Master's Dissertation
DOI
https://doi.org/10.11606/D.44.2021.tde-16072021-105728
Document
Author
Full name
Karla Ximena Morales Rodriguez
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2021
Supervisor
Committee
Rocha, Marcelo Monteiro da (President)
Avansi, Michelle Chaves Kuroda
Carneiro, Cleyton de Carvalho
Title in Portuguese
Avaliação de diferentes medidas de similaridade no SIMDISPAT - Novo Algoritmo de Simulação de Múltiplos Pontos
Keywords in Portuguese
Estatística de múltiplos pontos (MPS)
Imagem de treinamento
Template
Abstract in Portuguese
Title in English
not available
Keywords in English
not available
Abstract in English
The Geostatistics techniques implemented in the modeling of spatial phenomena were initially based on probability functions built from two-point statistics. However, over the years, different limitations were found, mainly in the reproduction of structures and complex geological features. In search of a solution, different researchers used statistics of more than two points, called multipoint simulation (MPS). MPS employs the concept of Training Image (TI), which is a conceptual geological model that represents spatial continuity. The objective of this work is to develop a new simulation algorithm based on MPS. The proposed program is called SIMDISPAT, based on concepts from the SNESIM algorithm (Single Normal Equations Simulation) and on concepts from the SIMPAT algorithm (Simulation with Patterns). The algorithm was written in the R programming language. The SIMDISPAT (Simulations with Distance and Pattern) method simulates images with complex geological characteristics, either in 2D or 3D. To verify the algorithm's efficiency, four synthetic databases described in the literature are used (three two-dimensional and one three- dimensional). In addition, four similarity distances are tested, two widely applied in MPS - the Manhattan distance and the Euclidean distance - the other two distances are common in different areas of the sciences - the Lorentz distance and the Cosine distance. To compare the results obtained by the algorithm, connectivity analysis was used to verify which of the distances best reproduces the characteristics of each TI, in addition, a visual method called multidimensional scaling (MDS) was used to explore the structure of similarity data. SIMDISPAT is able to effectively reproduce the different characteristics of the TI, with the distances of Manhattan, Euclidean and Lorentz, but at the Coseno distance it presents problems when reproducing the TI patterns. It is noteworthy that the Lorentz distance, which is not used in the MPS, satisfactorily reproduces the different characteristics of TI.