magistrsko delo
Uroš Petković (Author), Igor Škrjanc (Mentor)

Abstract

V magistrskem delu smo predstavili področje samorazvijajočih se sistemov, pri čemer smo se osredotočili na sisteme z mehko logiko. Osrednji del naloge je predstavitev in analiza obstoječega samorazvijajočega sistema eGAUSS+. Sistem temelji na rojenju glede na vrednost pripadnostne funkcije, ki je določena z Gaussovo funkcijo. Roji sistema se združujejo glede na primerjavo prostornin prostora, ki ga pokrivajo. Metodo smo modificirali in ji dodali dodatni vhodni parameter za omejitev velikosti rojev. Analizirali smo vpliv posameznih vhodnih parametrov sistema na samo rojenje. Pokazali smo, da dodani parameter bistveno izboljša ponovljivost nenadzorovanega rojenja. Metoda je uporabna za primer identifikacije vhodno-izhodnega dinamičnega sistema in za primer nadzorovanega razvrščanja. Preizkusili smo sistem na problemu vhodno-izhodne identifikacije dveh nelinearnih dinamičnih sistemov. Določili smo tudi interval zaupanja napovedi in raziskali vpliv predhodnega filtriranja regresorjev na rojenje in identifikacijo. Delovanje sistema smo ocenili še na primeru nadzorovanega razvrščanja in rezultate primerjali s preostalimi razvrščevalniki.

Keywords

samorazvijajoči se sistemi;tok podatkov;mehki sistemi;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FE - Faculty of Electrical Engineering
Publisher: [U. Petković]
UDC: 681.5(043.3)
COBISS: 69288963 Link will open in a new window
Views: 384
Downloads: 78
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Analysis of fuzzy evolving system based on Gaussian clustering
Secondary abstract: In this thesis, we presented the field of evolving systems with an emphasis on ones with fuzzy logic. The main focus of this work is a recently presented evolving eGAUSS+ system. The system is based on Gaussian clustering, where clusters are merged depending on the comparison between the sum of volumes of two clusters. The method was modified and a new input parameter for cluster volume control was added. We analysed the input parameters of the system, and evaluated their impact on clustering performance. We showed that the added input parameter significantly improves the repeatability of unsupervised clustering. This method can be used for input-output identification and supervised classification. The performance of input-output identification was evaluated on two different, nonlinear dynamical systems. We determined the confidence interval of model output and examined the effect of filtering of regressors on clustering and identification as well. The method performance was evaluated for supervised classification and the obtained results were compared with other classifiers.
Secondary keywords: evolving systems;data streams;fuzzy systems;
Type (COBISS): Master's thesis/paper
Study programme: 1000316
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za elektrotehniko
Pages: XVIII, 80 str.
ID: 12195564