Title Razvoj modela umjetnih neuronskih mreža za predviđanje emisija stakleničkih plinova s obzirom na sektorsku potrošnju energije u Republici Hrvatskoj
Title (english) Development of artificial neural network prediction model of greenhouse gas emissions based on sectoral energy consumption in Croatia
Author Tomislav Strahovnik
Mentor Tomislav Bolanča (mentor)
Mentor Marko Rogošić (mentor)
Committee member Vesna Tomašić (predsjednik povjerenstva)
Committee member Šime Ukić (član povjerenstva)
Committee member Mario Šiljeg (član povjerenstva)
Granter University of Zagreb Faculty of Chemical Engineering and Technology Zagreb
Defense date and country 2017-11-17, Croatia
Scientific / art field, discipline and subdiscipline TECHNICAL SCIENCES Interdisciplinary Technical Sciences Environmental Engineering
Universal decimal classification (UDC ) 66 - Chemical technology. Chemical and related industries 502/504 - Environmental science. Conservation of natural resources. Threats to the environment and protection against them
Abstract Predmet istraživanja u ovoj disertaciji je pronalaženje i razvoj optimalnog modela predviđanja emisija stakleničkih plinova s obzirom na realne projekcije sektorske potrošnje energije u Republici Hrvatskoj do 2030. U radu je u obzir uzeta samo potrošnja energije u energetskom sektoru, koji se sastoji od podsektora: industrija, promet, kućanstva, usluge, poljoprivreda i graditeljstvo, a predstavlja emisije stakleničkih plinova koje nastaju isključivo izgaranjem fosilnih goriva. Prvi korak bio je korigiranje dvaju postojećih scenarija neposredne potrošnje energije iz Strategije energetskog razvoja Republike Hrvatske iz 2009. godine. To su dva scenarija, tzv. BAU-scenarij (engl. Business-As-Usual) i održivi scenarij. Korekciju je bilo potrebno provesti s obzirom na realne i ostvarene iznose neposredne potrošnje energije u razdoblju nakon izbijanja gospodarske krize. Kriza se naime između ostalog očitovala i u smanjenju gospodarskih aktivnosti što je utjecalo i na pad neposredne potrošnje energije. Projekcije sektorske potrošnje energije za ciljanu godinu određene su ekstrapolacijom procijenjenih realnih godišnjih stopa rasta preuzetih za održivi scenarij iz Strategije energetskog razvoja Republike Hrvatske te je predložen novi scenarij potrošnje, tzv. korigirani BAU-scenarij koji uzima u obzir recentne gospodarske pokazatelje te prema kojem bi u 2020. neposredna potrošnja energije iznosila 293,33 PJ, a u 2030. 382,60 PJ. Na osnovi procijenjenih realnih godišnjih stopa rasta potrošnje energije koje su za 20 % umanjene u odnosu na stope iz korigiranog BAU-scenarija u razdoblju do 2020., a u razdoblju od 2021. do 2030. za 30 % umanjene u odnosu na stope iz korigiranog BAU-scenarija, predložen je i novi korigirani održivi scenarij prema kojem bi u 2020. neposredna potrošnja energije iznosila 283,11 PJ, a u 2030. 340,78 PJ. Pored dva korigirana scenarija neposredne potrošnje energije, a s ciljem utvrđivanja učinaka mogućih varijacija u projekcijama neposredne potrošnje energije na kretanje emisije stakleničkih plinova, u ovom su radu dodatno kreirane i četiri podvarijante korigiranog održivog scenarija s pretpostavljenim odstupanjima u neposrednoj potrošnji energije u rasponu od ± 20 %, odnosno 10 %. Za razvoj modela umjetnih neuronskih mreža za predviđanje emisija stakleničkih plinova testirane su dvije najčešće korištene arhitekture umjetnih neuronskih mreža, višeslojne perceptronske mreže i mreže s funkcijama s kružnom osnovicom. Prilikom traženja optimalne arhitekture neuronske mreže mijenjano je više različitih izvedbenih karakteristika, koji se očituju u različitim algoritmima za treniranje te u broju neurona u skrivenom sloju koji je varirao od 2 do 30. Kao ulazni podaci u modele umjetnih neuronskih mreža korištene su projekcije sektorske potrošnje energije, a izlazne vrijednosti bile su emisije stakleničkih plinova. Kao optimalne pokazale su se troslojne umjetne neuronske mreže s unaprednim vezama, sa šest neurona u skrivenom sloju. Kao optimalna za treniranje mreže pokazala se Levenberg-Marquardtova metodologija s Bayesovom regularizacijom. Rezultati pokazuju da se troslojne umjetne neuronske mreže mogu uspješno primijeniti za koreliranje postojećih podataka. Što se tiče predviđanja, korigirani BAU-scenarij pokazao je kontinuirano povećanje svih emisija stakleničkih plinova. Korigirani održivi scenarij pokazao je smanjenje razina emisija svih stakleničkih plinova u odnosu na korigirani BAU-scenarij. Promatrano smanjenje može se pripisati skupini mjera energetske učinkovitosti kojima se utječe na smanjenje neposredne potrošnje energije. Na osnovi usporedbe relativnog smanjenja emisija za svaki pojedini staklenički plin i za svaki pojedini scenarij i podvarijantu može se zaključiti kako se iznos maksimalnog relativnog smanjenja emisija kao i vrijeme kada se taj maksimum očekuje razlikuju. Model je pokazao da će se najveće smanjenje emisija u odnosu na korigirani BAU-scenarij ostvariti kod podvarijante 4 korigiranog održivog scenarija, a da će mjere za smanjenje emisija najprije početi djelovati kod podvarijante-1 korigiranog održivog scenarija. Glavno pitanje raspravljeno u disertaciji odnosi se na primjenjivost razvijenog modela za predviđanje emisija stakleničkih plinova s obzirom na projekcije potrošnje energije. Drugo važno pitanje jest kakav će biti utjecaj predloženih novih scenarija potrošnje energije na emisije stakleničkih plinova. Iako je Hrvatska već ispunila svoje ciljeve prema odredbama Sporazuma iz Kyota kao i ciljeve EU do 2030., upitna je cijena toliko željenog gospodarskog razvoja. Pred Hrvatskom je donošenje odluke u kojem od dva pravca želi ostvariti svoj dugoročni gospodarski, a time i energetski razvoj. Prvi pravac predstavlja ulaganje u daljnji razvoj obnovljivih izvora energije i primjenu čistih tehnologija u cilju ostvarivanja niskougljičnog i održivog razvoja, koje predlaže Strategija niskougljičnog razvoja RH, što u principu znači uvoz novih, ali i znatno skupljih tehnologija. Drugi pravac predstavlja poticanje jeftinijih fosilnih goriva i pripadajućih tehnologija koje su naravno štetnije po okoliš, ali čija je cijena znatno niža. Ukoliko Hrvatska izabere drugu mogućnost i na taj način postigne određeni industrijski rast te veću potrošnju fosilnih goriva, pitanje je hoće li i tada ispunjavati preuzete obveze i ciljeve prema važećim zakonima u EU. S ciljem stvaranja efektivne ravnoteže između gospodarskog oporavka i zaštite okoliša smatram da bi Hrvatska trebala u prvoj fazi nastaviti koristiti "prljavije" izvore energije kako bi ostvarila cjenovno konkurentnije gospodarstvo, a u drugoj fazi postupno prelaziti na skuplje, ali i po okoliš čišće obnovljive izvore energije i tehnologije koje će omogućiti prelazak s gospodarstva temeljenog na fosilnim gorivima na niskougljično gospodarstvo i održivi razvoj. Primjena ispitanog modela umjetnih neuronskih mreža za predviđanje emisija stakleničkih plinova pokazala se prikladnom s obzirom na utrošeno vrijeme kao i na preciznost primijenjenih ulaznih podataka, te bi zasigurno uvelike skratila vrijeme potrebno za donošenje smjernica potrebnih za novu strategiju energetskog razvoja.
Abstract (english) The subject of research in this dissertation is to find and develop the optimal model for predicting greenhouse gas emissions with regard to realistic projections of sectoral energy consumption in the Republic of Croatia by 2030. The research took into account only the energy consumption in the energy sector, which consists of sub-sectors: industry, transport, households, services, agriculture and construction, and represents the greenhouse gas emissions that arise solely from burning fossil fuels. The first step was correction of the two existing scenarios of final energy consumption from the 2009 Energy Development Strategy of the Republic of Croatia. These two scenarios are the so-called BAU (Business-As-Usual) scenario and sustainable scenario. The correction was necessary with regard to real and realized amounts of final energy consumption in the period after the outbreak of an economic crisis. The crisis was, among other things, manifested in the reduction of economic activities, which also affected the decline of final energy consumption. Projections of the sectoral energy consumption for the target year were determined by extrapolation of the estimated real annual growth rates assumed for a sustainable scenario from the Energy Development Strategy of the Republic of Croatia and a new consumption scenario was proposed, the so-called corrected BAU scenario that takes into account recent economic indicators, according to which the final energy consumption in 2020 and 2030 would be 293.33 PJ and 382.60 PJ, respectively. Based on the estimated real annual energy consumption growth rates that are reduced by 20% in relation to the rates from the corrected BAU scenario in the period up to 2020, and in the period 2021-2030 reduced by 30% compared to the rates from the corrected BAU scenario, a new corrected sustainable scenario was proposed, according to which in 2020 and 2030 the final energy consumption would be 283.11 PJ and 340.78 PJ, respectively. In addition to the two corrected scenarios of final energy consumption, in order to determine the effects of possible variations in projections of final energy consumption on the variation of greenhouse gas emissions, four additional subvariants of corrected sustainable scenarios were created with deviations of final energy consumption ranging from ± 20%. To develop a model of artificial neural networks (ANN) for predicting greenhouse gas emissions, the two most commonly used architectures of artificial neural networks, multilayer perceptron networks and radial basis function networks were tested. When looking for the optimal architecture of neural network, several different performance characteristics were altered, i.e. training algorithms were varied and the number of neurons in the hidden layer was scanned in the range from 2 to 30. Sectoral energy consumption projections were used as inputs in artificial neural network models; the output values were greenhouse gas emissions. The optimal architecture was found to be the three-layer feed-forward neural network with six neurons in the hidden layer. The Levenberg-Marquardt methodology with Bayesian regularisation was proved to be the optimal for ANN training. The results showed that those three-layer artificial neural networks could be successfully applied to correlate existing data. As for the prediction, the corrected BAU scenario was shown to produce a continuous increase of all greenhouse gas emissions. The corrected sustainable scenario was shown to produce a reduction of all greenhouse gases emissions with respect to the corrected BAU scenario. The observed decrease could be attributed to a group of energy efficiency measures which could affect the reduction of final energy consumption. Based on the comparison of relative emission reduction for each individual greenhouse gas and for each scenario and sub-variant, the variation of the maximum relative emission reduction was observed as well as the variation of time of the occurrence of the maximum. The model showed that the largest emission reduction with respect to the corrected BAU scenario would be achieved with subvariant 4 of the corrected sustainable scenario. In addition, it was shown that that emission reduction measures will first begin to act with subvariant-1 of the corrected sustainable scenario. The main issue discussed in the dissertation is linked to the applicability of the developed model for predicting greenhouse gas emissions with regard to energy consumption projections. Another important issue is the probable impact of the proposed new energy consumption scenarios on greenhouse gas emissions. Although Croatia has already fulfilled its goals under the provisions of the Kyoto Protocol as well as the EU goals by 2030, the price of so much desired economic development is questionable. Croatia has to make a decision about the future direction of its long-term economic and energetic development, and it has to be done very soon. The first path is to invest in the further development of renewable energy sources and the application of clean technologies in order to achieve low-carbon and sustainable development, as proposed by the Low-carbon Development Strategy of the Republic of Croatia. This in principle means the import of novel but expensive technologies. The second is the promotion of cheaper fossil fuels and associated technologies that are of course more harmful to the environment, but whose prices are considerably lower. In case of Croatia choosing the second option and thus achieving substantial industrial growth and greater consumption of fossil fuels, the question arises whether it will meet the assigned commitments and targets under applicable EU legislation. In order to create an effective balance between economic recovery and environmental protection, it might be concluded that Croatia has to continue to use "dirtier" energy sources in the first period to achieve a more competitive economy. In the second period the steps towards the transition to more expensive and cleaner renewable energy sources technologies are to be gradually applied that would enable the smooth transition from the fossil fuel economy to the low-carbon economy and sustainable development. The application of the tested model of artificial neural networks for predicting greenhouse gas emissions was shown to be appropriate with regard to the time spent as well as the accuracy of the input data used and it would certainly greatly shorten the time needed to make the guidelines for creating new energy development strategy.
Keywords
staklenički plinovi
neposredna potrošnja energije
umjetne neuronske mreže
Strategija energetskog razvoja Republike Hrvatske
Levenberg-Marquardtova metodologija s Bayesovom regularizacijom
Keywords (english)
greenhouse gases
final energy consumption
artificial neural networks
Energy Development Strategy of the Republic of Croatia
Levenberg-Marquardt methodology with Bayesian regularisation
Language croatian
URN:NBN urn:nbn:hr:149:895680
Promotion 2017
Study programme Title: Engineering Chemistry - Doctoral Course Study programme type: university Study level: postgraduate Academic / professional title: doktor/doktorica znanosti, područje prirodnih znanosti, polje kemija (doktor/doktorica znanosti, područje prirodnih znanosti, polje kemija)
Type of resource Text
Extent 136 str. ; 30 cm
File origin Born digital
Access conditions Open access
Terms of use
Created on 2023-12-06 11:56:30