Article On Soft Computing
An overview of soft computing
Pooja Gaikwad*
Abstract
Soft computing, as opposed to traditional computing, deals with approximate models and gives solutions to complex real-life
problems. Unlike hard computing, soft computing is tolerant of imprecision, uncertainty, partial truth, and approximations. In
effect, the role model for soft computing is the human mind. Soft computing is based on techniques such as fuzzy logic, genetic
algorithms, artificial neural networks, machine learning, and expert systems. Although soft computing theory and techniques
were first introduced in 1980s, it has now become a major research and study area in automatic control engineering. The
techniques of soft computing are nowadays being used successfully in many domestic, commercial, and industrial applications.
With the advent of the low-cost and very high performance digital processors and the reduction of the cost of memory chips it is
clear that the techniques and application areas of soft computing will continue to expand. This paper gives an overview of the
current state of soft computing techniques and describes the advantages and disadvantages of soft computing compared to
traditional hard computing techniques.
1. Introduction
One of the problems in traditional control systems is that complex plants cannot be accurately described by
mathematical models, and are therefore difficult to control using such existing methods. Soft computing on the other
hand deals with partial truth, uncertainty, and approximation to solve complex problems. Dr Zadeh who is the
pioneer of fuzzy logic quoted that “the guiding principle of soft computing is to exploit the tolerance for
imprecision, uncertainty, and partial truth to achieve tractability, robustness, low solution cost, better rapport with reality”. Because of its features such as intelligent control, nonlinear programming, optimization, and decision
making support, soft computing has become popular and has drawn research interest from people with different
backgrounds, Jang et al .
It is becoming difficult to control the growing complexity of modern machinery using traditional control systems
techniques. For example, many nonlinear and time-variant plants with large time delays cannot easily be controlled
and stabilized using traditional techniques. One of the reasons for this difficulty is the lack of an accurate model that
describes the plant. Soft computing is proving to be an efficient way of controlling such complex plants.
Zadeh pointed out that soft computing is not a single method, but instead it is a combination of several methods,
such as fuzzy logic, neural networks, and genetic algorithms. All these methods are not competitive, but are
complimentary to each other and can be used together to solve a given problem . It can be said that soft computing
aims to solve complex problems by exploiting the imprecision and uncertainty in decision making processes.
Fig. 1 shows the conventional and soft computing based problem solution principle as suggested by Gupta and
Kulkarni. The left diagram shows the traditional hard computing approach where an exact model of the plant under
investigation is available and traditional mathematical methods are used to solve the problem. The right diagram
shows soft computing approach where only an approximate model of the plant may be available, and the solution
depends upon approximate reasoning techniques.
Fuzzy control has been in use for over two decades to solve complex control problems, Driankov et al . In
addition, many instrumentation problems are being solved using fuzzy logic principles as reported by Russo .
Neural networks, although a newer concept, have also been used by many people to solve complex automatic
control problems, including the demanding servo problems .
In addition to solving automatic control problems, soft computing has also been used in diverse applications such
as in intelligent speech recognition , communications, fields of signal processing, heavy current systems,
design and manufacturing, pattern recognition, and many more applications.
This paper is an overview of soft computing techniques and describes some of the commonly used techniques to
solve complex problems with soft computing methods, such as fuzzy logic, neural networks, genetic algorithms, and
expert systems.
2. Fuzzy logic
The concept of fuzz logic was introduced by Zadeh3
as a method for representing human knowledge that is
imprecise by nature. Fig. 2 shows the basic configuration of a fuzzy logic system.
The fuzzification interface transforms the crisp input value into a fuzzy linguistic value. The fuzzification is
always necessary in a fuzzy logic system since the input values from existing sensors are always crisp numerical
values. The inference engine takes the fuzzy input and the fuzzy rule base and generates fuzzy outputs. The fuzzy
rule base is in the form of “IF-THEN” rules involving linguistic variables. The last processing element of a fuzzy
logic system is the defuzzification which has the task of producing crisp output actions.
Perhaps one of the biggest advantage of fuzzy logic is that it offers a practical way for designing nonlinear
control systems which are difficult to design and stabilize using traditional methods.
3. Artificial neural networks
Artificial neural networks (ANN), or neural computing is one of the rapidly growing fields of research, attracting
researchers from a wide variety of engineering disciplines, such as electronic engineering, control engineering, and
software engineering.
ANNs are information processing systems that are inspired by the way biological nervous system and the brain
works. ANNs are usually configured for specific applications, such as pattern recognition, data recognition, image
processing, stock market prediction, weather prediction, image compression, and security and loan applications.
Neural networks aim to bring the traditional computers a little closer to the way human brain works. ANNS work
best if the relationship between the inputs and outputs are highly non-linear. ANNs are highly suitable for solving
problems where there are no algorithms or specific set of rules to be followed in order to solve the problem.
A neural network is a large network of interconnected elements, inspired by the human neurons. Each neuron
performs a little operations and the overall operation is the weighted sum of these operations.
A neural network has to be trained so that a known set of inputs produces the desired outputs. Training is usually
done by feeding teaching patterns to the network and letting the network to change its weighting function according
to some previously defined learning rules. The learning can either be supervised, or unsupervised. In supervised
learning the network under investigation is trained by giving it inputs and matching output patterns. i.e. the
outcomes are known for specific inputs. In unsupervised learning the output of the network is trained to respond to
input patterns.
Some of the advantages and disadvantages of neural networks are:
- ANNs are not universal tools for solving problems as there is no methodology for training and verifying an ANN.
- The result of an ANN depends upon the accuracy of the available data
- Excessive training may be required in complex ANN systems
- ANNs can deal with incomplete data sets
- ANNs are successful in prediction and forecasting applications
An ANN is basically composed of three layers: input, hidden layer, and output, where each layer can have
number of nodes. Backpropagation algorithm is used in most ANN networks as a method to train the network.
Here, output of the neural network is evaluated against desired output, and if the results are not as expected, the
weights between layers are modified and the process is repeated until a very small error remains.
4. Genetic algorithms
Genetic algorithms18,19 are parts of artificial intelligence and fuzzy computing and they are mainly used to solve
various optimization problems encountered in real-life applications. The basic idea of a genetic algorithm is to
mimic the natural selection in nature in order to find a good selection for an application. Genetic algorithm is
basically a model of machine learning inspired by the process of evolution in nature. A genetic algorithm can be
used for finding solutions complex search problems found in engineering applications. For example, they can search through various designs and components to find the best combination that will result in overall better and cheaper
design.
Genetic algorithms are used in many diverse fields nowadays, such as climatology, biomedical engineering,
code-breaking, control engineering, games theory, electronic design, and automated manufacturing and design.
The basic processes in genetic algorithms are:
- Initialization, where an initial population is created randomly.
- Evaluation, where each member of the population is evaluated and the fitness of the individuals are assessed based on how well they fit the desired requirements.
- Selection, where only the ones that fit the desired requirements are selected.
- Crossover, where new individual are created by combining best aspects of the existing individuals. At the end of this it is expected to create individuals that are closer to the desired requirements. The process is repeated from the second step until a termination condition is finally reached.
5. Expert systems
An expert system, also known as a knowledge based system, is a computer based system that can make
intelligent decisions by emulating the decision making abilities of human experts. Expert systems are rule based
systems and they are part of the artificial intelligence. Expert systems have the abilities that they can change their
decisions and make new decisions based on the external factors. Some expert systems are designed to take place of a
human in an application, while some others are designed to aid the human. Some application areas of expert systems
are: online medical systems for diagnosing a problem, financial loan/credit decisions, legal matters, robotics, and
engineering design. One of the main problems in expert systems is the knowledge acquisition.
The main components of an expert system are: knowledge base, interface engine, and user interface. The
knowledge base is probably the most important part of any expert system. This is where the intelligence of the
system is stored. Expert systems in general can acquire new knowledge by their sensors or by training and extend
their knowledge bases so that they can easily respond to new problems. The knowledge is stored in the form of INTHEN-ELSE statements. The interface engine is between the knowledge base and the user. The interface engine
makes decisions by following the conditions and the requirements before it comes to an outcome and presents a
solution to the user. The user interface is usually in the form of natural language used daily by the user in everyday
life. There are basically two types of programming languages: algorithmic and symbolic. Traditional programming
languages such as Pascal, Basic C, and Fortran are algorithmic, also known as procedural languages, where it is
difficult to implement logical inferences in these languages. Several symbolic languages have been developed over
the years for expert systems development, such as Prolog, Lisp, Clips and so on.
6. Conclusions and the future
Intelligent systems and hence soft computing techniques are becoming more important as the power of computer
processing devices increase and their cost is reduced. Intelligent systems are required to make complex decisions
and choose the best outcome from many possibilities, using complex algorithms. This requires fast processing
power and large storage space which has recently become available in recent years to many research centres,
universities, and technical colleges at a very low cost.
With the power and the recognition of the Internet of Things (IoT) concept, the need for using soft computing
techniques and building intelligent systems have become more important than ever. Nowadays, most soft computing
applications can be handled efficiently by low-cost but super-fast microcontrollers.
Already we see the use of fuzzy logic, artificial neural networks, and expert systems in many everyday domestic
appliances, such as washing machines, cookers, and fridges. Many industrial and commercial applications of soft
computing are also in everyday use and this is expected to grow within the next decade.
It is the author’s opinion that the soft computing theory and techniques and its applications will grow rapidly
together with the use of IoT devices in future domestic, industrial and commercial markets.
References
Comments
Post a Comment