Napovedovanje multivariatnih časovnih vrst geoprostorskih podatkov z uporabo konvolucijsko-povratnih nevronskih mrež

magistrsko delo

Niko Uremovič (Author), Niko Lukač (Mentor), Marko Bizjak (Co-mentor)

Abstract

V magistrskem delu predstavimo nov pristop za napovedovanje multivariatnih časovnih vrst geoprostorskih podatkov. Pripravimo pregled obstoječih pristopov k napovedovanju časovnih vrst prostorskih podatkov. Predstavimo koncepte na katerih temelji konvolucijsko-povratna nevronska mreža ConvLSTM in njeno teoretično osnovo. Z uporabo ConvLSTM pri napovedovanju upoštevamo tako časovne odvisnosti med spremenljivkami, kot tudi prostorske odvnisnosti med podatki v sosednjih točkah. Metodo preizkusimo na primeru napovedovanja več spremenljivk onesnaženosti zraka za več merilnih postaj na različnih lokacijah in jo primerjamo s sorodnimi deli.

Keywords

multivariatne časovne vrste;geoprostorski podatki;napovedovanje časovnih vrst;konvolucijsko-povratne nevronske mreže;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2022
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[N. Uremović]
UDC:	519.2:004.8(043.2)
COBISS:	132919555
Views:	47
Downloads:	12
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Geospatial multivariate time series forecasting using convolutional reccurrent neural networks
Secondary abstract:	In this thesis we present a method for multivariate time series forecasting for geospatial data. We prepare an overview of existing methods for multivariate spatial time series forecasting. We present the theorethical background of the ConvLSTM neural network architecture and the concepts it is based on. By using ConvLSTM for geospatial time series forecasting, we account for both spatial and temporal dependencies in our data. We test the proposed method on the case of forecasting multiple variables of air pollution for multiple measurement stations and compare our results to related work.
Secondary keywords:	Multivariate time series;geospatial data;time series forecasting;convolutional recurrent neural networks;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	1 spletni vir (1 datoteka PDF (VII, 36 f.))
ID:	16309900