UNIT 1 : ARCHITECTURES - ANN

TWO MARK QUESTIONS:

1. What is Artificial Intelligence?

Artificial intelligence is a field of study that includes computational techniques for
performing tasks that apparently require intelligence when performed by humans. The
fundamental issues of artificial intelligence involve knowledge representation, search,
perception, and inference
2. What is meant by soft computing?
The term "soft computing" was introduced by Professor Lorfi Zadeh with the
objective of exploiting the tolerance for imprecision, uncertainty and partial truth to achieve
tractability, robustness, low solution cost and better rapport with reality.
3. .Differentiate hard computing and soft computing.

Sl.No Hard Computing Soft Computing

1 Precised models Approximate models
2 Symbolic logic reasoning Approximate reasoning
Traditional numeric modeling and Functional approximation and randomized
3
search search

4. What is the need for intelligent control / Soft computing?

i. Many real world problems are nonlinear dynamic in nature, hence mathematical
modeling is complex. To simplify this, intelligent control can be used.
ii. As the intelligence of software is more in intelligent control, the effort required for
the designer is reduced.
5. What is knowledge representation and what are its methods.
Knowledge representation refers to the data structure techniques and organizing
notations that are used in artificial intelligence (AI). These include semantic networks,
frames, logic, production rules, and conceptual graphs. Methods include Propositional
Calculus, Predicate Calculus, rule based system, symbolic reasoning system.
6. Write any four Intelligent or soft computing techniques.
 Artificial Neural Networks
 Fuzzy Logic Systems
 Genetic Algorithms
 Ant Colony Optimization
7. What is artificial neural network?
An artificial neural network (ANN) is an efficient information processing system
which resembles in characteristics with a biological neural network.
8. Draw the mathematical model of artificial neuron.

9. Draw the structure of McCulloch-Pitts neuron.

10. What are the types of learning in ANN?

 Supervised learning
 Unsupervised learning
 Reinforcement learning
11. What is supervised learning?
The learning here is performed with the help of a teacher. During training, the input
vector is presented to the network, which results in an output vector. This output vector is the
actual output vector. Then the actual output vector is compared with the desired (target)
output vector. If, there exists a difference between the two output vectors then an error signal
is generated by the network. This error signal is used for adjustment of weights until the
actual output matches the desired output (target).
12. Define unsupervised learning.
The learning here is performed without the help of a teacher. During training, a
sequence of input training vectors are provided to the neural network, but no target will
be specified. The net modifies the weights according to some training algorithm such
that most similar input vectors are assigned to the same output.
13. What is reinforcement learning?
The reinforcement learning is a form of supervised learning because the network
receives some feedback from its environment. However, the feedback obtained here is
only evaluative and not instructive .The external reinforcement signals are processed in
the critic signal generator and the obtained signals are sent to the ANN for adjustment of
weights properly so as to get better critic feedback in future. The reinforcement learning
is also called learning with a critic as opposed to learning with a teacher, which indicates
supervised learning.
14. Define perceptron.
Perceptron is a computer model or computerized machine devised to represent or
simulate the ability of the brain to recognize and discriminate.
15. What is back propagation network? What are the factors affecting its training?
Back propagation, an abbreviation for "backward propagation of errors", is a common
method of training artificial neural networks. Back propagation network is feed forward
multilayered neural network. The aim is to train the net to achieve a balance between the
net's ability to respond (memorization) and its ability to give reasonable responses to the
input that is similar but not identical to the one that is used in training (generalization).
16. What is an activation function?
The function to be applied over the net input of a neural network is called activation
function. The output of a network depends on the type of activation function.
17. What is learning rate? What is its function?
The learning rate is denoted by "α" It is used to control the amount of weight
adjustment at each step of training. The learning rate, ranging from 0 to 1, determines the
rate of learning at each time step.
18. Explain a multi-layer net.
A network that has more than one layer in its architecture is called multi-layer net.
This network will contain input layer, output layer and hidden layers.
19. Draw the structure of biological neuron.
Terminal Branches
Dendrites
of Axon

Axon
20. Give any 4 examples of activation function.
 Identity function

 Binary Step function

 Sigmoidal functions

 Ramp function

16 MARK QUESTIONS:

1. Explain Back propagation network with its architecture, training and testing
algorithms.
[Page: 64 to 73 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on architecture and flow chart)
2. Draw and Explain the perceptron with its learning algorithm
[Page: 49 to 56 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on architecture and flow chart)
3. Draw and Explain the Mculloh-Pitts neuron with its learning algorithm
[Page: 27 to 28 in Principles of soft computing by Sivanandam and Deepa]
4. Explain different types of learning in ANN.
[Page: 20 to 22 in Principles of soft computing by Sivanandam and Deepa]
5. One problem in Mculloh_Pitts & one problem in Perceptron.
[Page: 33 to 45 in Principles of soft computing by Sivanandam and Deepa]
(See all solved problems)
 UNIT 2 : NEURAL NETWORKS FOR CONTROL

TWO MARK QUESTIONS:

1. What is a feedback network?

If the output of the network is taken and fed back to modify the next process of the
network, then it is known as a feedback network.
2. Write any 4 applications of ANN.
 Load forecasting
 Weather prediction
 Medical diagnosis
 Speech recognition
3. What is Hopfield network?
The Hopfield network is an auto-associative fully interconnected single-layer
feedback network. It is also a symmetrically weighted network. When this is operated in
discrete line fashion it is called as discrete Hopfield network and its architecture as a
single-layer feedback network can be called as recurrent.
4. What is associative memory?
A neural net that is trained to associate a set of input vectors with a corresponding
set of output vectors is called associative memory.
5. What is auto associative memory?
A neural net that is trained to associate a set of input vectors with a corresponding
set of output vectors is called associative memory. If the desired output vector is same
as the input vector, the net is an auto associative memory.
6. What is hetro-associative memory?
A neural net that is trained to associate a set of input vectors with a corresponding
set of output vectors is called associative memory. If the desired output vector is
different from the input vector, the net is a hetro-associative memory.
7. What is the other name of Auto associative net?
The other name for auto associative net is Content addressable nature.
8. Justify why Artificial Neural Network is called adaptive system during training.
During training or learning phase the weight vectors are continuously updated
according to some learning rule till the target is achieved, hence this phase is called
adaptive.
16 MARK QUESTIONS:
1. Explain discrete time Hopfield networks.
[Page: 110 to 114 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on architecture and algorithm)
2. Explain Process identification using ANN.
Back propagation neural network was used for process identification. The
improvements in neural network identification are possible by building upon
established identification techniques. The back propagation neural network can learn a
functional mapping between input and output based on a set of training examples.
The determination of a functional relationship among variables is called
identification. Traditionally, regression techniques have been used to identify models
from process data. These process models can then be used by model predictive
controllers such as QDMC.
NEURAL NETWORKS:
A neural network consists of a large number of simple processors called
neurons. A typical neuron is shown in Fig.

The neuron has n inputs, X1,X2,...,Xn. These inputs can come from other units, or from
some external source. The output of the unit y for the neural networks is given by the
corresponding activation function used in the network.
The purpose for using the neural network is to obtain a mapping from a vector
X to a vector Y. The size of the input and output layers are fixed by the number of
components of X and Y, respectively. The user specifies the number of neurons N h in
the hidden layer. The accuracy of the mapping is relatively insensitive to N h over a
wide range of values provided that the number of neurons used is greater than the
minimum number needed to implement the mapping. For a network with a single
hidden layer, and a user-specified hidden layer size, the number of weighting terms in
the network is:
Nt=NXNh+NhNY+Nh+NY
Nt = the total number of weighting terms,
NX= the dimension of X,
NY = the dimension of Y, and
Nh= the size of the hidden layer.
The set of input and output vectors that the network is trained on is commonly
referred to as the training set.

IDENTIFICATION OF DYNAMIC PROCESS MODELS:

Model predictive control technology has been available for many years. The
use of dynamic process models allows the user to systematically design a control
system without resorting to ad hoc tuning methods.
Dynamic process model formulation:
The process model for a sampled data system can be formulated in its most
general form as:
yt = f(yt-1 , yt-2 , yt-3 ,….., ut-1 , ut-2 , ut-3 ,….., et , θt ,t) -------------------(1)
where,
f = specified functional relationship,
yt = process output at time t,
ut = measured process input at time t,
et = unmeasured process input at time t,
t = discrete time index (integer values only), and
θt = set of parameter values at time t.
The inputs and outputs can be single (scalar) or multivariable (vector) values.
The identification process for a given set of input-output can be described in three
steps:
1. Postulate a structure to the model (f).
2. Calculate the best estimate of θt.
3. Validate the resulting model.
The identification process is often iterative in nature. For example, the results
of the validation step may suggest a better structure f for describing the process.
Dynamic model identification work performed for identification is almost
always done assuming a simpler model structure than the previous mentioned
equation (1). It is usually assumed that the desired model is time-invariant and linear,
leading to a model of the form:
 ∞
y t =∑ gi ut−i +e t
i=1

where gi are the process parameters of the system

Traditional identification vs neural networks:
Dynamic model identification, using models such as output error (OE), auto
regressive moving average exogenius (ARMAX) and finite impulse response (FIR)
structures, is commonly based on several assumptions regarding the true process
system, the most important of which are linearity and time-invariance. OE and
ARMAX require the additional assumption that the process order can be identified,
and FIR assumes that the user can estimate the process settling time to a step input.
These assumptions allow a local experiment to be extrapolated globally to other
operating conditions. The validity of the extrapolation depends of course on the
validity of the underlying assumptions. Experience has shown that for a number of
process systems, these assumptions are reasonable. If the process is not time-
invariant, periodic re-identification of the process model may be required. If the
process is nonlinear with respect to operating conditions, a number of linear models
may be required to span the range of process operation.
The assumptions made in neural network identification are considerably less
than those made in classical identification. The user must define the process inputs
and outputs for both types of identification. For classical identification technology,
the user must specify a structure to the nature of the relationship between the inputs
and the outputs, while for neural network identification the user must only specify a
topology for the network that is sufficient to describe the input-output mapping. The
neural network does not try to either assume or determine the functional relationship
between the input and output variables. Instead, the network simply creates a mapping
from input to output based on the examples upon which it is trained.
The primary advantage of this neural network approach compared to classical
dynamic model identification is that we avoid specification of the model structure,
which can be very difficult. This is especially true when the true process structure is
not in the allowable model set for the identification method selected. OE, ARMAX
and FIR all require a carefully controlled experiment over a linear operating range of
the process. For nonlinear process, the data set must be separated into subsets where
the process is approximately linear. Neural networks allow input-output mappings to
be detected without any understanding of the functional relationship. The
disadvantage of this approach is the need to specify a set of inputs which span the
region of interest, since without knowledge of the true process structure the results of
 an experiment cannot be extrapolated outside the region where the experiment was
conducted.
In summary, neural network identification requires less user input in terms of
defining the structural relationship between the process inputs and outputs than
classical identification. However, the neural network identification experiment may
need to be much larger-perhaps by orders of magnitude-than the classical
identification experiment.
An example for the prediction of a dynamic process using back propagation
neural network using robust criterion is shown in below figure. The final prediction of
the network (dashed line) was close to the true system (solid line).

3. Explain Neuro controller for inverted pendulum.

 The inverted pendulum is a classic example of a naturally unstable system. Its
dynamics are basic to tasks involving the maintenance of balance, such as walking
and the control of rocket thrusters.
The inverted pendulum task involves a pendulum hinged to the top of a
wheeled cart that travels along a track, as shown in Fig.1.

The
cart and pendulum are constrained to move within the vertical plane. The state at time
t is specified by four real-valued variables:
 the angle between the pendulum and vertical and the angular velocity (θt & θt )
 the horizontal position and velocity of the cart (ht & ht )

The inverted pendulum system was simulated using the following equations of
motion.

where the units of θ, h, and time t are radians, meters, and seconds,
respectively, and where g is the acceleration due to gravity (9.8 m/sec2), F t the output
of the action network (±10 N), m c the mass of the cart (10 kg), m the mass of the
pendulum plus the cart (1.1 kg), and l the distance from the pivot to the pendulum’s
center of mass (0.5 m).
 The goal of the inverted pendulum task is to apply a sequence of right and left
forces of fixed magnitude to the cart such that the pendulum is balanced and the cart
does not hit the edge of the track. A zero-magnitude force is not permitted. Bounds on
the angle and on the cart’s horizontal position specify the states for which a failure
signal occurs. There is no unique solution-any trajectory through the state space that
does not result in a failure signal is acceptable. The only information regarding the
goal of the task is provided by a failure signal, which signals either the pendulum
falling past ±12 deg or the cart hitting the bounds of the track at ±2.4 m.

Solution Using Neural Networks:

The architecture of a network and the computations performed by each unit
specify a function from input to output vectors. The function is parameterized by the
numerical connection weights between units and on the inputs to the network. The
function is altered by a learning method that adjusts the values of the weights.
Learning can be based on several forms of evaluative feedback. Supervised
learning methods, the most commonly used in neural networks, require a training set
of data consisting of input vectors and corresponding desired output vectors. Such
methods cannot be applied to tasks for which the desired output of the network is not
known. The inverted pendulum, as we have defined it, is such a task: the correct
action for most states is not even well-defined, since many trajectories are possible
that indefinitely avoid failure.
If the desired output is not available, the performance of the network must be
evaluated indirectly by considering the effect of its output on the environment with
which the network interacts. Reinforcement learning methods can be applied when
this effect is measured by changes in an evaluation signal, or reinforcement.
The action network learns to select actions as a function of states. It consists of
a single unit having two possible outputs, one for each of the two allowable control
actions of pushing left or right on the cart with a fixed-magnitude force. The output of
the unit is probabilistic-the probability of generating each action depends on the
weighted sum of the unit’s inputs, i.e., the inner product of the input vector and the
unit’s weight vector.
Initial values of the weights are zero, making the two actions equally probable.
The action unit learns via a reinforcement learning method. It tries actions at random
and makes incremental adjustments to its weights, and, thus, to its action probabilities,
after receiving nonzero reinforcements. The only nonzero reinforcement present in the
inverted pendulum task is a failure signal. Learning good actions is extremely slow
when based on this rare and delayed signal.
A second mechanism is needed to allocate the blame for the failure among the
actions in the sequence leading to the failure. This mechanism is provided by the
evaluation network, which also consists of a single unit. The evaluation unit learns the
expected value of a discounted sum of future failure signals by means of a temporal-
difference (TD).
The output of the evaluation unit is the inner product of the input vector and
the unit's weight vector. Assuming the input vector is a representation of the inverted
pendulum's state and that the weights are developed by the TD method, the scalar
output of the evaluation unit provides a ranking of the states. The difference in the
unit's output on the transition from one state to another is used to judge the
effectiveness of the previous action. An increase in the evaluation signifies a
transition to a state having a weaker prediction of failure and that the probability of
the preceding action should be increased. Similarly, the probability of repeating an
action that precedes an evaluation decrease should be lowered. In this way, the change
in the evaluation network's output serves as reinforcement during the possibly long
periods between failures. However, the learned evaluation function is not always
helpful, particularly before much experience has been gained. The learning methods
for updating both the evaluation and action networks must deal with this uncertainty.
The control described here was motivated by a learning system called BOXES
which learned to control an inverted pendulum using a state representation of discrete
regions. The regions of the state space were formed by the intersections of six
intervals along the θ dimension and three intervals along the θ, h, and h dimensions,
making a total of 162 regions. The resulting networks are shown in Fig. 2. Each unit
receives the 162 binary input components, and the evaluation unit's output directs the
learning process for both units.
The TD method allows learning to occur continuously using the learned
evaluation function and differences in its output as reinforcement, rather than waiting
for further failures.
 UNIT 3 : FUZZY systems

TWO MARK QUESTIONS:

1. What is fuzzy logic?

Fuzzy logic is a form of multi-valued logic to deal with reasoning that is
approximate rather than precise.
2. What is the difference between fuzzy set and crisp set?

Sl.No. Fuzzy Set Crisp Set

elements of the set have a degree elements of the set definitely do belong
1
of membership in the set to the set
2 Natural language Précised Value
3. Define fuzzification and defuzzification.
Fuzzification is the process of conversion of crisp value to fuzzy value.
Defuzzification is the process of converting the output of the fuzzy rules into a scalar,
or non-fuzzy value.
4. Write any 4 properties of fuzzy set.
 Commutative
 Associative
 Distributive
 De-morgan
5. What is a membership function?
A membership function (MF) is a curve that defines how each point in the
input space is mapped to a membership value (or degree of membership) between 0
and 1.
6. Write few types of membership function.
 Triangular
 Trapezoidal
 Gaussian
 Gbell
7. What are the limitations of Fuzzy system?
 It is a rule based system rather than based on definite mathematical expression
Therefore solution is obtained using deductive reasoning which may not be
exact solution
 In order to not to miss a possible solution or to increase the accuracy, the
number of rules framed must be quite large making the working tiresome.
8. Mention any two applications of fuzzy logic.

 Washing machine

 Traffic control
 Medical diagnosis
 Speed control of machines
9. What is alpha or lambda cut set and what is cardinality of a Fuzzy set?
The set Aλ is a crisp set called lambda cut( or alpha cut) set of a fuzzy set A,
where Aλ = { x/ μA (x) ≥ λ } . This is a crisp set derived from its parent fuzzy set.
The cardinality of a fuzzy set A, the so-called SIGMA COUNT, is expressed as a
SUM of the values of the membership function of A, μA(x):cardA= μA(x1) + μA(x2) + …
μA(xn) = ΣμA(xi), for i=1..n
10. What are the different Fuzzy relation operations?
 AND – maximum membership value
 OR – minimum membership value
 NOT – ( 1- membership value)
11. Define a Fuzzy Cartesian product
Let A be a fuzzy set on universe X and B is a fuzzy set on universe Y . The
Cartesian product between A and B will result in a fuzzy relation such that
A x B = R C X x Y , where R has a membership μR (x,y) = min [ μA (x) , μB (y) ]
12. Define Fuzzy controller.
By a fuzzy logic controller (FLC) we mean a control law that is described by a
knowledge-based system consisting of IF...THEN rules with vague predicates and a
fuzzy logic inference mechanism.
13. What are the advantages of Fuzzy controller?
 Very robust and Can be easily modified
 Can use multiple inputs and outputs sources
 Much simpler than its predecessors (linear algebraic equations)
 Very quick and cheaper to implement
14. What is Fuzzy linqistic variable?
A linguistic variable is a variable whose values are words or sentences in a natural
or artificial language. Eg:Tall, High, Slow, Medium
15. Explain Centre of Gravity method of defuzzification.
 It finds the point where a vertical line would slice the aggregate set into two equal
masses. Mathematically this center of gravity (COG) can be expressed as:

16. Explain the difference between conventional

control system and Fuzzy control
The main difference between a conventional
control system and a fuzzy logic controlled system is not only in the type of logic
(Boolean or fuzzy) but in the inspiration. The former attempts to increase the efficiency
of control algorithms; the latter is based on the implementation of human understanding
and human thinking in control algorithms
17. What is rule base in FLC?
These rules are based on the knowledge and experience of a human expert within
that domain. Fuzzy rules are of the general form if antecedent (s) then consequent(s)
Eg: If Traffic is more Drive slowly
18. Define Fuzziness of Fuzzy set
Fuzziness can be defined as an intrinsic property of fuzzy sets as abstract
mathematical objects, without taking into account what kind of information a Fuzzy set
is representing.
19. What is a Fuzzy function?
The fuzzy function gives, for a certain value of the control variable, its
membership value (from 0 to 1) for the set defined by the shape and numerical
parameters.
20. What is ANFIS?
ANFIS is Adaptive Neuro-Fuzzy Inference System which is a combination of
Artificial neural networks and fuzzy logic systems.
21. What are different defuzzification methods?
 Max-membership principle
 Centroid method
 Weighted average method
 Mean–max membership
22. Draw the block diagram of fuzzy logic system.
23. What are the types (classification) of fuzzy logic controller?
Mamdani and Sugeno controllers.

16 MARK QUESTIONS:

1. Explain fuzzy logic control system with block diagram.

[Page: 375 to 377 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on architecture and explanation each block)
2. Explain the different defuzzification methods.
[Page: 313 to 319 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on types, diagram and formulae)
3. Explain the properties of fuzzy set.
[Page: 263 in Principles of soft computing by Sivanandam and Deepa]
(If asked for 16 marks try to prove each property with your own example)
4. One problem in fuzzy properties.
[Go through the problem solved in class, or pages: 264 to 270 in Principles of soft
computing by Sivanandam and Deepa]
5. Explain ANFIS architecture.
[Go through the attached notes for ANFIS architecture. Concentrate much on
architecture diagram and layear, formulae]
 UNIT 4 : application of FUZZY logic systems

TWO MARK QUESTIONS:

1. Write any 4 practical applications of fuzzy logic?

 Washing machine

 Traffic control
 Medical diagnosis
 Speed control of machines
2. A modern fully automatic washing machine is fixed with what type of intelligent
control? Define the input and output of the system
The washing machine will be fitted with Fuzzy logic intelligent control. The
inputs will be the weight of the dirty cloth, time to be taken for washing and the
output will be the amount of washing powder and water to be used.
3. Name the shapes of the membership function available in fuzzy logic tool box
 Trapezoidal
 Triangular
 Gaussian
 Sigmoidal
 L function
4. An intelligent door locking system in vehicles deploys what kind of control?
This will use the combination of Fuzzy logic and Neural Network because
they combine rule base as well as knowledge base.
5. Name any three commercial software used for soft computing techniques.
 MATLAB
 PSPICE
 ETAP
6. How stability of Fuzzy logic control is found?
In the stability analysis, the fuzzy controller is replaced with an appropriate
describing function which represents the nominal model of the fuzzy controller. The
Nyquist stability Criteria is then performed to determine the stability of the fuzzy
 control system. Additionally, this analysis can predict the existence of the limit
oscillation.
7. Write any two applications of fuzzy logic in electrical engineering.
 Control of Electrical machines
 Stability assessment of power systems.

16 MARK QUESTIONS:

1. Explain fuzzy logic application for liquid level control.

Consider the tank with liquid in it shown in Figure 1.

We want to design a controller that will either maintain the liquid level at a
desired point. Suppose that the tank in Figure 1 is 10 feet tall and the tank is empty.
We want to fill the tank to a level of 5 feet, so we make the current set-point, w,
equal to 5. The idea is to fill the tank to the desired set-point as quickly and smoothly
as possible. We want to minimize the amount of overshoot, or the time that the tank
has a level greater than the set-point value before it finally settles down. The current
level at any time, t, is designated as h. Liquid flows out of the tank through an open
valve. This flow is designated by the letter q. Liquid flows into the tank by means of
a pump. The pump flow, Q, can be regulated by the controller. The tank cross-
sectional area is designated by the letter A. Below equation describes the mass
balance for the liquid in the tank as a function of time:

Flow out of the tank, q, through the outlet pipe and the valve is described as,
where ɸ is a friction coefficient for flow through both the small exit pipe and the
valve. It can be calculated with fair precision, or better, measured. The term Ap
represents the cross-sectional area of the small exit pipe. The gravitational constant,
g, is equal to 32.2 ft s −2 in US engineering units. Figure 2 shows the block flow
diagram for this example.

Fuzzy Control:

A fuzzy control system connects input membership functions, functions

representing the input to the controller, e, to output membership functions that
represent the control action, u. For controlling the liquid level of the tank, we want to
design a controller that will allow us to change the set-point either up or down, and
one that will correct itself in the case of overshoot. A simple fuzzy control system
designed for our tank-level set point-tracking problem consists of three rules.
 If the Level Error is Positive Then the Change in Control Action is Positive.
 If the Level Error is Zero Then the Change in Control Action is Zero.
 If the Level Error is Negative Then the Change in Control Action is Negative.
 The input membership functions are shown in Figure 3. We should also notice
the “dead band” or “dead zone” in the membership function Zero between ±3 inches.
This is optional and is a feature commonly used with on–off controllers. It is easy to
implement with a fuzzy controller and is useful if the control engineer wishes to
minimize control response to small transient-level changes. This step can save wear
and tear on equipment.

The output membership functions for this controller are shown in Figure 4. In
this figure, the defuzzified output value from the controller is a fractional value
representing the required pump output for the desired level change. It is defined by
the following expression:

where the term Qsp represents the pump output (gallons per minute) required to
maintain the set-point level. The term Q i is the new pump output requested by the
controller. If Δu>0, then the Range is defined as Q max − Qsp, where Qmax is the
maximum pump output. If Δu < 0, then the Range is defined as Q sp. The term Qsp
must be calculated using a steady state mass balance for the tank or it must be
estimated in some fashion.
 The ranges of the fuzzy output sets Positive and Negative are +2.0 to 0.0 and
−2.0 to 0.0, respectively. Since the Change in Controller Action is a fraction
between either 0.0 and 1.0 or 0.0 and −1.0, it is clear that we will never obtain a
control action outside the range of −1.0 to 1.0. Our defuzzification technique will
require that we include numbers up to 2.0 in the fuzzy set or membership function
Positive and numbers down to −2.0 in the fuzzy set Negative. Even though numbers
of this magnitude can never be generated by our fuzzy system, we can still include
them in our fuzzy sets. The users can define their fuzzy sets however they wish. The
fuzzy mathematics described in earlier chapters is capable of handling objects of this
type. The user has to define the fuzzy sets so that they make sense for the particular
problem. In our case we are going to use the centroid technique for defuzzification.
We therefore need to extend our membership functions so that it is possible to obtain
centroids of ±1.0. We need this capability in order for the control system to either
turn the pump on “full blast” or turn it completely off.
2. Explain Fuzzy based motor control.
Fuzzy logic system can be used to perform the speed control of any electrical
motors. Here we discuss how a fuzzy logic system can be used to control the speed
of a DC shunt motor using field control method.
 In the field control method of speed control of DC shunt motor, the speed of
the motor can be varied by varying the field excitation. If there is any load change in
the motor, the line current varies and there will be a speed change. The purpose of
the fuzzy controller is to maintain a fairly constant speed of the motor as load is
changed. This speed control is accomplished by adjusting the field current. The input
variables to the fuzzy controller are therefore motor speed and field current, while
the field current is also an output variable. Fuzzy rules for this problem thus appear
in the form,
IF Speed is ………and field current (at that speed) is ……..
THEN field current (required to effect speed control) is …….
Consider a DC shunt motor with rated voltage of 220 V, having an armature
resistance of 0.2Ω, a rated field current of 2 A and a rated full load speed of 1500
rpm.
The field excitation of the motor and the speed error (difference between
actual speed and reference or set speed) may be taken as the input to the fuzzy
controller and the field excitation to be additionally given shall be taken as output of
fuzzy. The fuzzy linguistic variables for the inputs and the output can be created as
shown in the below tables.

Variable 1: Speed error

Fuzzy Variable Description Numerical Range
MN More negative -400 to -150 RPM
 N Negative -200 to -50 RPM
Z Zero -100 to 40 RPM
P Positive 20 to 100 RPM
MP More Positive 75 to 300 RPM

Variable 2: Field Current

Fuzzy Variable Description Numerical Range
S Small 1.2 – 1.3 A
BN Below Normal 1.25 – 1.4 A
N Normal 1.4 – 1.7 A
AN Above Normal 1.65 – 1.9 A
L Large 1.9 – 2 A

The fuzzy system is represented by a fuzzy associative memory (FAM)

mapping that associates the input variables to the output variable, as shown in the
below. The output variable is the field current required to restore the motor speed to
the normal range. The membership diagrams for both variables are given below.

Since there are 5 linguistic variables for each input, the total number of rules
will be 25. The input-output relation is described in the below table.

SPEED ERROR (Actual Speed-Reference Speed)

 MN N Z P MP

S L AN S AN L

FIELD CURRENT BN N BN BN AN AN

N AN AN N AN AN

AN N N AN N N

L BN BN L BN BN

An example rule from above table is given as follows.

If speed error is more negative(MN) and field current is small(S),
then field current to be applied is large(L)
similarly 25 rules need to be generated in the rule base of the fuzzy logic
system. Once the rules are generated, the fuzzy control system can be applied for the
control of motor. The block diagram representing the speed control of DC shunt
motor is give below.

3. Explain aircraft landing using fuzzy control.

[Page: 378 to 383 in Principles of soft computing by Sivanandam and Deepa]

(Concentrate much on diagrams, tables, membership diagrams)
4. Explain fuzzy PID control

Fuzzy controllers are inherently nonlinear controllers, and hence fuzzy control
technology can be viewed as a new, cost effective and practical way of developing
nonlinear controllers. The major advantage of this technology over the traditional
control technology is its capability of capturing and utilizing qualitative human
experience and knowledge in a quantitative manner through the use of fuzzy sets,
fuzzy rules and fuzzy logic. However, carrying out analytical analysis and design of
fuzzy control systems is difficult not only because the explicit structure of fuzzy
controllers is generally unknown, but also due to their inherent nonlinear and time-
varying nature.
Fuzzy logic controllers are normally built with three distinct components, as
shown in Fig. 1

Two processes in fuzzy control require heuristic rules and membership

functions to encode the desired system response characteristics and controller
dynamics. It is not clear that these fuzzy transformations should be based upon only
some basic understanding of the physical system. This is a significant problem in the
design of various fuzzy controllers, and is the basic justification for the reason of
using the well-known PID controller as the underlying structure for a new design.
Derivation of the nonlinear fuzzy PID controller:

A fuzzy controller has fuzzification, rule base, and defuzzification

components. The first problem is how to formulate the fuzzification process using
the common triangular and trapezoidal functions. The general PID controller
equation is given by,

represent the tracking error, change in error, and change in error rate,
respectively. Then the general PID controller equation can be re-written as,

The three error

terms are comparable to the input signals shown in Fig. 1.
 As presented in Fig. 2, the simplest input membership function used for e I, eP,
and eD in the fuzzification process are the same, and have two straight lines followed
by constant
hold
when the
values
exceed
some

predetermined threshold. The threshold parameter, L, is specified to define the

maximum and minimum values for the fuzzification process. For example, for an
input value greater than )L, the positive membership value function (designated by p)
starts at a small value, goes to a value of one half at 0, and finally goes to 1 when the
input value is greater that L. The negative membership function (designated by n) is
the opposite.

The output membership functions are not as simple since there are more possible
outcomes. In our case, there are also two crossing straight lines followed by constant
holds, but there are two extra triangular membership functions. These are shown in
Fig. 3. As before, the parameter, L, defines the minimum and maximum outputs, but
there are two new output terms centered at ±L/3, respectively.
 Given these membership functions, it is now possible to present the inference
composition rules. These can be expressed as follows:

where the ‘&’ symbol represents the fuzzy ``and'' operation, and the symbol
means ``is a member of '', and the symbol represents the combination. Note the
symbol represents the ``large'' value in the corresponding direction of either positive
(p) or negative (n).
 As there are three different input components in the control law corresponding
to the proportional, integral, and differential terms, it is necessary to view all the
possible combinations as a cube with a limiting edge value of L. Fig. 4 represents the
three-dimensional cube with the boundaries drawn at the value of L. the
defuzzification rules will be constructed by dividing the cube into 48 sectors with
known characteristics.
 This new fuzzy PID controller is will be able to control linear and nonlinear
plants effectively. The fuzzy PID controller is also able to tolerate many poor
selections or inadequate implementations of the controller gains which would make
most conventional controllers unstable. Since nonlinear effects will be encountered
in many complex systems, such as robotic manipulators, the ability of the fuzzy PID
controller to tolerate these un-modeled nonlinear and gain-value variation factors is a
substantial improvement over conventional linear PID controllers in real-world
applications.

UNIT 5 : genetic algorithms

TWO MARK QUESTIONS:

1. What is gradient search?

Gradient descent is a first-order iterative optimization algorithm. To find a
local minimum of a function using gradient descent, one takes steps proportional to
the negative of the gradient of the function at the current point.
2. Define genetic algorithms.
Genetic Algorithms are derivative-free stochastic optimization methods based
on the concepts of natural selection and evolutionary processes.
3. Define crossover and mutation in GA.
In genetic algorithms, crossover is a genetic operator used to vary the
programming of a chromosome or chromosomes from one generation to the next. It is
analogous to reproduction and biological crossover, upon which genetic algorithms
are based. Cross over is a process of taking more than one parent solutions and
producing a child solution from them.
Mutation is a genetic operator used to maintain genetic diversity from one
generation of a population of genetic algorithm chromosomes to the next. It is
analogous to biological mutation.
4. What are the types of crossover in GA?
 Single point crossover
 Two point crossover
 Uniform crossover
5. Define single point and two point crossover.
The traditional genetic algorithm uses single point crossover, where the two
mating chromosomes are cut once at corresponding points and the sections after the
cuts exchanged.
In two-point crossover, two crossover points are chosen and the contents between
these points are exchanged between two mated parents.
6. Define constrained and unconstrained optimization.
Constrained optimization is optimizing a function with the variables having
certain limits. Eg: Generator limit in Economic Dispatch
Unconstrained optimization is optimizing a function which allows the
variables to have any value.

7. Define uniform crossover.

Each gene in the offspring is created by copying the corresponding gene from
one or the other parent chosen according to a random generated binary crossover
mask of the same length as the chromosomes. Where there is a 1 in the crossover
mask, the gene is copied from the first parent, and where there is a 0 in the mask the
gene is copied from the second parent. A new crossover mask is randomly generated
for each pair of parents. Offsprings, therefore contain a mixture of genes from each
parent. The number of effective crossing point is not fixed, but will average L/2.
8. What are the different selection methods in GA?
 Roulette wheel selection
  Tournament selection
 Rank based selection
 Greedy selection
9. Define population in GA.
A population is a collection of individuals. A population consists of a number of
individuals being tested, the phenotype parameters defining the individuals and some
information about search space.
10. Write any two applications of GA in power systems.
 Economic Dispatch
 Optimal Power Flow
 Unit Commitment
11. What are the characteristics of GA.
 Derivative freeness
 Flexible
 Stochastic
 Iterative nature
12. What are the components of GA?
 Chromosomal Representation
 Initial Population
 Fitness Evaluation
 Selection
 Crossover and Mutation
13. Draw the basic GA cycle.

14. Write the

pseudo code
for simple GA.
{
initialize population;
 evaluate population;
while TerminationCriteriaNotSatisfied
{
select parents for reproduction;
perform recombination and mutation;
evaluate population;
}
}
15. What is Roulette wheel selection
Roulette selection is one of the traditional GA selection techniques . The
principle of roulette selection is a linear search through a roulette wheel with the slots
in the wheel weighted in proportion to the individual’s fitness values. A target value is
set, which is a random proportion of the sum of the fit nesses in the population. The
population is stepped through until the target value is reached.
16. What are the reproduction operators in GA?
 Crossover
 Mutation
17. Define Objective function in an optimization problem.
Objective function is an equation to be optimized given certain constraints and
with variables that need to be minimized or maximized using nonlinear programming
techniques.

18. Mention the role of fitness function in GA and what are the requirements of GA
Genetic Algorithm works on the principle of survival of fittest. A fitness
function quantifies the optimality of a solution (chromosome) so that that particular
solution may be ranked against all the other solutions.
19. The popularity of GA is attributed by which factors?
 Concept is easy to understand
 Supports multi-objective optimization
 Good for “noisy” environments
 Inherently parallel
 Easily distributed
20. What are the advantages of GA over conventional algorithms?
 Parallelism
 Liability
 Solution space is wider
  The fitness landscape is complex
 Easy to discover global optimum
 The problem has multi objective function
 Only uses function evaluations.
 Easily modified for different problems.
 Handles noisy functions well.
21. What are the limitations of GA? When does one go for GA?
The limitations of Genetic Algorithm are,
 The problem of identifying fitness function
 Definition of representation for the problem
 Premature convergence occurs
 Cannot use gradients.

16 MARK QUESTIONS:

1. Explain the process involved in genetic algorithm.

[Page: 402(sec.5.8) to 405 in Principles of soft computing by Sivanandam and Deepa]
(Concentrate much on flow chart and explanation)
2. One problem in GA
[Page: 418 to 422 in Principles of soft computing by Sivanandam and Deepa
Understand mathematical steps and you should be able to solve any problem
Similar problem is solved in class also]

3. Explain various selection and crossover methods in GA.

[Page: 407 to 414 in Principles of soft computing by Sivanandam and Deepa]
4. Explain GA application to Economic dispatch problem.
 The main objective of the economic dispatch problem is to minimize the total
power generation cost of the system. If C i is considered as the fuel cost of i th unit and
‘N’ is the total number of generating units, then the objective function is given as,

(∑ )
N
Objective Function F ,=min C i ( P gi ) ⁡
i=1

This objective function is subject to the following constraints.

 Equality constraint: load generation balance ie., the total generation must
be equal to the total load demand neglecting the losses.
 Inequality constraints: upper and lower limits on generating units output
Consider the below example system having 3 units and the total demand for
this system is 800 MW. The cost equations for the 3 units are given as,

The minimum and

maximum generating limits of each unit are 0 MW and 800 MW respectively.

Objective Function:

The objective function for this problem is given by,

( )
3
Objective Function F ,=min ∑ C i ( P gi ) ⁡
i=1

= min(C1+C2+C3)

C1+C2+C3 = 1100+ 5.3*P(1) + 5.5*P(2) + 5.8*P(3)+ 0.004*P(1)^2+ 0.006*P(2)^2+

0.009*P(3)^2
(this function is obtained by adding the generation cost
of all the 3 units of the given problem)

Constraints:
There are both equality and inequality constraints in the economic dispatch
problem.

Equality Constraint:

Equality constraint is the power balance equation which is given by,

 P1+P2+P3=800

Inequality Constraints:

The limits of generation given in problem are the inequality constraints. It is

given by,
0 MW ≤ P1 ≤ 800 MW
0 MW ≤ P2 ≤ 800 MW
0 MW ≤ P3 ≤ 800 MW
Genetic Operations:
 The above flow chart represents the basic GA cycle. This needs to be done for
the economic dispatch problem. Below is the step by step procedure.
Step 1:
Create fitness function / objective function and constraint functions for the economic
dispatch problem which are discussed already.
Step 2:
Assume initial population size as 20. Use Roulette selection which is one of
the traditional GA selection techniques. The principle of roulette selection is a linear
search through a roulette wheel with the slots in the wheel weighted in proportion to
the individual’s fitness values. A target value is set, which is a random proportion of
the sum of the fit nesses in the population. The population is stepped through until the
target value is reached.
Step 3:
Perform crossover and mutation.

Use single point cross over and perform bitwise mutation.

Step 4:
Evaluate the fitness function for the children produced. ie., find the value of total cost
for the obtained power values. Compare the cost value with previous iteration/
generation value.

Step 5:
Repeat the procedure/steps until the difference between previous generation cost
value and current generation objective function value becomes negligible. The
resultant will be the power scheduling considering economic dispatch.