Primjena dubokih neuronskih mreža za predviđanje dnevne koncentracije lebdećih čestica u zraku

master's thesis

• Title (english): Application of deep neural networks for the prediction of particulate matter concentration in the air
• Author: Valentino Petrić
• Mentor: Željka Ujević Andrijić (mentor)
• Committee member: Željka Ujević Andrijić (predsjednik povjerenstva)
• Committee member: Dajana Kučić Grgić (član povjerenstva)
• Committee member: Gordana Pehnec (član povjerenstva)
• Granter: University of Zagreb; Faculty of Chemical Engineering and Technology; Zagreb
• Defense date and country: 2022-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Chemical Engineering

Abstract

Cilj ovog rada je razviti modele za procjenu koncentracije lebdećih čestica u zraku pomoću metoda umjetne inteligencije. U radu su primijenjene metoda nasumičnih šuma i metoda neuronskih mreža koristeći programski jezik Python i pripadajuće knjižnice. Mjerni podaci korišteni u ovom radu odnose se na austrijski grad Graz čiji su meteorološki podaci mjereni na pet postaja u gradu: Sjever, Istok, Zapad, Jug i Don Bosco. Uz meteorološke podatke na postajama je mjerena i koncentracija lebdećih čestica u zraku. Uz lokalne meteorološke podatke korišteni su i satelitski podaci preuzeti sa stranice Climate Data Store. Koristeći lokalne i satelitske podatke razvijeni su modeli za svih pet postaja te je ispitivano koji podaci će pridonijeti boljoj kvaliteti modela. Rezultati su pokazali da su modeli razvijeni sa lokalnim meteorološkim podacima nešto točniji u usporedbi s modelima razvijenih sa satelitskim meteorološkim podacima. U ovom radu se ispitivalo i kako podaci za promet mjereni na šesnaest prometnica u gradu utječu na koncentraciju lebdećih čestica. Uvođenjem prometa kao dodatne ulazne varijable za razvoj modela postignuto je poboljšanje kvalitete modela posebno na skupu s lokalnim podacima. U konačnici su se svi dostupni podaci spojili u dva skupa, pri čemu su prvi skup činili lokalni i satelitski meteorološki podaci, a drugi skup su činili lokalni i satelitski meteorološki podaci kojima je pridodan i promet. Kod algoritma nasumičnih šuma skup sa svim podacima, tj. skup sa lokalnim i satelitskim meteorološkim podacima i prometom se pokazao najboljim, dok je kod algoritma neuronskih mreža taj skup bio nešto lošiji od skupa s lokalnim meteorološkim podacima i prometom. Razvijeni modeli pokazali su zadovoljavajuću točnost te se u budućnosti mogu primijeniti za predviđanje koncentracije lebdećih čestica na novim skupovima podataka.

Abstract (english)

The goal of this study is the application of artificial intelligence methods to develop models for estimating the concentration of particle matter in the air. Random forest regression and neural networks models were developed using the Python programming language and its libraries. Meteorological data used in this study is related to the Austrian city of Graz measured at five stations, namely: North, East, West, South and Don Bosco. In addition to meteorological data, the concentration of particle matter in the air is also measured at the stations. Additionally, satellite data from the Climate Data store is also used. Using these data, models were developed for all the stations and it was observed which data would contribute more to the final quality of the model. It has been proven that models with locally acquired meteorological data are better than models with satellite meteorological data. Furthermore, traffic data obtained by measuring traffic on sixteen roads in the city is also used in this study. Traffic data has been added to the set with local and satellite meteorological data. It has been proven that traffic as an additional variable improved the quality of the model’s assessments. This increase was more pronounced in the set with local data. In the end, all the above data is combined into two sets, where the first set consisted of local and satellite meteorological data, and the second set consisted of local and satellite meteorological data to which traffic was also added. With the random forest algorithm, the model developed using the set with all the data, i.e. the local and satellite meteorological with traffic, turned out to be the best. However, with the neural network algorithm, the model using all data show slightly worse results in comparison with neural network model developed on set with local meteorological data and traffic. The developed models showed satisfactory accuracy and can be applied to estimating the concentration of particle matter in the air using the new data set.

• Keywords: lebdeće čestice; strojno učenje; Graz; zaštita okoliša; meteorološki podaci; promet
• Keywords (english): particulate matter; machine learning; Graz; environmental protection; meteorological data; traffic
• Language: croatian
• URN:NBN: urn:nbn:hr:149:596824
• Study programme: Title: Chemical Engineering - Graduate study; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka kemijskog inženjerstva (magistar/magistra inženjer/inženjerka kemijskog inženjerstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2023-03-10 13:42:18