Вы находитесь на странице: 1из 10

Course Name: Intelligent Control System

Course No.: EEE-6505


1. What is intelligent control system? State the different tools of intelligent control with
examples.

Ans: Intelligent Control System: Intelligent control is a class of control techniques that use various artificial
intelligence computing approaches like neural networks, Bayesian probability, fuzzy logic, machine learning,
evolutionary computation and genetic algorithms.

Tools for intelligent control: fuzzy controllers, neural networks and genetic algorithms.

Examples : The goal of this expository paper is to bring forth the basic current elements of soft computing (fuzzy logic,
neural networks, genetic algorithms and genetic programming) and the current applications in intelligent control. Fuzzy
sets and fuzzy logic and their applications to control systems have been documented. Other elements of soft computing,
such as neural networks and genetic algorithms, are also treated for the novice reader. Each topic will have a number of
relevant references of as many key contributors as possible.

2. What is hybrid intelligent control system (HICS)? Give the different combinations of HICS?

Ans: Hybrid intelligent system denotes a software system which employs, in parallel, a combination of
methods and techniques from artificial intelligence subfields as:

 Neuro-fuzzy systems
 hybrid connectionist-symbolic models
 Fuzzy expert systems
 Connectionist expert systems
 Evolutionary neural networks
 Genetic fuzzy systems
 Rough fuzzy hybridization
 Reinforcement learning with fuzzy, neural, or evolutionary methods as well as symbolic reasoning
methods.
3. Describe the applications of intelligent control in the industrial field.
4. Describe briefly the different types of genetic representations in a string or, individuals.
1.4 Genetic Representations
GENETIC ENCODING: A genetic representation, also known as genetic encoding, describes the elements of the genotype
and how these elements are mapped into a phenotype. A suitable genetic representation should be devised so that (a)
the recombination and mutation operators have a high likelihood of generating increasingly better individuals, and (b)
the set of all possible genotypes have a high likelihood of covering the space of optimal solutions for the problem at
hand.
Therefore, the choice of a genetic representation can benefit from knowing some properties of the search space. For
example, if one intends to evolve a digital electronic circuit, the genetic representation may use a discrete alphabet
that has some correspondence to the components of the circuit so that mutations of the genotype are more likely to
map into meaningful phenotypes. Instead, if one intends to evolve an analog circuit, the genetic representation may
include some real-valued elements to describe the parameters of components such as resistors and capacitors, or at
least allocate more characters to genes that describe analog components at a sufficiently fine granularity.
Since artificial evolution is often used for problems that are ill-defined or poorly understood, the choice of a suitable
genetic representation is not a simple affair. In this section we describe some common representations that rely on a
one-to-one correspondence between the genotype and phenotype space. These representations do not include gene
regulation dynamics.
1.4.1 Discrete Representations
BINARY REPRESENTATION : The individual is described by a sequence of l discrete values drawn from an alphabet with
cardinality k. For example, genetic algorithms (Holland 1975; Goldberg 1989), which are a particular class of evolutionary
algorithms, often resort to a binary alphabet 0, 1 with cardinality k = 2. In a few cases, this binary representation can be
directly interpreted as a phenotype, such as in the description of the configuration string of field programmable gate
arrays (a specific type of reconfigurable digital electronic circuits) that will be described later.
1.4.2 Real-Valued Representations
The genotype consists of a set of n numbers belonging to the domain of real numbers, typically represented as floating-
point values. This representation

is suitable for solutions that require high-precision parameter optimization, such as in the case of the description of a
wing profile. In this case each number will represent the value of a parameter describing the wing curvature. If we know
that the problem does not require a high resolution of the parameter space, we may well use a binary representation
and allocate a suitable number of bits for each parameter value, as we have seen above. However, the two types of
representations require quite different genetic operators, as we will see later in this chapter.

1.4.3 Tree-Based Representations


These representations are suitable for describing hierarchical structures with branching points and conditions. They are
notably used in genetic programming, a particular class of evolutionary algorithms used for evolving computer
programs, but are also applied to the description of electronic circuits, construction procedures, and experiment
planning, to mention a few.
FUNCTIONS TERMINALS: A tree-based representation is composed of a finite set of functions and of a finite set of
terminals. The choice of these two sets depends on the problem to be solved and on some prior knowledge of the
solution space.
CLOSURE: Closure means that all the functions should accept any element of the terminal set and any value returned by
the functions in the function set.
SUFFICIENCY: Sufficiency means that the choice of functions and terminals should allow the generation of programs that
represent the solution to the problem.

5. Describe the fitness function of an artificial evolutionary system (P-22-23).


Fitness Functions:
MULTIPLE OBJECTIVES: Fitness functions often attempt to optimize multiple objectives of the problem at hand. For
example, one may wish to evolve the design of an airplane wing that maximizes lift, minimizes drag, and is composed of
the smallest number of pieces.
FITNESS EVALUATION: Evaluating the fitness of individuals is often the most time-consuming part of an evolutionary
algorithm The quality of the evolved solutions depends on how exhaustive the evaluation of individuals has been.
6. Describe different types of selection procedure of individuals in the artificial evolutionary system (P-23-26).

Selection and Reproduction:


SELECTION PRESSURE: The selection pressure indicates the percentage of individuals that will create offspring for the
next generation.

Figure 1.10 Roulette wheel selection. Left: Roulette wheel representation of proportionate selection. Each slot
corresponds to an individual and the size of the slot is proportional to the reproduction probability of the individual.
Center: All individuals in the population obtain similar fitness and therefore there is little chance that the best individual
will make more offspring than the other individuals. Right: One individual obtains much higher fitness than all other
individuals and therefore almost all individuals in the next generations will be copies of itself.

PROPORTIONATE SELECTION:
In proportionate selection, assuming that the fitness is non-negative, the probability p(i) that an individual i makes a
copy of its own genome is given by the ratio between its own fitness value f(i) and the sum of the fitness values
of all individuals in the population

ROULETTE WHEEL: One way of visualizing this process is to think of a roulette wheel where each slot corresponds to one
individual in the population and the size of a slot is directly proportional to the reproduction probability p(i) of that
individual (figure 1.10, left).

RANK-BASED SELECTION: Rank-based selection does not suffer from the above-mentioned problems because it is based
on the rank of the individual in the population instead of its absolute fitness.

GENERATIONAL REPLACEMENT: In generational replacement, by far the most frequently used, the newly produced
offspring replace the entire old population of individuals.
7. Describe briefly different types genetic operators (i.e., crossover and mutation) (P-26-29).

Genetic Operators: Genetic operators capture the effects of biological mutations on the genotype. In this section we
will describe only a subset of genetic operators that are applicable to the frequently used genetic representations
described earlier. These operators are designed to modify genotypes of fixed length that include only coding regions.
Other types of genetic operators that can modify the length of genetic strings, such as deletion, insertion, and
duplication, are used for more advanced representations tailored to evolve specific problems. We will see an example of
those representations when we describe the evolution of electronic analog circuits.
Genetic operators introduce diversity in the population and allow the exploration of novel solutions. The combination,
or crossover, of genetic material from two parents may, under some conditions, exploit useful genetic building blocks in
the two parents. Since genetic crossover is emphasized as an important source of evolution in genetic algorithms, we
will describe it separately from other types of mutations.

1)Crossover: Crossover makes sure that offspring inherit characteristics from parents by creating pairwise
recombinations of the genomes of selected individuals.
RECOMBINATION This operator is also known as recombination. The newly created offspring are randomly paired and
parts of their genotypes are swapped by the crossover operator with a probability pc. Crossover operators come in
different forms, which are tailored to the genetic representations. The idea behind genetic recombination is that some
of the resulting offspring may benefit from the synergistic effect that results from the combination of sub solutions
found by the two parents.
2) Mutation: Mutation operates at the level of the individual. Mutations are small random modifications of the
genotype that allow evolution to explore variations of existing solutions.

8. Different types of evolutionary measures, i.e., fitness landscape, fitness graph, population diversity, neutral paths,
All-Possible-Pairs-Diversity, Entropic Diversity (P-29-32).

FITNESS LANDSCAPE: The evolutionary search space is often described as a fitness landscape, which can be visualized as
a multidimensional surface obtained by associating a fitness value to all possible individuals that can be obtained from
the genetic representation (figure 1.13). Since it is impossible to sample all possible individuals for any realistic problem,
in practice one considers a few hundred or thousand genotypes generated by randomly sampling the genetic
representation. For the sake of visualization, often fitness landscapes are collapsed to a two-dimensional graph where
all sampled individuals are lined on the same axis according to some ordering criterion (for example, string distance
or sampling order). If most of the fitness values are equal or zero, the evolvability of the system is very low. In this case,
it is advisable to use large populations to change the genetic representation.
FITNESS GRAPH: The fitness graph instead is a visualization of performance of an evolutionary algorithm across
generations (figure 1.14). Researchers typically plot the average fitness of the population and the fitness of the best
individual at every generation. Since artificial evolution builds upon a restricted pool of individuals, multiple runs with
different initialization of the population are necessary to draw any solid conclusion about the quality of the final
solution.

POPULATION DIVERSITY:

NEUTRAL PATHS: A neural pathway that serves to connect relatively distant areas of the brain or nervous system is a
bundle of neurons, known collectively as white matter.

ALL-POSSIBLE-PAIRS DIVERSITY:
ENTROPIC DIVERSITY: An alternative measure of diversity that is especially useful for nonbinary genetic alphabets is the
entropic diversity:

9. Different types of evolutionary algorithms (P-33-34).

10. What is evolutionary electronics? Define the circuit topology, circuit sizing, and routing of a circuit.
Solved

11. What are the limitations of conventional circuit designing?

Evolutionary electronics has the potential for being one of those alternative approaches because it appears ideally
suited to contexts where the limitations of the conventional approach are most apparent, such as
• design problems where the circuit specifications are given in terms that are difficult to formalize in the way required
by conventional design practices (for example, a global measure of performance) but are naturally expressed as an
evolutionary problem;
• problems where systematic design techniques are scarce or missing altogether and the progress in the generation of
satisfying solutions is trusted to the expensive and possibly uncertain insight of the human designer;
• design problems where existing systematic design techniques manage the complexity of circuit design by imposing a
number of constraints that result in a waste of devices or performance.
12. Define briefly about evolutionary electronics? What are the advantages of evolutionary electronics over
conventional approach? (P-42-45)
Solved

13. What is the difference between evolved analog and digital circuits? (P-49-52, where Box 1.2 is not needed).

14. How the digital circuit is designed by evolutionary process? (P-62-63)

Digital circuit is designed by evolutionary process:


The setup of an evolutionary electronics experiment requires the specification of a genetic representation for the
circuits and that of a fitness function representing the goal of the experiment. Many genetic representations can be
devised for digital circuits and, in fact, most of the creativity of the evolutionary experimenter is typically expressed
here. In a first class of popular representations the genome is a list of blocks of fixed and predefined structure, each
defining the nature of a circuit component and its connections to the other components (see, for example, the genetic
representation described in figure 1.28). Another popular representation is used in the case of intrinsic evolution with
reconfigurable devices, and is based on the direct use of the string of configuration bits as the genome. Program-based
genetic representations such as genetic programming are another popular kind of representation and will be considered
in more detail below, in the discussion of analog evolutionary design.
Considering now the definition of the fitness function, if the design goals include a conventional specification of the
required combinational input/output function or that of the sequential input/state/output behavior, there is a natural
priority for the realization of those requirements. In this case the fitness function is typically defined in terms of a count
of the items of the required behavior that are correctly realized by the candidate solution. For example, for the
realization of a combinational circuit realizing a given truth table, the fitness could be the number of output entries of
the truth table that are correctly realized. In some cases it is advisable to include a mechanism of penalization of the
circuit structures that discourages the matching of almost all entries of the truth table with trivial functions that are
difficult to improve to an exact solution (Zebulum et al. 2002, p. 174). Once the priority objective of functionality has
been realized, other objectives such as the number of devices in the circuit, the input/output delay, the maximum
admissible clocking speed, the power consumption, and so on can be also taken into account as objectives of the
evolutionary process.
We consider now three examples of evolutionary digital design which illustrate some of the most common
circumstances in which the evolutionary approach represents a valid alternative to the conventional digital design
approach. In the first example the recourse to evolution is motivated by the fact that the description is not given in a
form equivalent to a truth table or to a state diagram required by conventional design. In the second example the
objective is the synthesis of circuits that realize logic functions often used as digital building blocks using fewer resources
than with conventional synthesis techniques. In the third example, the goal is to explore the potential of evolution by
letting it go beyond the limits of the conventional abstractions of digital design.

Вам также может понравиться