Reusch Computational Intelligence, Theory and Applications

International Conference 8th Fuzzy Days in Dortmund, Germany, Sept. 29-Oct. 01, 2004 Proceedings
1. Auflage 2006
ISBN: 978-3-540-31182-9
Verlag: Springer Berlin Heidelberg
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, Band 33, 769 Seiten

Reihe: Advances in Intelligent and Soft Computing

This book constitutes the refereed proceedings of the 8th Dortmund Fuzzy Days, held in Dortmund, Germany, 2004. The Fuzzy-Days conference has established itself as an international forum for the discussion of new results in the field of Computational Intelligence. All the papers had to undergo a thorough review guaranteeing a solid quality of the programme. The papers are devoted to foundational and practical issues in fuzzy systems, neural networks, evolutionary algorithms, and machine learning and thus cover the whole range of computational intelligence.

Written for: Engineers, scientists and graduate students in Computational Intelligence

Keywords: Computational Intelligence, Fuzziness, Soft Computing.

Autoren/Hrsg.

Reusch, Bernd

Inhaltsverzeichnis

1;Preface;6
2;Contents;7
3;Evolutionary Algorithms;16
3.1;An Evolutionary Algorithm for the unconstrained Binary Quadratic Problems;17
3.2;Application of Genetic Algorithms by Means of Pseudo Gradient;31
3.3;Optimization by Island-Structured Decentralized Particle Swarms;39
3.4;Directed Mutation by Means of the Skew-Normal Distribution;49
4;Rule-Based Fuzzy Inference;66
4.1;Smooth Extensions of Fuzzy if-then Rule Bases;67
4.2;Pre-validation of a Fuzzy Model;75
4.3;Multiresolution Fuzzy Rule Systems;79
5;Data Characterization through Fuzzy Clustering;95
5.1;Fuzzy Clustering of Macroarray Data;97
5.2;Fuzzy Clustering: Consistency of Entropy Regularization;109
5.3;Fuzzy Long Term Forecasting through Machine Learning and Symbolic Representations of Time Series;123
5.4;Fuzzy Prototypes Based on Typicality Degrees;139
6;Plenary Talk;154
6.1;The Power of Zadeh’s Protoforms: Towards General Problem Formulations in Fuzzy Multistage Control and Group Decision Making;155
7;Fuzzy Control;158
7.1;Fuzzy Logic Fluid Therapy Control System for Renal Transplantation;159
7.2;Interpolative Fuzzy Reasoning in Behaviour-based Control;173
7.3;Fuzzy Modeling of Offensive Maneuvers in an Air-to-Air Combat;185
8;Recent Advances in Theoretical Soft Computing;200
8.1;Approximation of Fuzzy Functions by Extended Fuzzy Transforms;201
8.2;Fuzzy Control as a General Interpolation Problem;211
8.3;Galois Connections with Truth Stressers: Foundations for Formal Concept Analysis of Object-Attribute Data with Fuzzy Attributes;219
8.4;Fuzzy Transforms in Removing Noise ;235
8.5;Safe Modelling of Fuzzy If–Then Rules;245
8.6;Perception-Based Logical Deduction;251
9;Towards Intelligent Decision Support Systems via Soft Computing;266
9.1;Fuzzy Object-oriented Modelling with Metadata Attributes in C;267
9.2;Strategies for Decision Making in the Conditions of Intuitionistic Fuzziness;277
9.3;Fuzzy Linguistic Summaries in Text Categorization for Human-Consistent Document-Driven Decision Support Systems;285
9.4;An Application of Intuitionistic Fuzzy Relational Databases in Football Match Result Predictions;295
9.5;Generalized Net Model for Adaptive Electronic Assessment, Using Intuitionistic Fuzzy Estimations;305
10;Fuzzy Logic in Decision Support;313
10.1;Analytic Hierarchy Process Based on Fuzzy Analysis;315
10.2;A Fuzzy-Ga Hybrid Technique for Optimization of Teaching Sequences Presented in ITSs;325
10.3;Consistency Conditions for Fuzzy Choice Functions;331
11;Applications of Fuzzy Systems;340
11.1;A Fuzzy Logic Application to Environment Management System: A Case Study for Goksu Streams Water Quality Assesment;341
11.2;Combination Rule of Normal Degrees on Automated Medical Diagnosis System (AMDS);353
11.3;Generation of Representative Symptoms Based on Fuzzy Concept Lattices;363
12;Connectives;370
12.1;On the Direct Decomposability of Fuzzy Connectives, Negations and Implications Based on t-norms and t-conorms on Product Lattices;371
12.2;The Cancellation Law for Addition of Fuzzy Intervals;383
12.3;Generic View On Continuous T-Norms and T-Conorms;391
13;Intelligent Techniques for Knowledge Extraction and Management;398
13.1;Mining Class Hierarchies from XML Data: Representation Techniques;399
13.2;Generalizing Quantification in Fuzzy Description Logics;411
13.3;Fuzzy Types: A First Step Towards Lazy Types in the .NET Framework;427
13.4;Fuzzy Induction via Generalized Annotated Programs;433
13.5;Evaluating Fuzzy Association Rules on XML Documents;449
14;Plenary Talk;464
14.1;Ubiquitous Robot;465
15;Fuzzy Image Processing;475
15.1;Combining Fuzzy Logic and Kriging for Image Enhancement;477
15.2;Optical Quality Control of Coated Steel Sheets Using Fuzzy Grey Scale Correlograms;489
16;Plenary Talk;496
16.1;Fuzzy-Methods in Knowledge Discovery;497
17;Evolutionary Algorithms;499
17.1;Action Games: Evolutive Experiences;501
17.2;Training Sets Co-evolving Multilayer Perceptrons along;517
17.3;Improving Parallel GA Performances by Means of Plagues;529
17.4;Hybrid Evolutionary Algorithms for Protein Structure Prediction under the HPNX Model;539
18;Aggregation Operators;550
18.1;Quasi-Copulas on Discrete Scales;551
18.2;Basic Classification of Aggregation Operators and Some Construction Methods;559
18.3;Homogeneous Aggregation Operators;569
18.4;1-Lipschitz Aggregation Operators, Quasi-Copulas and Copulas with Given Opposite Diagonal;579
18.5;Fuzzy Measures and Choquet Integral on Discrete Spaces;587
19;Neural Networks;597
19.1;Modular Neural Network Applied to Non-Stationary Time Series ;599
19.2;A Feedforward Neural Network based on Multi-Valued Neurons ;613
19.3;Least Squares Support Vector Machines for Scheduling Transmission in Wireless Networks;627
19.4;Neural Networks for the Control of Soccer Robots;635
20;Neuro-Fuzzy Systems;644
20.1;Universal Approximator Employing Neo-Fuzzy Neurons;645
20.2;Combined Learning Algorithm for a Self-Organizing Map with Fuzzy Inference;655
20.3;Fuzzy/Neural Connection Admission Controller for Multimedia Tra.c in Wireless ATM Networks;665
21;Fuzzy Mathematics;682
21.1;Limits of Functional Sequences in the Concept of Nearness Relations;683
21.2;On the Law of Large Numbers on IFS Events;691
21.3;An Axiomatic Approach to Cardinalities of IF Sets;695
22;Fuzzy Optimization;707
22.1;Sensitivity Analysis for Fuzzy Shortest Path Problem;709
22.2;Fuzzy Coloring of Fuzzy Hypergraph;717
22.3;Nonlinear Optimization with Fuzzy Constraints by Multi-Objective Evolutionary Algorithms;727
23;Poster Contributions;738
23.1;Comparison of Reasoning for Fuzzy Control;739
23.2;Evolving Scientific Knowledge;747
23.3;Coding of Chaotic Orbits with Recurrent Fuzzy Systems;753
23.4;Genetic-Based Tuning of Fuzzy Dempster-Shafer Model;761
23.5;A Novel Design for Classifying Multi-field Internet Packets Using Neural Networks;771
23.6;Modeling Uncertainty in Decision Support Systems for Customer Call Center;777
23.7;A New GA-Based Real Time Controller for the Classical Cart-Pole Balancing Problem;785
23.8;Depth Control of Desflurane Anesthesia with Cardiovascular-based an Adaptive Neuro-Fuzzy System;801
23.9;Ultrasound Intensity and Treatment Time Fuzzy Logic Control System for Low Cost Effective Ultrasound Therapy Devices;811
24;Author Index;823

Leseproben

2 The principle of the new evolutionary algorithm. (p. 4)

2.1 The structure of the algorithm.

Hybrid EAs are frequently used for solving combinatorial problems. These methods improve the quality of the descendent solution for example with the application of a local search procedure, SA, or TS. The constitution of these systems corresponds to an extension of an EA: for instance a local search procedure is applied at every step of the EA cycle.

The new EA unlike former hybrid EAs based on a single stage, uses a 2-stage algorithm structure in order to speed up convergence and to produce higher quality results. The first stage is a quick "preparatory" stage that is designated to improve the quality of the initial population. The second stage is a hybrid EA with some special operators.

Let us discuss the 2 EAs (stages) in more detail:

1. The first stage forms some solutions at random and then tries to improve them by randomly generating descendents. The descendent may replace the most similar one of the former solutions.

2. The second stage is a hybrid ES. The algorithm uses two different recombination operations, and concatenated, complex neighbourhood structures for the mutations. The recombination operation is a uniform or single-point recombination or otherwise simple copy-making.

In selecting the parents, priority is given to the best, highest objective/fitness function value: the algorithm selects the fittest solution with 0.5 probability and another solution with 0.5/t probability (where t is the size of the population).

By mutation we applied varying number of bit-flip and a special bit-flip (bit- flip-flop). We form the neighbourhood structure using: some bit-flip-flops + some bit-flips.

The quality of the solutions is improved with a local search procedure. We applied the randomized k-opt local search (Merz and Katayama 2001). Finally in order to keep the diversity of the population we use a filter and a restart procedure. The filter selects only the best of the solutions close to each other, the other ones are deleted.

The restart begins the second stage again, if the fittest solution didn’t change in the last generations. It replaces the weakest solutions with new ones (70% of the population), and it applies the local search procedure on a part of the new individuals.

3 The new algorithm

3.1 The characteristics of the EAs
The main functions and characteristics of the EAs are the following:

Initial population. The same population and individuals are used in all stages. The first individuals of the P population are randomly generated from S. These are the first "solutions".

Fitness function. The algorithm uses the objective function f(x) as fitness function.

Selection operator. In the first stage descendents are randomly selected from S, without the application of any further operators (recombination, mutation). In the second stage the algorithm selects two different parents from the population: the first of them is the most appropriate solution with 0.5 probabilities.