Kartiranje ležišta ugljikovodika metodom umjetnih neuronskih mreža

master's thesis

• Title: Kartiranje ležišta ugljikovodika metodom umjetnih neuronskih mreža : diplomski rad
• Title (english): Mapping of hydrocarbon reservoirs using artificial neural networks : master’s thesis
• Author: Marijan Šapina
• Mentor: Tomislav Malvić (mentor)
• Committee member: Tomislav Malvić (predsjednik povjerenstva)
• Committee member: Josipa Velić (član povjerenstva)
• Committee member: Zdenko Krištafor (član povjerenstva)
• Granter: University of Zagreb; Faculty of Mining, Geology and Petroleum Engineering; (Department of Petroleum and Gas Engineering and energy); Zagreb
• Defense date and country: 2016-06-24, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mining, Petroleum and Geology Engineering; Petroleum Mining

Abstract

Računalno kartiranje ležišta ugljikovodika moguće je raznim metodama i tehnikama, primjerice
metodama geostatističke procjene, ali i geomatematičkim metodama, kao što su primjena umjetnih
neuronskih mreža. Metoda umjetnih neuronskih mreža se primarno koristi u slučaju manjkavih
podataka i podataka u nelinearnom odnosu, a karakteristika te metode je da simulira postupak
ljudskog učenja uvježbavanjem parametara nizom ponavljanja. U ovomu radu je statističkom
analizom i usporedbom s geostatistiĉkim interpolacijskim metodama potvrđena postavljena
hipoteza o uspješnom kartiranju izvornim algoritmom umjetne neuronske mreţe izrađenom u
programu za statističku obradu podataka „R“ na primjeru kartiranja dubine EK markera
(elektrokarotaţnog markera) Rs5 u području Bjelovarske subdepresije unutar Dravske depresije u
hrvatskom dijelu Panonskog bazenskog sustava. Arhitektura mreže koja je dala najbolji izlaz je
mreža s dva skrivena sloja, od kojih se prvi sastojao od 10 neurona, a drugi od 6. Korišten je
algoritam s povratnim postupkom. Kros-validacijom su uspoređene dvije metoda kartiranja, s time
da je algoritmom neuronske mreže dobivena vrijednost sume kvadrata pogrješke 16294,5, a
običnim krigiranjem 14638,35.

Abstract (english)

Abstract: Computerized mapping of hydrocarbon reservoirs can be achieved using a wide range of methods and techniques, for example, geostatistical interpolation. The use of geomathematical methods is also possible, for example, the use of artificial neural networks. Artificial neural networks are primarily used in case of flawed data and data that are in a non-linear relation. The characteristic of this method is that it simulates the process of human learning by training and optimizing parameters through a series of repetitions. In this thesis, using statistical analysis and comparison with geostatistical interpolation methods, the hypothesis of successful mapping using this original artificial neural network algorithm has been confirmed. The algorithm has been made in “R“, a statistical computing software, and is used for mapping of depth of e-log border “Rs5“ in the Bjelovar Sub-depression, as a part of the Drava Depression in the Croatian part of the Pannonian Basin System. The neural network architecture that produced best responses is a network with two hidden layers, with 10 and 6 neurons, respectively. Back-propagation algorithm was used. Two methods were compared by cross-validation and the neural network produced a mean squared error as 16294.5, and Ordinary Kriging produced 14638.35.

• Keywords: dravska depresija; geostatistika; kartiranje; povratni postupak; umjetne neuronske mreže
• Keywords (english): artificial neural networks; backpropagation; Drava depression; geostatistics; mapping
• Language: croatian
• URN:NBN: urn:nbn:hr:169:369345
• Study programme: Title: Petroleum engineering; specializations in: Petroleum engineering, Energy production and supply, Environmental protection in petroleum engineering; Course: Energy production and supply; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka naftnog rudarstva (magistar/magistra inženjer/inženjerka naftnog rudarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2016-08-31 12:24:36