An Algorithm for Improving Quality of a Fuzzy Rule Base

• Autorzy: Izabela Rejer

Warianty tytułu

Algorytm poprawy jakości bazy reguł modelu rozmytego

• Języki publikacji: EN

Abstrakty

EN

One of the most popular approaches taken in the process of automatic creation of a rule base of a fuzzy model is a classic grid partitioning of an input space of the analyzed system. The basic feature of this approach is that a rule base created on the basis of this grid contains a lot of rules which can be eliminated from the model without a loss in a model precision. The unnecessary rules exist in the model because the grid is applied over the whole input space, regardless of the distribution of data points. Hence, in case of an unequal data distribution, some parts of the input space of the analyzed system are not covered by data points, which means that some rules are unsupported by any data point. Of course rules not supported by data points should not take part in the inference process because they can produce incorrect results in future model application. However, when these rules are eliminated from the rule base, a fuzzy model becomes an incomplete one. The theory states that an incomplete model either should not be used in practice at all or should be used only in some, very precisely defined, regions of an input space. In fact, there is one more possibility. The model completeness can be extorted by restoring some of previously discarded rules - of course after necessary changes in their conclusions. The aim of this article is to present an approach to chain systems modelling which can be used for: eliminating rules, calculating new conclusions for some of them and adding them again to the model. (original abstract)

Jednym z bardziej popularnych podejść w procesie automatycznej budowy bazy reguł modelu rozmytego jest prostokątny podział przestrzeni wejściowej badanego systemu. Podstawową cechą takiego podejścia jest to, że baza reguł stworzona na podstawie takiego podziału zawiera reguły nadmiarowe, to jest reguły, które mogą zostać wyeliminowane z modelu bez spadku jego dokładności. Eliminacja reguł z modelu rozmytego stwarza jednak niebezpieczeństwo, że powstały model będzie modelem niekompletnym. Celem artykułu jest zaprezentowanie autorskiego podejścia, pozwalającego na eliminację nadmiarowych reguł z bazy reguł modelu rozmytego, niedopuszczającego do pojawienia się nieciągłości w bazie reguł. (abstrakt oryginalny)

Słowa kluczowe

EN

Fuzzy systems; Algorithms

PL

Systemy rozmyte; Algorytmy

• Czasopismo: Studia i Materiały Polskiego Stowarzyszenia Zarządzania Wiedzą
• Rocznik: 2011
• Tom: 42
• Strony: 204--217

Twórcy

autor

Izabela Rejer

• Uniwersytet Szczeciński

Bibliografia

• Bronsztejn I.N. at al, Modern Compendium of Mathematic, Polish Science Publisher Warsaw, 2004.
• Graham R.L., An Efficient Algorithm for Determining the Convex Hull of a Finit Planar Set, "Information Processing Letters" 1, 1972: p. 132-133.
• Klesk P., The method of setting suitable extrapolation capabilities for neuro-fuzzy models of multidimensional systems, PhD Thesis, Technical University of Szczecin, 2005.
• Klęsk P., Algorithm for automatic definition of validated and nonvalidated region in multidimensional space, 10th International Conference on Advanced Computer Systems, 2003.
• Kóczy L.T., Hirota K., Approximate reasoning by linear rule interpolation and general approximation, "International Journal of Approximate Reasoning"., vol. 9, 1993.
• Krivsky S., Lang B., Verified computation of Higher Dimensional convex hulls and the solution of linear systems, "Electronic Journal on Mathematics of Computation", 2003.
• Masters T., Practical Neural Networks Recipes in C++, Academic Press Inc., 1993.
• Niederliński A., Polynomial and neural input-output models for control - a comparison, Fourth International Symposium on Methods and Models in Automation and Robotics, Międzyzdroje, 1997.
• Ojala T., Neuro-fuzzy systems in control, Master of Science Thesis, Tampere University of Technology, Department of Electrical Engineering, Finland, 1994.
• Pavia R.P., Dourado A., Structure and parameter learning of neuro-fuzzy systems: A methodology and a comparative study, "Journal of Intelligent & Fuzzy Systems", No. 11, 2001.
• Piegat A., Fuzzy Modelling and Control, Physica-Verlag, New York 1999.
• Piegat A., Rejer I., Mikołajczyk M., Application of Neural Networks in Chain Curve Modelling, Artificial Intelligence and Soft Computing ICAISC 2006 - Lectures Notes in Artificial Intelligence, Springer-Verlag, Berlin, 2006.
• Preparate F.C., Hong S.J., Convex Hulls of Finite Sets of Points in Two and Three Dimensions, Commun. ACM, vol. 20, no. 2, 1977: p. 87-93.
• Rejer I., Mikołajczyk M., A Hypertube as a Possible Interpolation Region of a Neural Model, Lectures Notes in Artificial Intelligence, Springer-Verlag, Berlin, 2006.
• Rejer I., Reduction of rules of a fuzzy model with a chain model, Advances in Information Processing and Protection, red. Pejaś J., Saeed K., Springer, 2007.
• Rutkowska D., Piliński M., Rutkowski L., Neural networks, genetic algorithms and fuzzy systems, Warsaw Scientific Publishing House Ltd., 1999.
• Weisstein, E.W., Convex hull, Published by Wolfram Research. From Math World, http://mathworld.wolfram.com/ConvexHull. html, 2006.
• Wong K. at al., Fuzzy Rule Interpolation for Multidimensional Input Spaces With Applications: A Case Study, "IEEE Transaction on Fuzzy Systems", Vo. 13, No. 6, December, 2005.
• Xiong N., Evolutionary learning of rule premises for fuzzy modelling, "International Journal of Systems Science", Vol. 32, No. 9, 2001.