Predviđanje protoka vozila u mreži prometnica

master's thesis

• Author: Andrea Stanić
• Mentor: Tomislav Šmuc (mentor)
• Committee member: Tomislav Šmuc (predsjednik povjerenstva)
• Committee member: Goran Igaly (član povjerenstva)
• Committee member: Robert Manger (član povjerenstva)
• Committee member: Ivica Nakić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2018-11-27, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

U ovom radu je istražen problem predviđanja protoka prometa u mreži prometnica. Složenost procesa koja je iza protoka prometa je iznimno velika i pri izgradnji modela treba voditi računa o mnogo čimbenika koji utječu na promet, a koji su navedeni u prvom poglavlju. U drugom poglavlju je dan teoretski uvod u vremenske nizove te ARMA, ARIMA, SARIMA i STARIMA modele. Za svaki od navedenih modela je opisan neki od postojećih pristupa za modeliranje protoka prometa. Treći dio rada je posvećen dubokom učenju te rekurentnim neuronskim mrežama. Opisana su dva različita pristupa, a svaki od njih koristi prilagođenu verziju rekurentnih neuronskih mreža. Prvi pristup koristi difuzijsku konvolucijsku rekurentnu neuronsku mrežu, a drugi dvosmjernu LSTM neuronsku mrežu. Posljednje poglavlje sadrži primjenu STARIMA modela i jednostavne LSTM neuronske mreže na podacima koji simuliraju protok prometa u Varšavi.

Abstract (english)

This thesis’ emphasis has been on the problem of traffic flow forecasting in the road network. The process behind traffic flow is extremely complex, and when creating a model one should take into account many of the factors affecting traffic flow. The first chapter of the thesis explains those factors. In the second chapter, there is a theoretical introduction to time series and ARMA, ARIMA, SARIMA, and STARIMA models. Some of the existing traffic flow modeling approaches have been described for the above-mentioned models. The third part of the thesis includes an introduction to deep learning and recurrent neural networks. Two different approaches have been described, each of which uses a custom version of recurrent neural networks. The first approach uses a diffusion convolutional recurrent neural network, and the second bidirectional LSTM neural network. The last chapter includes the use of STARMA model and simple LSTM neural networks on data simulating the traffic flow in Warsaw.

• Keywords: predviđanje toka prometa; vremenski nizovi; rekurentne neuronske mreže
• Keywords (english): traffic flow forecasting; time series; recurrent neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:217:867004
• Study programme: Title: Mathematical Statistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-06-19 09:50:16