Date of Award

January 2019

Document Type


Degree Name

Doctor of Philosophy (PhD)


Petroleum Engineering

First Advisor

Vamegh Rasouli


The Three Forks Formation, as the lower part of the Bakken petroleum system is a complex reservoir with variable mineralogy, thin bed characteristics, and low permeability. Advanced logging tool and techniques are required to characterize and estimate water saturation (Sw), porosity, and mineralogy in this type of formations.

In this research, to overcome these challenges, we used three different methods to estimate Sw. The first model was based on an integrated petrophysical workflow developed to evaluate the reservoir quality. In the second approach, Sw was estimated from dielectric measurements, which is independent of resistivity. The two models showed good agreements with core measurement results. In a third attempt, the application of machine learning and deep learning algorithms were applied to estimate Sw. This was with the aim of generalizing the results to the entire extent of the Three Forks in the Williston Basin. The results suggest the use of the three algorithms (support vector machine regression, random forest regression, and backpropagation neural network) complementary to each other for Sw estimation. These methods captured the complexity of the Three Forks Formation where the laminations are in abundance with a complex pore size distribution.

On the other hand, the NMR T2 Log Mean was applied to investigate the pore size distribution and its relation with Sw. The average T2 Log Mean values of equal to or greater than 8 msec was defined as a cutoff corresponding to oil-bearing interval within the Three Forks Formation.

In the second phase of the research, series of laboratory experiments were carried out. The results showed that Three Forks Formation is composed of dolostones, quartz, feldspars, calcite, pyrite, anhydrite, and clay minerals mainly illite. The dolomite is the most abundant mineral. Also, the hysteresis response of core permeability and porosity as a function of net effective stress was used to develop stress-dependent permeability and porosity relationship in a wide range of stresses.

Finally, to define the rock types’ classification based on well logs and routine core data, in addition to the deterministic models, neural network technique, based on Indexation and Probabilistic Self-Organizing Map, was used to integrate the carbonate reservoir heterogeneity and geological properties in rock types characterization. Accordingly, six rock types were defined and the results showed that the PRT1 and PRT2 rock types appear to be the best reservoir zones in Three Forks Formation.