nach oben

2008 | Buch

Kapitel lesen Erstes Kapitel lesen

Soft Computing

Techniques and its Applications in Electrical Engineering

verfasst von: Dr. D. K. Chaturvedi

Verlag: Springer Berlin Heidelberg

Buchreihe : Studies in Computational Intelligence

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

Intuitive consciousness/ wisdom is also one of the frontline areas in soft computing, which has to be always cultivated by meditation. This book is an introduction to some new fields in soft computing with its principal components of fuzzy logic, ANN and EA and it is hoped that it would be quite useful to study the fundamental concepts on these topics for the pursuit of allied research.

The approach in this book is to provides an understanding of the soft computing field, to work through soft computing using examples, to integrate pseudo - code operational summaries and Matlab codes, to present computer simulation, to include real world applications and to highlight the distinctive work of human consciousness in machine.

"I believe the chapters would help in understanding not only the basic issues and characteristic features of soft computing, but also the aforesaid problems of CTP and in formulating possible solutions. Dr. Chaturvedi deserves congratulations for bringing out the nice piece of work." Sankar K. Pal, Director Indian Statistical Institute

Inhaltsverzeichnis

Frontmatter

1. Introduction to Soft Computing

Soft computing (SC) is a branch, in which, it is tried to build intelligent and wiser machines. Intelligence provides the power to derive the answer and not simply arrive to the answer. Purity of thinking, machine intelligence, freedom to work, dimensions, complexity and fuzziness handling capability increase, as we go higher and higher in the hierarchy as shown in Fig. 1.1. The final aim is to develop a computer or a machine which will work in a similar way as human beings can do, i.e. the wisdom of human beings can be replicated in computers in some artificial manner.

Intuitive consciousness/wisdom is also one of the important area in the soft computing, which is always cultivated by meditation. This is indeed, an extraordinary challenge and virtually a new phenomenon, to include consciousness into the computers.

Soft computing is an emerging collection of methodologies, which aim to exploit tolerance for imprecision, uncertainty, and partial truth to achieve robustness, tractability and total low cost. Soft computing methodologies have been advantageous in many applications. In contrast to analytical methods, soft computing methodologies mimic consciousness and cognition in several important respects: they can learn from experience; they can universalize into domains where direct experience is absent; and, through parallel computer architectures that simulate biological processes, they can perform mapping from inputs to the outputs faster than inherently serial analytical representations. The trade off, however, is a decrease in accuracy. If a tendency towards imprecision could be tolerated, then it should be possible to extend the scope of the applications even to those problems where the analytical and mathematical representations are readily available. The motivation for such an extension is the expected decrease in computational load and consequent increase of computation speeds that permit more robust system (Jang et al. 1997).

2. Life History of Brain

The history of our quest to understand the brain is certainly as long as human history itself. Use this extensive timeline to meander through some of the high-lights (and low-lights) of this great journey of understanding. There are many evidences of ancient civilization which show that people were conducted surgery on head (Brain). In Hindu religion, Lord Ganesh had the head of elephant, Incarnation of Lord Narsingha, etc. Today due to fast progress of neuro-science, we are at the verge of understanding that how brain functions and what is the relationship between mind and brain, which may provide a basis for understanding consciousness.

Plato hypothesized that the brain was the seat of the soul and also the center of all control. It is somewhat surprising that he came to this correct conclusion in spite of the fact he rejected experiment and observation, and believed that true knowledge came only from pure reasoning and thought such as that involved in mathematics. It is also mentioned that our pleasure, joys, laughter and jests as well as our sorrows, pains, grief and tears every thing is closely controlled by the brain condition.

3. Artificial Neural Network and Supervised Learning

Artificial neural networks are biologically inspired but not necessarily biologically plausible. Researchers are usually thinking about the organization of the brain when considering network configurations and algorithms. But the knowledge about the brain’s overall operation is so limited that there is little to guide those who would emulate it. Hence, at present time biologists, psychologists, computer scientists, physicists and mathematicians are working all over the world to learn more and more about the brain. Interests in neural network differ according to profession like neurobiologists and psychologists try to understanding brain. Engineers and physicists use it as tool to recognize patterns in noisy data, business analysts and engineers use to model data, computer scientists and mathematicians viewed as a computing machines that may be taught rather than programmed and artificial intelligentsia, cognitive scientists and philosophers use as sub-symbolic processing (reasoning with patterns, not symbols), etc.

A conventional computer will never operate as brain does, but it can be used to simulate or model human thought. In 1955, Herbert Simon and Allen Newell announced that they had invented a thinking machine. Their program, the logic theorist, dealt with problems of proving theories based on assumptions it was given. Simon and Newell later developed the general problem solver, which served as the basis of artificial intelligence (AI) systems. Simon and Newell believed that the main task of AI was figuring out the nature of the symbols and rules that the mind uses. For many years AI engineers have used the "top-down" approach to create intelligent machinery. The top-down approach starts with the highest level of complexity, in this case thought, and breaks it down into smaller pieces to work with. A procedure is followed step by step. AI engineers write very complex computer programs to solve problems. Another approach to the modelling of brain functioning starts with the lowest level, the single neuron. This could be referred to as a bottom-up approach to modelling intelligence.

4. Factors Affecting the Performance of Artificial Neural Network Models

Artificial neural network is widely used in various fields like system’smodelling, forecasting, control, image processing and recognition, and many more. The development of multi-layered ANN model for a particular application involves many issues which affect its performance. ANN performance depends mainly upon the following factors:

Network

Problem complexity

Learning Complexity.

5. Development of Generalized Neuron and Its Validation

More recently, ANNs and fuzzy set theoretic approach have been proposed for many different industrial applications. A number of papers have been published in the last two decades. An illustrative list is given in bibliography. Both techniques have their own advantages and disadvantages. The integration of these approaches can give improved results.

In the previous chapter, the performance aspect of ANN has been discussed in detail. To overcome some of the problems of ANN and improve its training and testing performance, the simple neuron is modified and a generalized neuron is developed in this chapter.

In the common neuron model generally the aggregation function is summation, which has been modified to obtain a generalized neuron (GN) model using fuzzy compensatory operators as aggregation operators to overcome the problems such as large number of neurons and layers required for complex function approximation, which not only affect the training time but also the fault tolerant capabilities of the artificial neural network (ANN) (Chaturvedi 1997).

6. Applications of Generalized Neuron Models

In the earlier chapter, the development details of GN models have been studied. GN models have also been tested on benchmark problems. It is found that the GN models are much better than multilayered ANN in all the benchmark problems, which encouraged to use for different problems like modeling and simulation of electrical machines, short term electrical load forecasting and various control applications.

7. Introduction to Fuzzy Set Theoretic Approach

The scientists have been trying to develop an intelligent machine similar to human beings since last many years. There are many points of similarity and differences between computers and human processing. A comparison between computer (machine system) and human system is given in Table 7.1. The computer is a logical machine, works on the basis of precise logic. Today we have very fast computers with large memory to store data. Then also these computers could not help us in answering simple questions as given below.

8. Applications of Fuzzy Rule Based System

The basic concepts of fuzzy system were described in the earlier chapter. There are numerous applications of fuzzy systems in various fields such as operations research, modeling, evaluation, pattern recognition, control and diagnosis, etc. Fuzzy systems theory is the starting point for developing models of ambiguous thinking and judgment processes, the following fields of application are conceivable:

Human models for management and societal problems;

Use of high level human abilities for use in automation and information systems;

Reducing the difficulties of man-machine interface;

Other AI applications like risk analysis and prediction, development of functional device.

Fuzzy systems are quite popular in control applications. In standard control theory, a mathematical model is assumed for the controlled system, and control laws that minimize the evaluation functions are determined; but when the object is complicated, mathematical models cannot be determined and one cannot figure out how to decide on the evaluation functions. In these cases, skilled individuals perform control functions by using their experience and intuition to judge situations on the basis of what they think.

9. Genetic Algorithms

Origin with a protozoa (prime unicell animal) to existence of human (most developed living being) in nature as a result of evolution, is the main theme, adopted by genetic algorithms (GA), one of the most modern paradigm for general problem solving. Since the paradigm simulates the strategy of evolution, it is surprisingly simple but powerful, domain free approach to problem solving. GAs are gaining popularity in many engineering and scientific applications due to their enormous advantages such as adaptability, ability to handle non-linear, ill defined and probabilistic problems. As the approach is domain free, it has wide scope of applications and most of the optimization problems can be handled successfully with this approach.

The emergence of massively parallel computers made these algorithms of practical interest. There are various well known programs in this class like evolutionary programs, genetic algorithms, simulated annealing, classifier systems, expert systems, artificial neural networks and fuzzy systems. This chapter discusses a genetic algorithm – which is based on the principle of evolution (survival of fittest). In such algorithms a population of individuals (potential solution) undergoes a sequence of transformations like mutation type and crossover type. These individuals strive for survival; a selection scheme, biased towards fitter individuals, selects the next generation. After some number of generations the program converges to the optimal value.

10. Applications of Genetic Algorithms to Load Forecasting Problem

Evolutionary programs are gaining popularity in many engineering and scientific applications due to their enormous advantages such as adaptability, ability to handle non-linear, ill-defined and probabilistic problems. Specific reference to genetic algorithms (Gas), some parameters that influence the convergence to the optimal value are the population size (popsize), the crossover probability (Pc) and the mutation propability (Pm). Normally these values are prescribed initially and do not vary during the execution of the program, although these parameters greatly affect the performance of GA.

The present chapter deals with the development of an improved genetic algorithm (IGA) by introducing a variation in the values of the parameters like population size (popsize), the crossover probability (Pc) and the mutation propability (Pm). The aim of this variation is to minimize the convergence time. This work presents a method of dynamically varying the parameters of operation of the GA program using fuzzy state theory (FST) so that the final convergence is obtained in a shorter time.

Also, in this chapter a function has been developed and optimized for long-term load forecasting problem using IGA. This technique does not require any previous assumption of a function for load forecasting, further, it does not need any functional relationship between dependent and independent variables. The results obtained by this technique are compared with the data available from central electricity authority (CEA), India to demonstrate the effectiveness of the proposed technique.

11. Synergism of Genetic Algorithms and Fuzzy Systems for Power System Applications

The power system of today is a complex interconnected network having four major components – generation, transmission, distribution and loads. Electricity is being generated in large hydro, thermal and nuclear power stations, which are normally located far away from the load centers. Large and long transmission networks are wheeling the generated power from these generating stations to different distribution systems, which ultimately supply the load. The distribution system is that part of the power system which connects the distribution substations to the consumers’ service-entrance.

Earlier the utilities were mainly concerned about the optimal dispatch of active power only, but evolvement of competition has also resulted in the optimal dispatch of reactive power. When only total cost is minimized by real power scheduling of available generator in a system, the optimal power flow (OPF) corresponds to Active Power Dispatch. Some of the solution techniques successfully used for active power dispatch include classical co-ordination methods based on Lagrangian multiplier approach (Chowdhury and Rahman 1990), Linear programming (LP) based methods (Stott and Hibson 1978; Stott and Marinho 1979), quadratic programming (QP) approach (Nanda et al. 1989), Gradient method using steepest descent technique (Dommel and Tinney 1968) and Newton's methods (Sun et al. 1984; Maria and Findlay 1987). A comprehensive review of various optimization techniques available in the literature is reported in references (Happ 1977; Sasson and Merril 1974; Carpantier 1985). The classical method of optimization is relatively simple, fast and requires less memory space but sometimes it is unable to handle the system constraints effectively and sometimes convergence is not obtained. The LP based method involves approximation in linearizing the objective function and constraints and may result in zigzagging of the solution. Gradient based methods compute the derivative of the function at each step. They require a close initial guess and in general suffer from convergence difficulties and may stuck to local minima.

12. Integration of Neural Networks and Fuzzy Systems

The literature reveals that the ANN and Fuzzy set theoretic approaches have been often used for non-linear and complex problems such as load forecasting in power system. The integration of these approaches gives improved results as compared to conventional techniques. Both the modeling techniques have their own merits and demerits as follows:

Fuzzy models possess large power in representing linguistic and structured knowledge by fuzzy sets and performing fuzzy reasoning by fuzzy logic in qualitative manner and usually rely on the domain experts to provide the required knowledge for a specific problem. Further, the compensatory operators in the fuzzy models as connectives are found quite suitable and produce results, which are very close to the actual results (Mizumoto 1989).

On the other hand, neural network models are particularly good for non-linear mappings and for providing parallel processing facility to simulate complex system. The neural network models are developed via training.

Furthermore, while the behavior of fuzzy models can be understood easily due to their logical structure and step-by-step inference procedures. Neural network models act normally as a black box, without providing explanation facility.

13. ANN – GA-Fuzzy Synergism and Its Applications

The feed-forward backpropagation artificial neural networks (ANN) are widely used to control the various industrial processes, modelling and simulation of systems and forecasting purposes. The backpropagation learning has various drawbacks such as slowness in learning, stuck in local minima, requires functional derivative of aggregation function and thresholding function to minimize error function etc. Various researchers suggested a number of improvements in simple back-propagation learning algorithm developed.

In this paper, a program is developed for feed-forward artificial neural network with genetic algorithm (GA) as the learning mechanism to overcome some of the disadvantages of backpropagation learning mechanism to minimize the error function of ANN.

Genetic algorithm (GA) simulates the strategy of evolution and survival of fittest. It is a powerful domain free approach integrated with ANN as a learning tool. The ANN – GA integrated approach is applied to different problems to test this approach. It is well known that the GA optimization is slow and depending on the number of variables. To improve the convergence of GA, a modified GA is developed, in which, the GA parameters are modified using five fuzzy rules and concentration of genes is suggested.

Backmatter

Titel: Soft Computing
verfasst von: Dr. D. K. Chaturvedi
Verlag: Springer Berlin Heidelberg
Electronic ISBN: 978-3-540-77481-5
Print ISBN: 978-3-540-77480-8
DOI: https://doi.org/10.1007/978-3-540-77481-5