Unaprijedna neuronska mreža s jednim i dva sakrivena sloja u predikciji ponašanja kaotičnog dinamičkog sustava

undergraduate thesis

• Title (english): Feedforward neural network with one and two hidden layers in prediction of chaotic dynamic system
• Author: Luka Ceraj
• Mentor: Dubravko Majetić (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2015-03-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering; Precision Mechanical Engineering

Abstract

Klasične statističke metode obrade signala daju slabe rezultate u predviđanju ponašanja dinamičkih sustava, a poseban problem predstavljaju sustavi s nelinearnostima, vremenski promjenjivim parametrima i kaotični sustavi. Promjenom parametara kaotični sustavi pokazuju širok raspon različitih nelinearnih ponašanja, te zato predstavljaju dobar problem za ispitivanje raznih tehnika obrade signala, među kojima su i neuronske mreže. U ovom radu biti će prikazan jedan takav problem, gdje će se pomoću unaprijedne statičke neuronske mreže s povratnim prostiranjem pogreške pokušati predvidjeti ponašanje nelinearnog kaotičnog sustava pod nazivom Glass-Mackeyeva jednadžba. Usporediti će se rezultati učenja neuronske mreže s jednim i s dva sakrivena sloja, ispitati utjecaj različitih aktivacijskih funkcija sakrivenih slojeva na učenje mreže, te utjecaj momentuma prvog i drugog reda. Prikazati će se i utjecaj broja neurona pojedinog sakrivenog sloja neuronske mreže s dva sakrivena sloja na učenje mreže. Naučena mreža biti će testirana na njoj nepoznatom skupu ulaznih podataka.

Abstract (english)

Classical statistical signal processing methods give poor results in predicting the behavior of dynamic systems and special issue represent systems with nonlinearity, time-varying parameters and chaotic systems. By changing their parameters chaotic systems show a wide range of different nonlinear behavior and therefore represent a good problem for testing a variety of signal processing techniques, including the neural networks. This paper will be presented with one such problem, where Error-Back Propagation Neural Network with Static Neuron Models will be used in prediction of the behavior of nonlinear chaotic system called Glass-Mackey equation. Learning results of neural networks with one and two hidden layers will be compared, impact of different activation functions of hidden layers on network learning will be examined, and the impact of the momentum of the first and second order. Influence of the number of neurons in each hidden layer of neural network with two hidden layers on learning of the network will be shown. Trained network will be tested with input data not known to network.

• Keywords: predikcija; kaotični sustavi; unaprijedna statička neuronska mreža; povratno prostiranje pogreške; Glass-Mackeyev sustav; sakriveni slojevi; aktivacijske funkcije; momentum prvog reda; momentum drugog reda
• Keywords (english): prediction; chaotic systems; feedforward static neural network; error-back propagation; Glass-Mackey system; hidden layers; activation functions; first order momentum; second order momentum
• Language: croatian
• URN:NBN: urn:nbn:hr:235:152044
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-05-22 12:53:22