis and design of MRAS
Gradient Approach
Lyapunov functions
Passivity theory
The gradient method was used by Whitaker in the original work on the MRAS.
This approach is based on the assumption that the parameters change slowly than the
other variables in the system. This assumption is essential for the computation of the
sensitivity derivatives that are needed in the adaptation mechanism.
Model following is an important part of the MRAS.
8.2.1 The MRAS problem in general

For a system with adjustable parameters the model reference adaptive method
gives a general approach for adjusting the parameter so that the closed loop transfer
function will be close to a prescribed model. This is called model following problem. One
important question is how small we can make the error e. This depends both on the
model, the system and the command signal. If it is possible to make the error equal to
zero for all command signals, then perfect model following is achieved. In this research it
is tried to make the impedance difference which is the impedance error as small as
possible.
process
Regulator
uc u y
Ru=Tuc –Sy
Fig. 8.3 closed loop system with general linear controller.

Analysis of this case gives good insight into the design problem. It also provides a
way of unifying treatment of the MRAS.
Consider a single-input, single-output system, which may be either a continuous
time model or a discrete-time model:
8.10
Where u is the control signal and y the output signal. The symbols A and B denote
polynomials in either the differential operator p or the forward shift operator q. It is
assumed that A and B are relatively prime. Furthermore, it is assumed that deg A≥ deg B,
i.e., that the system is proper (the continuous-time case) or causal (the discrete-time
case). The polynomial A is assumed to be monic, i.e., the first coefficient is unity.
Assume that we want to find a regulator such that the relation between the
command signal uc and the desired output signal ymis given by
8.11
Where Am and Bm are polynomials in the differential operator or the forward shift
operator. A general shift operator. A general linear control law can be described as
Ru = Tuc– Sy 8.12
Where R,S, and T are polynomials. This control law represents a negative
feedback with the transfer operator – S/R and a feedforward with the transfer operator
T/R. Elimination of u between Eqs. 8.10 and 8.12 gives the following equation for the
closed-loop system:
(AR+BS)y=BTuc 8.13
It follows from Eq 8.13 that the characteristic polynomial of closed-loop system
is AR+BS. The remaining factor can be interpreted as observer dynamics. There are thus
three type of factors of the characteristic polynomial: canceled process zeros given B +,
desired model poles given by Am, and observer poles given by the observer polynomial
Ao. Hence
AR+BS =B +AoAm 8.14
Which is called the Diophantine equation. (Some authors prefer to call the Bezout
identity.) It follows from this equation that B+divides . Hence
R=B+R1 8.15
+
Dividing Eq. (5) by B gives
AR1+ B –S=AoAm 8.16
Now require that the relation in Eq. (8.13) between the command signal ucand the process
output y should be equal to the desired closed-loop response given by eq.(8.11). The

specifications must also be such that B divides Bm; otherwise there is no solution to the
design problem. Hence
Bm=B –B|m
T=AoB|m 8.17
To complete the solution of the problem we must give conditions to guarantee
that there exist solutions to Eq.(8.16) that give a proper (continuous time) or causal
(discrete-time) control law:
degAo≥ 2 deg A – deg Am– deg B+ –1 8.18
The control law of Eq. (8.12), with the controller polynomials given by Eqs.
(8.15),(8.16), and (8.17), gives perfect model-following if the compatibility conditions of
Eqs. (8.18) and (8.19) are fulfilled. Notice that the design above contains the solution of
the Diphantine equation (Eq. 8.16) and is thus not suited for a direct adaptive controller.
Assume, however, that all zeros are canceled Then
AoAmy=AR1+b0S 8.19
Multiply this by y and use the model equation of Eq. (8.10). This gives
AoAmy =BR1u +b0Sy
=b0(Ru + Sy) 8.20
The polynomials on the left-hand side are known, and those on the right-hand side
are the unknown controller parameters. The T polynomial is obtained directly from Eq.
(8.17). The reparameterized model of Eq. (8.20) can now be used to estimate the
unknown regulator parameters using the methods Least squares and regression models.
This leads to a direct MRAS.
8.3 The Gradient Approach
The gradient approach to model-reference adaptive control is developed in this

section. This is a fundamental idea in the MRAS approach. The parameter adjustment
scheme is usually called the MIT rule because the work was done at the Instrumentation
Laboratory (now the Draper Laboratory) at MIT [171]. A discussion of a particular
example illustrates the idea and reveals a fundamental difficulty. Modified adjustment
rules that avoid the difficulty are then given.

The MIT Rule

Assume that we attempt to change the parameters of the regulator so that the error
between the outputs of process and the reference model is driven to zero. Let e denote the
error and θ the parameters. Introduce the criterion
J(θ)= e2 8.21
To make J small it is reasonable to change the parameters in the direction of the negative
gradient of J, i.e.,
8.22
Other variables in the system, then the derivative can be evaluated under the
assumption that θis constant. The derivative is the sensitivity derivative of the system.
The adjustment rule of Eq. (8.22), where is the sensitivity derivative, is commonly
referred to as the MIT rule.
J(θ) = 8.23
The adjustment rule becomes
8.24
This is actually the way that the first MRAS was implemented. An even simpler
implementation is obtained by using the update rule
This is called the sing-sign algorithm. A discrete-time version of this algorithms

is used in telecommunications, where simple implementation and fast computations are
required.
Adaptation of a feedforward gain

Consider the problem of adjusting a feedforward gain, Let the model and the
process have the transfer function G(s). The error is
e = y –ym =G(P)θuc– G(p)θ0uc

where uc is the command signal, ym the model output, y the process output, θ the
adjustable parameter, and p= the differentiation operator. The sensitivity derivative is
The MIT rule then gives
If the sign of the parameter θ0 is known, the parameter can be introduced.
The rate of change of the parameter should thus be made proportional to the product of
the error and the model output.
Notice that no approximations were needed. When the MIT rule is applied to
more complicated problems, it is necessary to use approximations to obtain the sensitivity
derivatives.
Fig 8.4 Waveform of a First order system

TCSC DESIGN
The capacitor and inductor design is depending on the following equation
8.25
XTCSC - TCSC reactance XL - Capacitive reactance XC – Inductive reactance α - firing

angle
So if the XL = XC as per the equation XTCSC = infinitive so then no power is
transferred through it.
The design of inductor and capacitor will be decided by the XL/XC .

XL
= 0.133
XC Is the nominal value of the reactance ratio . But depending on the
XL
application we can choose the value of X C value 0.1 to 0.3 .[51]
We have chosen the value of 0.133.
So,
XL
= 0.133
XC
XL = 2пfL
1
XC =
2πfC
The value of the C is assumed to be 22uF
we get
XC = 120.59 Ω and
XL
= 0.133
applying it to X C
We get XL = 16.211Ω
from XL = 2пfL
We get the value of L= 0.043 H

XL
the ratio should be within 0.1< < 0.3
XC
So,
XL > 16.22 ohms
So,
Reference impedance is chosen for Inductive mode is 19 < Zref < 60 ohms
similarly Xc > 120 ohm in capacitive mode so we are choosing 120 < Zref <136 ohms
Fig 8.5 General simulink model of MRAS

One of the important features of MRAS is the adjustment mechanism which
receives the input from the model as well as from the controller. The main aim of the
adjust mechanism is to process the error signal in coordination with the model reference
which will be developed which behaves similar to the original systems. Any controller
types such PI or Fuzzy logic controller can be used as the adjustment mechanism. Here in
this research work PI controller along with Artificial Neural Network has been used as
the mechanism.
This research work carried out has a main aim to enhance the stability of the
power system by using adaptive controller strategies to an optimally located Flexible AC
Transmission System (FACTS) devices using evolutionary technique. The MRAS
comprises of control mechanism to modify the impedance of the TCSC under
consideration wherein the impedance value is received as an output of the error block
after comparison with actual and the model from the feed back path. The control
mechanism which is a part of MRAS used in this work is Artificial neural network.
This research work carried out has a main aim to enhance the stability of the
power system by using adaptive controller strategies to an optimally located Flexible AC
Transmission System (FACTS) devices using evolutionary technique. The system
performance could be enhanced further by improving its stability. The stability can be
improved by using controllers. After optimally locating a FACTS device the system is
applied with conventional controllers like PI , Fuzzy logic controllers and then the system
is tested for adaptive controller like MRAS based FACTS controller. The MRAS
comprises of control mechanism to modify the impedance of the TCSC under
consideration wherein the impedance value is received as an output of the error block
after comparison with actual and the model from the feed back path. The control
mechanism which is a part of MRAS used in this work is Artificial neural network.
The adaptive controller is a controller that can modify its behavior in response to
changes in the dynamics of the process and in the characteristics of disturbances.
Adaptive controllers also have their own parameters, which must be chosen. Controllers
without any externally adjusted parameters can be designed for specific applications in
which the purpose of control can be stated a priory, auto pilots for missiles and ships are
typical examples. In general an optimal control sign requires a nominal system model, ,
which represents the current operating point of the system. The implementation of
optimal control with a non-adaptive system model cannot be successful for a nonlinear
system without continuous support. In this work adaptive controllers like MRAS based
FACTS controllers have been included in the system to enhance the stability. Artificial
neural network based MRAS has been used to improve the stability as a part of
adjustment mechanism in the MRAS for controlling the firing angle of the TCSC.

The following cases have been considered to analyze the behavior of the system
considered:
CASE (1) MRAS along with PI controller in inductive mode

CASE (2) MRAS along with PI controller in inductive mode and ANN model
CASE (3) MRAS along with PI controller in capacitive mode
CASE (4) MRAS along with PI controller in capacitive mode and ANN model
For all the waveforms in the above four cases X axis time in seconds, y axis- Top power in
MW, middle Zreference yellow, Zmeasured pink in ohms, bottom firing angle in degrees .
CASE (1) MRAS adaptive controller along with PI controller in inductive mode
As seen from the above derivation it can be observed that the TCSC impedance
operates in two modes. The inductive and capacitive mode. In this research the TCSC
impedance is taken as the parameter which can be adjusted that is an adjustable parameter
so that the system will be close to the prescribed model. This is called the model
following problem. The reference model is the impedance reference model. It can be
observed from the Literature that the TCSC impedance is not chosen as a controllable
parameter in the MRAS. The model also simulates fast changes in the power system.
The changes in the power system which affect the transmission line parameters are the
sudden change in load, closure of circuit breaker, or a fault. Whenever there is a change
in voltage or current in the line, it affects the measured impedance of the transmission
line. The measured impedance which is also the TCSC impedance is compared with the
reference model continuously. The error is the difference between the reference
impedance and the measured impedance and the command signal is the command to the
SCR in the TCSC which is proportional to the error. A command signal which is
proportional to the error is applied and the angle of the triggering pulse changes
according to predetermined look up table given in the SIMULINK model. The TCSC
impedance or the measured impedance changes and is again compared with the reference
model and the process continues till 5 seconds which is the time chosen throughout the
research. The impedance for inductive mode is Reference impedance is chosen for

Inductive mode is 19 ohms < Zref < 60 ohms. In this research the range of impedance
chosen for inductive mode is 20ohms to 55 ohms. The variation is for [0, 1.5, 2.5, 3.5, 4.5
] seconds , [20 25 35 45 55] ohms respectively. The variation in the reference impedance
simulates the variations in the parameters of the transmission lines during faulty
situations. As per the MRAS principle the system is required to adapt itself for the
variations in the reference impedance which is also the reference model in the Model
reference adaptive system. This is demonstrated using the SIMULINK model of the test
system with the MRAS controllers.
In this research the case (1) SIMULINK model used is as shown below.
Fig 8.6 Simulink model of MRAS based controller in inductive mode
The SIMULINK model consists of programmable voltage source , transmission

line , TCSC model , the control system model which consists of the MRAS , Z reference
model , firing unit, scopes and plotting medium (main variables). The detailed model is
as shown below:
Fig 8.7 Zreference look-up table Fig 8.8 Zref source block parameters
The above SIMULINK model is run for 5 seconds and the waveforms of power,
measured impedance and reference impedance and firing angle alpha are plotted against
time.


Fig 8.9 waveforms of MRAS based controller in inductive mode
Analysis of the result :

It can be observed from the waveforms that as and when the reference impedance
changes the system adapts itself to the changes in the selected parameter based on the
reference model. (As per the Model reference adaptive control.)
It can be observed from the waveform that when the reference model, reference
impedance changes, the measured impedance changes accordingly. The change in
measured impedance brings about change in the angle of SCR in the TCSC. When the
firing angle changes the TCSC impedance changes and this brings about change in the
power flow.
It can be observed that the variations are not smooth. The power even though
within the limits there are lots of fluctuations which may cause the system to become
unstable. The measured impedance also fluctuates. Even though the system is adapting
itself to the changes the adaption is not a smooth one rather the measured impedance
changes with fluctuations before it reaches the required value. The power flow is 181.4
MW. The fluctuations can be reduced by using ANN. It is explained in the next section.

8.4 A Basic Introduction to Neural Networks

The simplest definition of a neural network, more properly referred to as an
'artificial' neural network (ANN), is provided by the inventor of one of the first neuro
computers, Dr. Robert Hecht-Nielsen. He defines a neural network as:
"...a computing system made up of a number of simple, highly interconnected processing
elements, which process information by their dynamic state response to external inputs.
ANNs are processing devices (algorithms or actual hardware) that are loosely modeled
after the neuronal structure of the mamalian cerebral cortex but on much smaller scales.
A large ANN might have hundreds or thousands of processor units, whereas a mamalian
brain has billions of neurons with a corresponding increase in magnitude of their overall
interaction and emergent behavior. Although ANN researchers are generally not
concerned with whether their networks accurately resemble biological systems, some
have. For example, researchers have accurately simulated the function of the retina and
modeled the eye rather well.
8.4.1The Basics of Neural Networks
Neural neworks are typically organized in layers. Layers are made up of a number
of interconnected 'nodes' which contain an 'activation function'. Patterns are presented to
the network via the 'input layer', which communicates to one or more 'hidden layers'
where the actual processing is done via a system of weighted 'connections'. The hidden
layers then link to an 'output layer' where the answer is output as shown in the graphic
below.

Fig 8.10 Layers of ANN

Most ANNs contain some form of 'learning rule' which modifies the weights of
the connections according to the input patterns that it is presented with. In a sense, ANNs
learn by example as do their biological counterparts; a child learns to recognize dogs
from examples of dogs.
Although there are many different kinds of learning rules used by neural
networks, this demonstration is concerned only with one; the delta rule. The delta rule is
often utilized by the most common class of ANNs called 'backpropagational neural
networks' (BPNNs). Backpropagation is an abbreviation for the backwards propagation
of error.
With the delta rule, as with other types of backpropagation, 'learning' is a
supervised process that occurs with each cycle or 'epoch' (i.e. each time the network is
presented with a new input pattern) through a forward activation flow of outputs, and the
backwards error propagation of weight adjustments. More simply, when a neural network
is initially presented with a pattern it makes a random 'guess' as to what it might be. It
then sees how far its answer was from the actual one and makes an appropriate
adjustment to its connection weights. More graphically, the process looks something like
this:

Fig 8.11 Single Node example
Note also, that within each hidden layer node is a sigmoidal activation function which
polarizes network activity and helps it to stablize.
Backpropagation performs a gradient descent within the solution's vector space
towards a 'global minimum' along the steepest vector of the error surface. The global
minimum is that theoretical solution with the lowest possible error. The error surface
itself is a hyper paraboloid but is seldom 'smooth' as is depicted in the graphic below.
Indeed, in most problems, the solution space is quite irregular with numerous 'pits' and
'hills' which may cause the network to settle down in a 'local minimum' which is not the
best overall solution.
Fig.8.12 error space in ANN

Since the nature of the error space can not be known a priori, neural network
analysis often requires a large number of individual runs to determine the best solution.
Most learning rules have built-in mathematical terms to assist in this process which
control the 'speed' (Beta-coefficient) and the 'momentum' of the learning. The speed of
learning is actually the rate of convergence between the current solution and the global
minimum. Momentum helps the network to overcome obstacles (local minima) in the
error surface and settle down at or near the global minimum.
Once a neural network is 'trained' to a satisfactory level it may be used as an
analytical tool on other data. To do this, the user no longer specifies any training runs and
instead allows the network to work in forward propagation mode only. New inputs are
presented to the input pattern where they filter into and are processed by the middle
layers as though training were taking place, however, at this point the output is retained
and no back propagation occurs. The output of a forward propagation run is the predicted
model for the data which can then be used for further analysis and interpretation.
It is also possible to over-train a neural network, which means that the network
has been trained exactly to respond to only one type of input; which is much like rote
memorization. If this should happen then learning can no longer occur and the network is
referred to as having been "grand mothered" in neural network jargon. In real-world
applications this situation is not very useful since one would need a separate grand
mothered network for each new kind of input.
8.4.2 Difference between From Conventional Computing and ANN

To better understand artificial neural computing it is important to know first how
a conventional 'serial' computer and it's software process information. A serial computer
has a central processor that can address an array of memory locations where data and
instructions are stored. Computations are made by the processor reading an instruction as
well as any data the instruction requires from memory addresses, the instruction is then
executed and the results are saved in a specified memory location as required. In a serial
system (and a standard parallel one as well) the computational steps are deterministic,
sequential and logical, and the state of a given variable can be tracked from one operation
to another.

In comparison, ANNs are not sequential or necessarily deterministic. There are no

complex central processors, rather there are many simple ones which generally do
nothing more than take the weighted sum of their inputs from other processors. ANNs do
not execute programmed instructions; they respond in parallel (either simulated or actual)
to the pattern of inputs presented to it. There are also no separate memory addresses for
storing data. Instead, information is contained in the overall activation 'state' of the
network. 'Knowledge' is thus represented by the network itself, which is quite literally
more than the sum of its individual components.
8.4.3 Applications of Neural Networks
Neural networks are universal approximators, and they work best if the system we
are using them to model has a high tolerance to error. However they work very well for:
capturing associations or discovering regularities within a set of patterns;
where the volume, number of variables or diversity of the data is very great;
the relationships between variables are vaguely understood; or,
the relationships are difficult to describe adequately with conventional
approaches.
8.4.4 Limitations
There are many advantages and limitations to neural network analysis In

reference to backpropagational networks , there are some specific issues
Backpropagational neural networks (and many other types of networks) are in a sense the
ultimate 'black boxes'. Apart from defining the general architecture of a network and
perhaps initially seeding it with a random numbers, the user has no other role than to feed
it input and watch it train and await the output. In fact, it has been said that with
backpropagation, "you almost don't know what you're doing". Some software freely
available software packages (NevProp, bp, Mactivation) do allow the user to sample the
networks 'progress' at regular time intervals, but the learning itself progresses on its own.
The final product of this activity is a trained network that provides no equations or
coefficients defining a relationship (as in regression) beyond it's own internal
mathematics. The network 'IS' the final equation of the relationship.
Backpropagational networks also tend to be slower to train than other types of
networks and sometimes require thousands of epochs. If run on a truly parallel

computer system this issue is not really a problem, but if the BPNN is being
simulated on standard serial machine (i.e. a single SPARC, Mac or PC) training can
take some time. This is because the machines CPU must compute the function of each
node and connection separately, which can be problematic in very large networks
with a large amount of data. However, the speed of most current machines is such
that this is typically not much of an issue.
8.4.5 Advantages over Conventional Techniques
Depending on the nature of the application and the strength of the internal data
patterns we can generally expect a network to train quite well. This applies to problems
where the relationships may be quite dynamic or non-linear. ANNs provide an analytical
alternative to conventional techniques which are often limited by strict assumptions of
normality, linearity, variable independence etc. Because an ANN can capture many kinds
of relationships it allows the user to quickly and relatively easily model phenomena
which otherwise may have been very difficult or impossible to explain otherwise.
8.5 SIMULINK /MATLAB application of Neural network in

this research
Steps followed in ANN application in this research
Step 1 : Neural network can be simulated using MATLAB . It can be initiated by giving a
command called “nnstart” in the command window. Soon after which the following
window will be opened.

Step 2 : In that window any of the tool mentioned can be selected. The tools available are
(i) Fitting tool
(ii) Pattern recognition tool (iii) clustering tool (iv) Time series tool
In this research the Fitting tool is selected.
Step 3: After the selection of the appropriate tool the following window is opened.
Step 4: The tool gives the description about the tool and the neural network.
Then the next button is clicked and below window is displayed. In this window the Inputs
tab which has been selected has to be given as the input.
In this research input chosen is the comparator. (This is selected to reduce the transients
in dc link).

Step 5 : The output is taken from the output of PI controller.
Step 6 : After clicking the next button the following window is displayed. Here the
neurons need to be trained. The training system is taking 10 number of hidden neurons in
training. If we increase the neurons we get trained for the specified input and output.
Step 7: When the next button is clicked the following window opens up.
training button will be shown in the window. That shows the neural network is training
and then the training starts.
Vdc Voltage after applying Neural Network:

Step 8: This shows the how many number of iterations have been taken and the time
taken for training error on the training etc.
Case (2): Fig 8.13(a) and Fig 8.13(b) Simulink model of MRAS based controller with
ANN in inductive mode

Fig 8.13(a) Simulink model of the system with MRAS , PI controller and ANN
Fig 8.13(b) Model of the control system used with the ANN

Fig. 8.14 Waveforms of response for MRAS based controller with ANN
Analysis of the result :

The drawback of the MRAS with PI controller is that the regulator has limited
application as the system fails to respond for fast changes in the system parameter
considered that is the impedance of the transmission line. The fast changes in the
transmission line parameters occur when there is a fault, sudden change in the load,
sudden operation of circuit breakers etc. The above condition is simulated in the
SIMULINK as a periodical change in the reference impedance. The value of Z reference
chosen is 20 ohms, 25 ohms, 35 ohms, 45 ohms, 55 ohms for 0 seconds , 1.5 secs, 2.5
secs, 3.5 secs , 4.5 secs respectively
In the second case inorder to smoothen the waveform of the power and hence to reduce
the fluctuations the MRAS in inductive mode is applied with the ANN technique. In this
research the ann is used to train the impedance and hence to select the angle for triggering
the SCR in the TCSC.
The system considered is applied with an Artificial Neural network. The input to
NN block is the impedance of the system. In the first step the fitting tool is selected. The
fitting problems, will set the Neural network to map between a data set of numeric inputs
and a set of numeric targets.
The Neural network Fitting Tool will help to select data, create and train a network and
evaluate its performance using mean square error and regression analysis. A two layer
feed forward network with sigmoid hidden neurons and linear output neuron can fit
multi-dimensional mapping problems arbitrarily, given consistent data and enough
neurons in its hidden layer. The network will be trained with Levenberg-Marquardt back
propagation algorithm, unless there is not enough memory in which case scaled conjugate
gradient back propagation will be used.
The input selected is one among 601 sampled elements. Next the target need to be
selected.
The next step is validation and test data. In this step randomly 181 samples are
divided for training, validation, testing. Here 127 samples are selected for training, 27 for
validation and 27 for testing. The three kinds of sampling are used as follows:

Training: These are presented to the network during training and the network is adjusted
according to the error. In this research the error is the difference between reference and
measured impedance.
Validation: These are used to measure network generalization and to halt training when
generalization stops improving testing.
Testing: These have no effect on training and so provide an independent measure of
network performance during and after training.
After this the number of neurons is chosen. The neurons will be trained. In this
work the neurons are 10. The training automatically stops when generalization stops
improving as indicated by an increase in the mean square error of the validation samples.
The impedance of the system is compared with the reference impedance the error is
trained by splitting it into hidden 10 neurons. With the help of results obtained the firing
angle is suitably changed and the measured impedance is made almost equal to the
reference impedance. When the TCSC impedance changes due to the changes in the
firing angle the power flow voltage across the buses and the current flowing in the line
also changes corresponding to the TCSC impedance which controls the impedance of the
line. When the voltage and the current flow change, the power flow in the line also
changes. In this research the TCSC impedance is considered as the main parameter
throughout the analysis. As and when the impedance of the TCSC changes there is a
change in voltage, current and the power flow in the line. As seen by the waveforms the
following conclusion can be drawn:
(1) The reference waveform changes periodically
(2) The measured impedance changes according to the reference model. There is a
sharp shoot up of the measured impedance at 0.3 seconds due to switching effects.
After that there are no fluctuations but variations is observed as and when the ref.
impedance changes. This is an improvement when compared with the MRAS with
PI controller.
(3) The change in the measured impedance or the line impedance brings about change
in the power as seen from the waveform. But the power fluctuations is not
reduced . Hence system will not be stable.
The power flow is 181.2 MW.
Case (3):
In this case the for the same system considered the range of the reference model is
changed. The model considered in this section is capacitive range. That is the reference
impedance is varied from 120 ohms to 130 ohms. The TCSC impedance is tried to
maintain in this range only. The SIMULINK model is as shown below. The model used
for impedance calculation , angle calculation and the values of PI controller is also
shown.
Fig 8.15 Simulink model of MRAS based controller with ANN in inductive mode
Fig 8.16 Simulink model for Firing Angle calculation

Fig 8.17 Simulink model for impedance calculation
Fig 8.18 Impedance calculation (contd)
Fig 8.19 Function block parameters

Fig 8.20 waveforms of the response for MRAS in capacitive mode
Analysis of the result:
The output of the system under consideration for MRAS and PI based controller
in capacitive mode is as shown in the above fig.8.20. The following conclusions can be
drawn, initially due to switching action there are fast changes in the measured impedance.
Correspondingly the angle varies and hence the power flow varies. From the duration 0.5
seconds to 1.5 seconds the measured impedance is close to the reference impedance and
the error is less. Hence the angle is 90 degrees. At 1.5 seconds the reference impedance
changes to 122 ohms from initial value of 120 ohms. The measured impedance will not
change much and the error is more hence there is a drop in the angle and the angle
fluctuates around 85 degrees. At 2.5 seconds the reference impedance changes that is it
increases to 124 ohms. At 3 seconds it changes to 126 ohms, 3.5 seconds to 128 ohms
and at 4 seconds to 130 ohms and continued with that value till 5 seconds. At 2.5 seconds
both the measured impedance changes with lot of fluctuations and continues without
reaching a steady state. This is because the line becomes more capacitive and system fails
to respond. This is reflected on the angle and the power also. Hence it can be said that as
the line becomes more and more capacitive the MRAS loses its control and the MRAS
principle cannot be applied. The average power flow is 184.4 MW. The system is next
applied with a ANN, MRAS and PI controller in capacitive range.
CASE (4):
The system under consideration is applied with MRAS controller with ANN and
the system in capacitive mode. Fig 8.21 shows the SIMULINK model of the system
under consideration. It also shows the control system with Neural network , function
fitting block. The system is run for 5 seconds and the main variables are plotted as shown
as shown in the waveforms.
Fig 8.21 Simulink model of MRAS based controller with ANN in capacitive mode

Fig 8.22 ANN simulink model
Fig 8.23 Layers in ANN

Fig 8.24 waveform of the response for MRAS based controller with ANN in
capacitive mode
Analysis of the result:
It can be observed from the results plotted as waveforms that the capacitive range of
TCSC with MRAS and ANN makes the variation of the measured impedance a smooth
one. The reference impedance is varied as 0.5 seconds 120 ohms, 1.5 seconds 122 ohms,
2.5 seconds 124 ohms, 3 seconds 126 ohms, 3.5 seconds 128 ohms, 4.5 seconds 130
ohms. It can be seen that the change in ref. impedance is faster after 2.5 seconds. It is
observed from the waveform results of the measured impedance that at the beginning the
imp. Shoots up to 130 ohms at 0.1 seconds. But at 0.4 seconds it reaches 120 ohms same
as ref. imp. AS and when the ref. imp. Changes the meas. Imp. Also changes very
smoothly due to the presence of the ANN. The angle variations are also smooth and the
power flow is also smooth . The system is said to stable under the simulated situations.
The power flow is 194.2 MW which is the highest among all the controllers considered.

8.6 Discussion about the results

This chapter presents an effective comparison of various controllers used in the
system. First the system is applied with PI controller , which is a conventional controller.
The system is then tested for a intelligent controller that is the Fuzzy logic controller.
Adaptive controller called the Model Reference adaptive system is used. The TCSC
consists of an inductor and a capacitor as shown in figure. The range of TCSC impedance
is is chosen as per the limitation TCSC can be used in two modes like inductive mode
and capacitive mode. The figure 8.21 shows the PI controller based control of the TCSC
controller. The controller gives the above result as shown in waveforms in fig 8.24. The
first figure shows the output waveform of power flow in the system due to the response
of PI controller. The second one is the variation of impedance of the TCSC compared
with the reference impedance. Third one is variation of firing angle alpha , used for
firing the SCRs present in the TCSC . All the waveforms correspond to PI controller
applied to system after the location of TCSC. The PI controller is present in the feedback
path of the TCSC. As TCSC is located in series with the line and it brings in changes in
the impedance of the line, the impedance is taken as the reference. The reference
impedance is given as a step input in order to analyze the behavior of the system. This is
used to the simulate the changes in the system inorder to observe the behavior of the
system under changes. The yellow line shows the reference impedance and the pink line
shows the measured impedance of the TCSC. Initially due to the switching action there is
large fluctuations in the power. But as the reference impedance increases the measured
impedance or the ZTCSC increases. As soon as ZTCSCis included power flow is becomes
almost stable.. The power then settles to a constant value as as there is no change in the
impedance of the line. At 1.5 seconds there is a first change in the reference impedance.
The impedance level is increased. The power remains almost constant. That is for
changes the line impedance the PI controller changes the TCSC firing angle suitably in
accordance with the impedance changes.. For fast step changes in the reference
impedances it causes fluctuations in the firing angle of TCSC and as a result there is
fluctuation in the power flow due to the nature of PI controller It can also be shown that
for fast changes in the reference the PI controller does not suit well and hence an

improvisation is needed. Here we can observe the oscillation clearly as the pi controller
lacks the control when there is frequent change in the reference impedance. So the power
oscillates more and the measured impedance also oscillates. So total power injected
oscillates where we can‟t find the dynamic stability. So the PI controller lacks the control
in dynamic condition. Hence an improved method based on intelligent controller is used.
Here the same system is implemented with Fuzzy logic controller.
The system is also subjected to a three phase fault. The three phase fault is
considered as this is the most severe fault. The three phase fault is simulated at
2.25seconds.
Fuzzy based TCSC:

The fuzzy logic based TCSC tuning using Fuzzy logic controller is also
implemented to study the behavior of the system under intelligent controller. The Fuzzy
system used in this research uses minimum number of membership functions. The
membership functions used here have been chosen so as to limit the time taken to solve
that is 5 seconds. Actually the Fuzzy logic controller needs more membership functions
in the circuit which makes the network complex and occupies more memory. It takes
more time to solve the whole system when fuzzy inference system is used. When referred
to oscillations, there will be more oscillations before it settles down else we require more
number of member ship function to give good response as per the reference. So it is
complicated. If its complexity is acceptable then the Fuzzy gives good result. It can be
observed that there are many fluctuations before the system settles to a particular value.
Hence Fuzzy logic controller is not suited here for the time limitation considered.
Adaptive controller based TCSC:

An adaptive controller called Model reference adaptive controller system
[MRAS] is considered to show the behavior of the system under adaptive conditions.
Adaptive Control covers a set of techniques which provide a systematic approach for
automatic adjustment of controllers in real time, in order to achieve or to maintain a
desired level of control system performance when the parameters of the plant dynamic
model are unknown and/or change in time.

MRAS inductive range of impedance

Here the impedance is taken as the reference. That is because when TCSC is
included in the transmission line parameter changed is the impedance of the line. The
impedance of the line is taken as the reference model. The impedance is changed in the
range from 19 to 60. As we know the range of inductance of the TCSC should be
between 19 < Zref <60. The impedance is varied as 20 25 35 45 55 for the durations 0 1.5
2.5 3 3.5 seconds. It can be observed from the waveforms that as and when the reference
impedance varies the measured impedance of the line also varies which in turn varies the
power and the firing angle of the TCSC. It can be observed that the MRAS based
controller restricts the variation of the measured impedance within the range limits. The
measured impedance is calculated on line and the Zmeasured is compared with
Zreference (specified as the reference model) at each and every instant . Based on the
comparisons the measured impedance is kept within the limits. When there is variations
in the measured impedance the power varies . The firing angle of the SCR used in the
TCSC changes corresponding to the impedance variations thus restricting the power
within the limits.
MRAS capacitive range of impedance:
MRAS based TCSC:

The power(1), impedance(2), and alpha is presented in the diagram. Here the
performance compared to PI controller over MRAS based neural network can work
perfectly with the same oscillations created as in the PI controller. So it is visible that the
oscillations due to the reference impedance is not making heavy oscillations as in the PI
controller the trained neural network is used here to do the same operation of the PI
controller. The waveforms are given above.
The above results show the change in impedance doesn't cause any fluctuations compared
to PI controller the MRAS based ANN is the best because of best control over stability.

Summary of the results:
Table 8.1Comparison table of all the results obtained
Without With PI With Fuzzy MRAS

TCSC controller controller Inductive Inductive Cap PI Cap ANN
mode (PI) mode (ANN)
% Z error max Not 8% 7% 7.7% 30% variations 30%
applicable
% Zerror Min 1.5% 1.5% .5% 1.1% 1.2%
Power flow 171MW 189MW 188MW 187MW 187 MW 171MW 190.2
(with (with (with (fluctuations) MW(smooth)
oscillations) oscillations) oscillations)
Steady state 4.8 secs 4.15seconds Not reached 4.6 secs > 5 secs > 5 secs 4.1 secs
time within 5
seconds
Firing Min Not 78 deg oscillates 0 deg 0 deg. 35 deg. 0.5 deg
Angle applicable o from 0o to
Max 90 40 deg. 90 deg. 90 deg. 90 deg
variation 89o from 0 to
4.5 secs
8.7 Implementation of MRAS on an Actual System
:
Fig 8.25 Single line diagram of actual 14 bus part of Indian Power Network

8.8 Voltage Stability

A system experiences a state of voltage instability when there is a progressive or
uncontrollable drop in voltage magnitude after a disturbance, increase in load demand or
change in operating condition. The main factor, which causes these unacceptable voltage
profiles, is the inability of the distribution system to meet the demand for reactive power.
Under normal operating conditions, the bus voltage magnitude (V) increases as Q
injected at the same bus is increased. However, when V of any one of the system‟s buses
decreases with the increase in Q for that same bus, the system is said to be unstable
.Although the voltage instability is a localized problem, its impact on the system can be
wide spread as it depends on the relationship between transmitted P, injected Q and
receiving end V. These relationships play an important role in the stability analysis and
can be displayed graphically.
8.8.1 Classification of Voltage Stability
The time span of a disturbance in a power system, causing a potential voltage

instability problem, can be classified into short-term and long-term. The corresponding
voltage stability dynamics is called short term and long-term dynamics respectively
.Automatic voltage regulators, excitation systems, turbine and governor dynamics fall in
this short-term or „transient‟ time scale, which is typically a few seconds. Induction
motors, electronically operated loads and HVDC interconnections also fall in this
category. If the system is stable, short-term disturbance dies out and the system enters a
slow long-term dynamics. Components operating in the long-term time frame are
transformer tap changers, limiters, boilers etc. Typically, this time frame is for a few
minutes to tens of minutes. A voltage stability problem in the long-term time frame is
mainly due to the large electrical distance between the generator and the load, and thus
depends on the detailed topology of the power system.
8.9 Voltage Profile Enhancement using MRAS based TCSC

For the 14 bus system a three phase fault was initiated on line 7-10 at bus no 7 at
0.5 seconds and the fault was removed at 0.8 second
8.9.1 Voltage Profile Before Placing TCSC
Fig. 8.26 Simulink Model Before Placing TCSC

(a)
(b)
0.5 1 1.5 2 2.5 secs

(c)
Fig 8.27 waveforms of voltages X axis time in seconds Y axis

(a)Voltage dip at buses 1,2,3,4,5
(b)Voltage dip at buses 6,7,8,9,10
(c)Voltage dip at buses 11,12,13,14

8.9.2 Voltage Profile After Placing TCSC
Fig. 8.28 Simulink Model After Placing TCSC

(a)
(b)
0.5 1.0 1.5 2 2.5 secs

(c)
Fig 8.29 waveforms of voltage x axis time in seconds(a)Voltage dip at buses
1,2,3,4,5 after placing TCSC
(b)Voltage dip at buses 6,7,8,9,10 after placing TCSC
(c)Voltage dip at buses 11,12,13,14 after placing TCSC

By comparing the wave forms of voltages at buses it is seen that with MRAS
based TCSC the voltage dip is considerably reduced and steady state is reached there is
considerable damping Comparison of total power flow after and before placement of
TCSC with controller in sensitive lines:
Table 8.2 Comparison of power flow after and before placement of TCSC with
controller in sensitive lines
Line No Total power flow before Total power flow after
placing TCSC in MW placing TCSC in MW
1-2 457 460.2

7-10 33.07 43.07
9-11 126.6 137.3
8.10 Implementation of MRAS on a 5 bus system
Fig: 8.30 MRAS for Stability Improvement Without TCSC Transmission Line

Without TCSC Transmission Line Result
Fig. 8.31 Load Voltage
Fig. 8.32 Load Current

Fig. 8.33 Power
Fig 8.34 Un-Regulated Load Voltage

Fig 8.35 Un-Regulated Load Current
Fig. 8.36 Un-Regulated load Power

Fig.8.37 Un-Regulated load Power (scale modified)
CASE II: 5 Bus Simulink System With TCSC and MRAS
Fig.8.38 MRAS for Stability Improvement With TCSC Transmission Line

TCSC Transmission Line Result
Fig.8.39 Load Voltage
Fig.8.40 Load Current

Fig.8.41 Power
Fig.8.42 Regulated Load Voltage
Fig.8.43 Regulated Load Current

Fig.8.44 Regulated Load Power
Fig.8.45 Regulated Load Power (scale modified)
8.11 Conclusion
This chapter presents the development of adaptive controllers for tuning SCR in
FACTS. It is seen that adaptive controller with ANN gives improved dynamic
performance. It is also seen that MRAS, capacitive mode with ANN gives the best result.
Hence it is suggested for implementation. The efficacy is tested on IEEE 5 bus system
and voltage stability enhancement is shown by implementing on a 14 bus part of Indian
power network.

Model Reference Adaptive Controller 8.1 Objective

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Model Reference Adaptive Controller 8.1 Objective

Uploaded by

Copyright:

Available Formats

Adaptive controller strategies for FACTS devices in a power system to enhance stability

8.2 Basics of Model reference adaptive controller

PET Research centre, PESCE , Mandya 197

Fig. 8.1 Model reference adaptive control schematic

PET Research centre, PESCE , Mandya 199

where δ is small positive number and ø(t) = ∑φφT.

B. Pole-shift control concept

PET Research centre, PESCE , Mandya 200

Fig. 8.2 Closed-loop system configuration.

PET Research centre, PESCE , Mandya 201

The value of α can be selected to satisfy some optimized performance indices.

8.2.1 The MRAS problem in general

Fig. 8.3 closed loop system with general linear controller.

PET Research centre, PESCE , Mandya 203

PET Research centre, PESCE , Mandya 204

8.3 The Gradient Approach

The gradient approach to model-reference adaptive control is developed in this

PET Research centre, PESCE , Mandya 205

The MIT Rule

referred to as the MIT rule.

This is called the sing-sign algorithm. A discrete-time version of this algorithms

Adaptation of a feedforward gain

PET Research centre, PESCE , Mandya 206

adjustable parameter, and p= the differentiation operator. The sensitivity derivative is

The MIT rule then gives

If the sign of the parameter θ0 is known, the parameter can be introduced.

Fig 8.4 Waveform of a First order system

PET Research centre, PESCE , Mandya 207

XTCSC - TCSC reactance XL - Capacitive reactance XC – Inductive reactance α - firing

The design of inductor and capacitor will be decided by the XL/XC .

PET Research centre, PESCE , Mandya 208

Fig 8.5 General simulink model of MRAS

PET Research centre, PESCE , Mandya 210

CASE (1) MRAS along with PI controller in inductive mode

PET Research centre, PESCE , Mandya 211

Fig 8.6 Simulink model of MRAS based controller in inductive mode

The SIMULINK model consists of programmable voltage source , transmission

PET Research centre, PESCE , Mandya 213

PET Research centre, PESCE , Mandya 214

Fig 8.9 waveforms of MRAS based controller in inductive mode

Analysis of the result :

PET Research centre, PESCE , Mandya 215

8.4 A Basic Introduction to Neural Networks

8.4.1The Basics of Neural Networks

PET Research centre, PESCE , Mandya 216

Fig 8.10 Layers of ANN

PET Research centre, PESCE , Mandya 217

Fig 8.11 Single Node example

Fig.8.12 error space in ANN

PET Research centre, PESCE , Mandya 218

8.4.2 Difference between From Conventional Computing and ANN

PET Research centre, PESCE , Mandya 219

In comparison, ANNs are not sequential or necessarily deterministic. There are no

8.4.3 Applications of Neural Networks

There are many advantages and limitations to neural network analysis In

PET Research centre, PESCE , Mandya 220

8.4.5 Advantages over Conventional Techniques

8.5 SIMULINK /MATLAB application of Neural network in

PET Research centre, PESCE , Mandya 221

PET Research centre, PESCE , Mandya 222

Step 5 : The output is taken from the output of PI controller.

PET Research centre, PESCE , Mandya 223

PET Research centre, PESCE , Mandya 224