Petroleum potential assessment using an optimized fuzzy outranking approach: A case study of the Red River petroleum system,Williston Basin

Abstract

This paper presents a new approach, called optimized fuzzy Preference Ranking Organization METHod for Enrichment Evaluation (PROMETHEE), based on combining fuzzy logic, an outranking method and the cuckoo search optimization algorithm to assess petroleum potential in a spatial framework. The approach brings together flexibility and simplicity to solve petroleum exploration problem under uncertainty using experts’ knowledge and the information associated with the discovered oil pools simultaneously. The characteristics of the essential elements of the petroleum system are used as key criteria in the model. To exemplify the approach, a case study was undertaken in the Red River petroleum system of the Canadian portion of the Williston Basin. Eight datasets related to the selected criteria were integrated by the optimized fuzzy PROMETHEE to create a map that makes it possible to identify the areas of highest petroleum potential. All discovered oil pools in the Red River petroleum system were used in the verification process. The results indicated that the proposed approach can deal effectively with incomplete data and imprecise information, and can be efficiently used in petroleum exploration, thereby reducing the cost and risk of exploration.

Keywords

Red River petroleum system Williston Basin petroleum potential assessment outranking fuzzy PROMETHEE optimization cuckoo search algorithm

Introduction

The primary objective of exploration in the petroleum industry is to find economically recoverable reserves. Petroleum exploration is a costly process which involves a great deal of complexity and uncertainty. A decision to drill in an area without careful consideration of the petroleum potential assessment would most likely result in a dry hole. So, potential modeling is a crucial step in petroleum exploration. The evaluation of the potential of undiscovered petroleum accumulations involves integrating surface and subsurface datasets, including geological, geochemical, geophysical, petrophysical and geographical data.

Existing literature provides a number of methods for modeling petroleum potential in a spatial framework, which fall in three general categories: data-driven, knowledge-driven and hybrid approaches. On the one hand, spatial associations between discovered reservoirs and indirect evidence of the petroleum potential are quantitatively analyzed by data-driven methods (Carranza et al., 2008a; Coolbaugh et al., 2002; Magalhães and Filho, 2012). The extracted spatial relationships are then used to estimate the parameters of model by which input datasets are integrated to produce a petroleum potential map (Abedi et al., 2012a, 2012b; Blewitt et al., 2002; Carranza and Hale, 2002). There are several data-driven methods that can be used for modeling the petroleum potential of a region of interest, such as adaptive neuro-fuzzy inference system (ANFIS), neural networks (ANNs), Bayesian classifiers, evidential belief functions (EBFs), gene expression programming (GEP), logistic regression, support vector machines (SWMs) and weights of evidence. On the other hand, knowledge-driven methods do not require spatial analysis of discovered petroleum pools within a region of interest. They involve application of expert knowledge to assign the parameters of model (Carranza et al., 2008b; Kraipeerapun et al., 2005). Fuzzy logic, Boolean logic, outranking techniques, the Dempster-Shafer belief theory and index overlay are some examples of these methods.

In the past few years, a number of studies have been performed on the spatial or non-spatial assessment of petroleum potential using data-driven, knowledge-driven and hybrid approaches. Zargani et al. (2003) used weights of evidence method to integrate the main geological factors controlling petroleum occurrence in a geographic information system (GIS) environment in order to identify the areas with the highest petroleum potential for future exploration. The results were somewhat encouraging and 64% of the discovered petroleum accumulations of the Murzuq Basin of Libya were located in the favorable regions of the resulting posterior probability map. Chen and Osadetz (2005) simulated the spatial distribution of undiscovered oil and gas accumulations by a fractal model. The results of the simulation indicate that the spatial characteristics of the accumulations in the Rainbow play of the Western Canada Sedimentary Basin (WCSB) are fractal and there are areas with high potential for more investigation in the southwest and northeast of the play. An expert system entitled approximate fuzzy assessment (AFA) on the strength of fuzzy set and possibility theories for the evaluation of oil potential in the Hassi Messaoud fields of Algeria was developed by Tounsi (2005). Source rock, reservoir, trap and seal as the essential elements of a petroleum system were considered in the AFA expert system. Although the system could handle data with uncertainty, the petroleum potential was investigated in a non-spatial framework. Chen and Osadetz (2006b) proposed a model-based simulation approach using the Fourier transform algorithm to produce a petroleum accumulation map for the Rainbow play of the WCSB. The pre-1994 exploratory datasets were used for estimating the model parameters and conditioning the simulation. The final accumulation map successfully predicted the location of 68% of the post-1993 discovery wells. A GIS-based fuzzy multi-criteria evaluation approach for petroleum exploration in northern South America was developed by Bingham et al. (2012). They validated the results by comparing the favorability maps to the discovered fields and concluded that the approach is satisfactory for assessing the petroleum potential in frontier areas because it does not rely on prior probabilities. Two data-driven methods, the frequency ratio and EBFs, were used by Arab Amiri et al. (2015a) to evaluate the spatial distribution of oil pools in the Red River petroleum system of the Williston Basin. Seven sub-criteria of the elements of the petroleum system were selected as input datasets. The frequency ratio method showed slightly better performance in the petroleum potential modeling than EBFs. In another study, Arab Amiri et al. (2015b) applied a GIS-based EBFs model to create a petroleum resources potential map for Khuzestan Province of Iran. Ziyong et al. (2017) employed a GIS-based fuzzy logic model to investigate the gas potential of the Lower Carboniferous of the Chu-Sarysu Basin in Kazakhstan. The geological, geophysical and simulation datasets were integrated using the gamma operator to delineate the areas with the highest gas potential. The predictions of the proposed model were validated by the well-logging interpretation.

Some studies have also been conducted on the spatial characteristics of undiscovered petroleum accumulations. In a study by Gao et al. (2000), for example, each of the gas pools was first parameterized by the geological properties, location, shape and exploration history data. An object model on the basis of the characteristics of each pool and relationships between pairs of pools was then used for predicting the spatial distribution of undiscovered natural gas resources in the Middle Devonian Rainbow reef play of the WCSB. A similar study was carried out by Chen et al. (2000) in the Keg River Reef play of the WCSB. They developed a stochastic model to determine the possible locations and sizes of undiscovered pools.

Furthermore, several spatial approaches such as the index overlay (Grant et al., 1996), multivariate and Bayesian statistical methods (Chen and Osadetz, 2006a), Monte-Carlo simulation and geostatistical techniques (Ruffo et al., 2006), object-based stochastic model (Chen et al., 2008), hybrid fuzzy-probabilistic method (Roisenberg et al., 2009), fuzzy-belief and Bayesian-belief networks (Martinelli, 2012; Xie et al., 2011) have been proposed for risk assessment in petroleum exploration.

Petroleum potential modeling is considered as a multiple criteria decision-making (MCDM) problem, since there are a finite number of spatial elements (pixels or voxels) as alternatives which have to be ranked considering several, sometimes conflicting, criteria. Among various methods of MCDM, the Preference Ranking Organization METHod for Enrichment Evaluation (PROMETHEE), as an outranking approach, is particularly suitable for ranking applications (Albadvi, 2004; Al-Shemmeri et al., 1997; Goumas and Lygerou, 2000; Ufuk Bilsel et al., 2006). The methods of PROMETHEE family are regarded as partially compensatory models (Geldermann et al., 2000) that are easily understood and implemented (Zhao et al., 2019). They are suitable when units are incomparable or incommensurate, criteria metrics are difficult to aggregate, and variety of measurement scales are used (Linkov et al., 2004). Furthermore, fuzzy logic (Zadeh, 1965) has been found to be a very efficient mathematical tool to deal with imprecise and incomplete data (Roisenberg et al., 2009) and is adequate to handle uncertainty in petroleum geology (Chen and Fang, 1993). Consequently, combining fuzzy logic and PROMETHEE, called fuzzy PROMETHEE (FP), provides an effective and flexible knowledge-driven mechanism to solve the petroleum exploration problem under uncertainty.

It should also be noted that most knowledge-driven approaches for petroleum potential assessment such as FP are characterized by complex functional relationships and a large number of parameters that must be specified by experts. In most cases, the parameters cannot be determined with precision and models may suffer from expert-elicited values, hence the need for parameter optimization. Parameter optimization is the process in which model parameters are tuned for a particular problem. Several computational intelligence methods, such as cuckoo search (CS), pattern search, particle swarm optimization (PSO), artificial bee colony (ABC), genetic algorithms (GA) and firefly, were proposed in the literature. These algorithms estimate relevant parameters by minimizing, or maximizing, the objective function in the optimization process. The findings of several preliminary studies (Gandomi et al., 2013; Kumar and Chakarverty, 2011; Tuba et al., 2011; Yang and Deb, 2010; Civicioglu and BesdokEmail, 2013) show that CS is very promising and efficient in solving optimization problems and could outperform other metaheuristic algorithms such as GA, ABC and PSO, because it has relatively few parameters that must be tuned by the user. In fact, besides the population size, there is basically one parameter $p_{a}$ that has to be set in the CS algorithm (Yang and Deb, 2009).

Considering the above, optimized fuzzy PROMETHEE (OFP) is presented for petroleum potential assessment, in which FP provides a robust decision-making framework that has the capability to handle data with uncertainty and the CS algorithm is used to optimize weights of the criteria. In order to exemplify FP and OFP, a case study was undertaken in one of the petroleum systems of the Williston Basin.

To the best of our knowledge, the OFP approach has not been previously applied to the petroleum exploration problem. The PROMETHEE methods are able to model real-world problems with crisp input values (Brans and Mareschal, 1994; Brans et al., 1984; Brans and Vincke, 1985), this is not the case in most petroleum exploration activities, especially in frontier areas, where information is uncertain, incomplete and imprecise. Hence, one of the novelties of the presented research work is to utilize the trapezoidal fuzzy (TF) numbers to determine the shape of the membership functions for petroleum potential assessment under uncertainty. Bias is a serious problem in oil and gas exploration. Biased expert-elicited values for parameters included in a petroleum potential model are significantly associated with unsuccessful outcomes such as an uneconomic reservoir or a dry hole. So, another important novelty of the study includes optimizing the experts-assigned weights of the criteria by the CS algorithm.

Materials and methods

The goal of the proposed methodology is to perform petroleum potential assessment in a spatial framework. An overview of the methodology is presented in Figure 1. Having identified the evaluation criteria based on the characteristics of the petroleum system elements, eight datasets related to the criteria were prepared and converted to the raster data model. Cell values of the raster datasets were then stored in a spatial database. The generalized preference functions and a vector containing the triangular fuzzy weights were defined by the exploration experts to apply FP. For OFP, in addition to these, it is necessary to specify the parameters of the CS algorithm by which the triangular fuzzy weights of the criteria are optimized. The values of the net outranking flow were computed for alternatives or pixels both by FP and OFP. The net outranking flow values resulting from each method were subsequently converted to a petroleum potential map. Each map was reclassified into the potential classes ranging from no- to high-potential. Finally, the prediction performance of both methods was evaluated by overlaying a vector layer of the discovered oil pools upon the reclassified maps.

Figure 1.

Overall flow of the proposed methodology.

PROMETHEE and FP

PROMETHEE is a subgroup of MCDM methods which was developed by Brans et al. (1984) in the 1980s. The method uses the outranking methodology to rank a finite number of alternatives in accordance with several criteria. The success of various ranking formats of PROMETHEE in different applications is attributed to its simplicity, clearness, stability and flexibility (Brans et al., 1986).

The methods of PROMETHEE family are based on a set of alternatives $A = \{a_{1}, a_{2}, \dots, a_{n}\}$ which will be evaluated according to a set of criteria $F = \{f_{1}, f_{2}, \dots, f_{m}\}$ . A pairwise comparison is performed between two actions $a_{i}$ and $a_{j}$ to determine the intensity of preference of $a_{i}$ over $a_{j} (p_{k} (d_{k}), d_{k} = f_{k} (a_{i}) - f_{k} (a_{j}))$ , where $f_{k} (a_{i})$ represents the evaluation of $a_{i}$ corresponding to criterion $f_{k}$ , and $p_{k}$ is the preference function for the $k$ th criterion. The range of the preference scale values varies from 0 to 1 (no preference to full preference). The preference of $a_{i}$ over $a_{j}$ is separately determined for each criterion, and then the outranking-relation $π$ is calculated as follows

\{\begin{matrix} π (a_{i}, a_{j}) = \sum_{k = 1}^{m} w_{k} . p_{k} (f_{k} (a_{i}) - f_{k} (a_{j})), \forall a_{i} \in A \\ π (a_{j}, a_{i}) = \sum_{k = 1}^{m} w_{k} . p_{k} (f_{k} (a_{j}) - f_{k} (a_{i})), \forall a_{j} \in A \end{matrix}

(1)

where

w_{k} \in W = \{w_{1}, w_{2}, \dots, w_{m}\}

is a weight which represents the relative importance of each criterion and

m

is the total number of criteria. The preference index

π (a_{i}, a_{j})

represents the preference degree of

a_{i}

over

a_{j}

and

π (a_{j}, a_{i})

indicates with what degree

a_{j}

is preferred to

a_{i}

over all the criteria. Clearly,

π (a_{j}, a_{i}) \sim 1

suggests a strong global preference of

a_{i}

over

a_{j}

, while

π (a_{i}, a_{j}) \sim 0

indicates a weak global preference of

a_{i}

over

a_{j}

(Brans and Mareschal, 1990). Each alternative

a_{i}

is facing

(n - 1)

other alternatives in

A

. So, the positive outranking flow

ϕ^{+} (a_{i})

and the negative outranking flow

ϕ^{-} (a_{i})

are defined as

\{\begin{matrix} ϕ^{+} (a_{i}) = \frac{1}{n - 1} \sum_{\binom{j = 1}{j \neq i}}^{n} π (a_{i}, a_{j}) \\ ϕ^{-} (a_{i}) = \frac{1}{n - 1} \sum_{\binom{j = 1}{j \neq i}}^{n} π (a_{j}, a_{i}) \end{matrix}

(2)

where

n

represents the total number of actions.

ϕ^{+} (a_{i})

indicates how

a_{i}

is outranking other actions and

ϕ^{-} (a_{i})

expresses how

a_{i}

is outranked by all the others.

If a complete pre-order is necessary, a total ranking method will be used based on the net outranking flow evaluation. To obtain the net outranking flow, the positive outranking flow is subtracted from the negative outranking flow

{ϕ {(a_{i}) = ϕ}^{+} (a_{i}) - ϕ}^{-} (a_{i}), \forall a_{i} \in A

(3)

The higher the net outranking flow $ϕ (a_{i})$ , the better the action.

The PROMETHEE methods were extended by Goumas and Lygerou (2000) to deal with fuzzy input data along with crisp weights. Further enhancements were made by Geldermann et al. (2000) using fuzzy preferences and weights. They proposed the use of TF numbers to represent the uncertainties. This approximation seems to be sufficient and comprehensible for decision makers (Rommelfanger, 1988). Using TF numbers offers a particular advantage: it allows to represent an interval, a triangular fuzzy number and a specific value by a specific TF number (Zhang et al., 2009). The membership function for the TF intervals (Dubois and Prade, 1980; Rommelfanger, 1988, 1996) can be formulated as

μ (x) \{\begin{array}{l} 0 for x \leq m_{ℓ} - α or m_{u} + β \leq x \\ 1 - \frac{m_{ℓ} - x}{α} for m_{ℓ} - α < x < m_{ℓ} \\ \begin{array}{l} 1 for m_{ℓ} \leq x \leq m_{u} \\ 1 - \frac{x - m_{u}}{β} for m_{u} < x \leq m_{u} + β \end{array} \end{array}

(4)

where

β

and

α

are the right and left spreads of the TF interval

[m_{ℓ}; m_{u}]

. In this interval,

m_{ℓ}

and

m_{u}

are the lower and upper bounds of the numbers that certainly belong to the set of available values (Geldermann et al., 2000). This TF interval is represented by

\tilde{M} = ({m_{ℓ}; m_{u}; α; β)}_{L R}

. A triangular fuzzy number is a special case of the TF interval with

m_{ℓ} = m_{u}

, and a crisp number

n

is expressed with

m_{ℓ} = m_{u} = n

and

α = β = 0

. The FP algorithm can be summarized in the following steps:

Step 1: A generalized preference function $p_{k} (d)$ is specified for each criterion $f_{k}$ .

Step 2: A vector which contains the TF weights is defined.

{\tilde{w}}^{T} = [{\tilde{w}}_{1}; \dots; {\tilde{w}}_{k}] with {\tilde{w}}_{k} = (m_{ℓ}^{w}; m_{u}^{w}; α^{w}; β^{w})_{L R}

(5)

Step 3: The fuzzy outranking-relation $\tilde{π}$ is defined for all the alternatives $a_{i}$ , $a_{j} \in A$

\tilde{π} (a_{i}, a_{j}) = \sum_{k = 1}^{m} {\tilde{w}}_{k} \otimes p_{k} ({\tilde{f}}_{k} (a_{i}) ⊖ {\tilde{f}}_{k} (a_{j}))

(6)

with ${\tilde{f}}_{k} (a_{i}) = ({m_{ℓ}; m_{u}; α; β)}_{L R}$ and ${\tilde{f}}_{k} (a_{j}) = ({n_{ℓ}; n_{u}; γ; δ)}_{L R}$ . The preference degree for the comparison of $a_{i}$ and $a_{j}$ regarding to the criterion $f_{k}$ can be expressed as follows

p_{k} ({\tilde{f}}_{k} (a_{i}) ⊖ {\tilde{f}}_{k} (a_{j})) = (m_{ℓ}^{p k}; m_{u}^{p k}; α^{p k}; β^{p k})_{L R}

(7)

where

m_{ℓ}^{p k} = p_{k} (m_{ℓ} - n_{u})

m_{u}^{p k} = p_{k} (m_{u} - n_{l})

α^{p k} = p_{k} (m_{ℓ} - n_{u}) - p_{k} (m_{ℓ} - n_{u} - α + δ)

and

β^{p k} = p_{k} (m_{u} - n_{l} + β + γ) - p_{k} (m_{u} - n_{l})

The preference degrees are then multiplied by the weight of each criterion. Accordingly, the fuzzy outranking-relation $\tilde{π}$ is calculated by

\tilde{π} (a_{i}, a_{j}) = (m_{ℓ}^{π}; m_{u}^{π}; α^{π}; β^{π})_{L R}

(8)

where

m_{ℓ}^{π} = \sum_{k = 1}^{m} {(m}_{ℓ}^{w k} . m_{ℓ}^{p k})

m_{u}^{π} = \sum_{k = 1}^{m} {(m}_{u}^{w k} . m_{u}^{p k})

α^{π} = \sum_{k = 1}^{m} {(m}_{l}^{w k} α^{p k} + m_{l}^{p k} α^{w k} - α^{w k} α^{p k})

and

β^{π} = \sum_{k = 1}^{m} {(m}_{u}^{w k} β^{p k} + m_{u}^{p k} β^{w k} - β^{w k} β^{p k})

Step 4: The fuzzy positive and negative outranking flows are calculated for each alternative $a_{i} \in A$ as

\{\begin{matrix} {\tilde{ϕ}}^{+} (a_{i}) = \frac{1}{n - 1} \sum_{\binom{j = 1}{j \neq i}}^{n} \tilde{π} (a_{i}, a_{j}) \\ {\tilde{ϕ}}^{-} (a_{i}) = \frac{1}{n - 1} \sum_{\binom{j = 1}{j \neq i}}^{n} \tilde{π} (a_{j}, a_{i}) \end{matrix}

(9)

Step 5: Finally, the fuzzy scores are combined and the results are defuzzified and compared/ranked (Geldermann et al., 2000).

The CS algorithm

The CS is a metaheuristic optimization method which was developed by Yang and Deb (2009). The idea of the method is based on the brooding parasitism of some cuckoo species (Fister et al., 2014; Yang and Deb, 2014). The explorative random walk (global search) and exploitative random walk (local search) are the search capabilities of the CS algorithm, which are both controlled by a switching/discovery probability ( $p_{a}$ ). The exploitative random walk is intensive and accounts for about 25% of the total search time ( $p_{a} = 0.25$ ), whereas the explorative random walk takes about three times as long as the local search. This leads to the efficient exploration of the search space on the global scale, and accordingly, the global optimality can be obtained with higher probability (Yang and Deb, 2014). Yang and Deb (2009) used following idealized rules to apply CS algorithm as an optimization tool:

One egg (a set of solution co-ordinates) is laid randomly in other bird species’ nests by each cuckoo at a time;

The nests with high-quality eggs, or solutions, are carried forward to the next generation;

The number of host nests available for laying is fixed. A host bird can discover the alien egg with a probability $p_{a}$ lying between 0 and 1. If the egg is discovered, it can be either thrown away by the host bird, or the nest is abandoned and a new nest is built elsewhere.

According to the above rules, the CS algorithm can be summarized in the following steps (Figure 2).

Figure 2.

Pseudocode of the CS algorithm.

In CS, a pattern (i.e. solution) is associated with a host nest. Every pattern has individual attributes, each of which corresponds to a cuckoo egg (Hanoun et al., 2012). Lévy flights are used to carry out the global random walk as follows

X_{g + 1; i} = X_{g; i} + α \otimes L é v y (β)

(10)

where

g = (1, 2, \dots, MaxGeneration)

i = (1, 2, \dots, n)

\otimes

and

α > 0

denote the generation number, pattern number, entry-wise multiplication and step size scaling factor, respectively. The factor

α

must be associated with the scales of the optimization problem at hand (Yang and Deb, 2009). The

j

th attributes of the

i

th pattern are initiated by

X_{g = 0; j, i} = rand \cdot ({U b}_{i} - {L b}_{i}) + {L b}_{i}

(11)

where

{U b}_{i}

and

{L b}_{i}

are the upper and lower boundaries of the

j

th attributes, respectively. The value of each attribute is checked by the CS algorithm in every computation step. If the value exceeds the allowed search range, it will be replaced by the value of corresponding boundary (Ma et al., 2013).

Before starting iterative search process, the most successful pattern is detected as $x_{best}$ pattern by the CS algorithm. Among the various algorithms proposed for generating Lévy flights, the Mantegna’s (1994) algorithm works well in most of the optimization studies (Yang and Deb, 2009, 2010; Yang et al., 2012). Accordingly, the evolution phase of the pattern is initialized by detecting $ϕ$ (Gandomi et al., 2013), which is given as follows

ϕ = ({\frac{Γ (1 + β) \cdot \sin (π \cdot / 2)}{Γ (((1 + β) / 2) \cdot β \cdot 2^{(β - 1) / 2})})}^{1 / β}

(12)

where

Γ

denotes gamma function and

β

is the Lévy distribution parameter. According to Lévy distribution, the value

β = 1.5

is commonly used for the step size (Brajevic and Tuba, 2014).

After defining the donor vector $v$ (where $v = x_{i}$ ), the evolution phase of the $x_{i}$ pattern can be started. The following equation calculates the required step size of the $j$ th attributes

s_{j} = 0.01 \cdot ({\frac{u_{j}}{v_{j}})}^{1 / β} \cdot (v - x_{best})

(13)

where

v = randn [D]

and

u = ϕ \cdot randn [D]

. A uniform integer in the interval

[1, D]

is generated by the function

randn [D]

, and the donor pattern

v

is then randomly adjusted by

v = v + s_{j} \cdot randn [D]

(14)

The fitness of the random pattern is evaluated by the CS algorithm. The $x_{best}$ pattern is updated if a better solution is found during the evaluation. The infeasible patterns or solutions are revised by the crossover operator as follows

v_{i} = \{\begin{array}{l} x_{i} + rand \cdot (x_{r 1} - x_{r 2}), {if rand}_{i} > p_{a} \\ x_{i}, otherwise \end{array}

(15)

Location and geology

The Williston Basin is a roughly circular, indistinctly bounded, shallow depression on the North American Craton that contains numerous oil-rich formations (Gerhard et al., 1990; Heck et al., 2002; Ruzyla and Friedman, 1985). It covers approximately 300,000 square miles (Pollastro et al., 2013) across two provinces (Manitoba and Saskatchewan) in Canada and three states (Montana, North and South Dakota) in the United States. To the northwest, the basin is separated from the Alberta Basin by the Sweetgrass Arch and the Bow Island Arch; to the south, by the Transcontinental Arch and Black Hills Uplift; and, to the north and east, by the Canadian Shield (Flannery, 2006; Hitchon, 1996). Figure 3 shows the extent of the Williston Basin with the major geologic elements that bound it.

Figure 3.

Location map showing the main structural elements of the Williston Basin (Obermajer et al., 2000).

The petroleum system concept in the Williston Basin was first applied by Williams (1974) and Dow (1974) who detailed three petroleum systems, and recent studies have expanded the number to nine including the Red River, Winnipegosis, Bakken, Madison, Tyler, Winnipeg, Duperow, Birdbear and Deadwood (Lillis, 2013). In this study, we focused on the Red River petroleum system because it is a self-sourced and highly prolific system with significant remaining exploration potential (Nesheim, 2017).

The Red River strata are Ordovician in age and consist of three brining-upward cycles. The lowest cycle is the interval of interest in this study. The Yeoman Formation rests disconformably or unconformably on the Winnipeg or Deadwood Formations within the cycle. The Yeoman Formation is made up of dolomitized to undolomitized, burrow-mottled to unmottled lime mudstones and wackestones (Kreis and Kent, 2000). The lower two-thirds of the Yeoman Formation lack sufficient porosity and permeability to be considered a good reservoir unit. The uppermost third, which is known as the Upper Yeoman, is strongly mottled, sufficiently porous and permeable to form good quality reservoirs (Kendall, 1976; Kreis and Kent, 2000; Pu et al., 2003). Furthermore, the thin, commonly laminated, organic-rich interbeds (kerogenite or kukersite) within the Upper Yeoman are identified as the main source of petroleum in Red River reservoirs throughout the Williston Basin (Derby and Kilpatrick, 1985).

In the Red River petroleum system, hydrocarbons have mostly accumulated in structural traps with stratigraphic enhancement (Husinec, 2016). Drilling and seismic investigations indicate that most of the Red River pools are associated with small, low-relief, anticlinal structures formed by compactional drapes of the Red River strata over Precambrian paleohighs (Pu et al., 2003).

The lowest cycle of the Red River strata is sealed by the upper evaporitic unit of Lake Alma Member which is informally called Lake Alma anhydrite (Khan et al., 2006). Lake Alma anhydrite is composed of laminated dolomites, overlain by nodular and laminated anhydrite (El Taki and Pratt, 2009).

Criteria identification and selection

The petroleum system concept can be efficiently used in exploration and resource assessment, and provides a basis for proposing undiscovered commercial accumulations (Magoon, 1997). The petroleum system is a unifying concept that describes the dynamic, mobile formation of petroleum pools and encompasses all the essential elements and processes needed for oil and gas accumulations to exist (Magoon, 1999; Perrodon, 1992). The essential elements of a petroleum system include the source rock, reservoir rock, seal rock and overburden rock. Trap formation and the generation-migration-accumulation of petroleum are the processes. These interdependent elements and processes must be correctly placed in space and time so that organic matter contained in the source rock can be converted into oil and gas accumulations (Magoon and Dow, 1991). In this paper, some characteristics of the Red River self-sourced petroleum system elements are employed as criteria for modeling the petroleum potential. These criteria are briefly described below.

Sedimentary rocks rich in organic matter that form oil or gas are source rocks (Hyne, 2001). Effective source rocks are generating or have generated and expelled petroleum. They satisfy three geochemical requirements, including quantity, quality and thermal maturity (Peters and Cassa, 1994).

As shown in Table 1, there are several common ways to assess each of these factors. The thickness of source rock (D_s), total organic carbon (TOC) contents and thermal maturity (T_max) control the amount of petroleum generated and expelled from the source rock (Pitman et al., 2001). T_max is a pyrolysis parameter which measures the degree of maturation of organic matter in rocks. It is defined as the pyrolysis temperature at which the maximum amount of hydrocarbons is released from the decomposition of kerogen (Espitalié, 1986).

Table 1.

(a) Generative potential (quantity) of immature source rock, (b) kerogen type and expelled products (quality) and (c) thermal maturity (Peters and Cassa, 1994).

(a)		Rock-Eval (mg/g rock)
Potential (quantity)	TOC (wt. %)	S₁	S₂	Bitumen (ppm)	Hydrocarbons (ppm)
Poor	<0.5	<0.5	<2.5	<500	<300
Fair	0.5–1	0.5–1	2.5–5	500–1000	300–600
Good	1–2	1–2	5–10	1000–2000	600–1200
Very Good	2–4	2–4	10–20	2000–4000	1200–2400
Excellent	>4	>4	>20	>4000	>2400
(b)
Kerogen (quality)	Hydrogen index (mg HC/g TOC)		S₂/S₃	Atomic H/C	Main product at peak maturity
I	>600		>15	>1.5	Oil
II	300–600		10–15	1.2–1.5	Oil
II/III	200–300		5–10	1.0–1.2	Oil/gas
III	50–200		1–5	0.7–1.0	Gas
IV	<50		<1	<0.7	None
(c)	Maturation			Generation
Maturity	R_o (%)	T_max (°C)	TAI	Bitumen/TOC	Bitumen (mg/g rock)	Production index (S₁/(S₁+S₂))
Immature	0.20–0.60	<435	1.5–2.6	<0.05	<50	<0.10
Mature
Early	0.60–0.65	435–445	2.6–2.7	0.05–0.10	50–100	0.10–0.15
Peak	0.65–0.90	445–450	2.7–2.9	0.15–0.25	150–250	0.25–0.40
Late	0.90–1.35	450–470	2.9–3.3	–	–	>0.40
Postmature	>1.35	>470	>3.3	–	–	–

TOC: total organic carbon.

Among the quality parameters, the hydrogen index (HI) is often used in the analysis of petroleum source rocks (Pedersen et al., 1992). HI is a pyrolysis parameter which measures the quality of organic matter in a source rock (Sonnenberg, 2011). This parameter can be used for making a quick estimate of convertibility of organic matter to mobile hydrocarbons and determining the type of expelled products. For example, type I kerogen (HI of over 600 mg HC/g TOC) has up to 80% convertibility to mobile hydrocarbons, and yields a light, high-quality oil, while type III kerogen (HI of less than 150 mg HC/g TOC) with a convertibility of less than 40% generates mostly gas, plus varying amounts of waxy oil (Hallett, 2002). In this study, TOC, T_max, D_s and HI were selected as criteria for assessing the potential of the source rock.

A reservoir is a body of underground rock having sufficient porosity and permeability to store and transmit oil, water and gas (Speight, 2013). The most common reservoir rocks are sedimentary rocks, because on the one hand, they have generally higher porosity and permeability than igneous and metamorphic rocks, and on the other hand, they form under temperature conditions at which hydrocarbons may be preserved (Max et al., 2013). The thickness of reservoir rock (D_r) and depth to reservoir are two of the hydrocarbon-volume attributes that determine the volume of the hydrocarbon accumulation within the prospect (Crovell and Balay, 1994). Unlike thickness of reservoir rock, there is no clear relationship between reservoir depth and potential of reservoir rock. Therefore, in the absence of porosity and permeability data, only D_r was considered as criterion for the evaluation of the reservoir rock potential.

A trap is a geometrical configuration of rocks in which hydrocarbons can accumulate (Bradley and Powley, 1994). It must be sealed by a relatively impermeable geologic formation to stop upward migration of the hydrocarbons. There are two basic types of traps: structural traps which are the result of strata deformation by tectonic movements and stratigraphic traps where impervious strata seal the reservoir (Gudmestad et al., 2010; Jahn et al., 2008). The trapping potential of structures is indirectly related to roughness (R_r) and curvature (C_r) of reservoir top surface (Arab Amiri et al., 2015a; Dolson, 2016; Ericsson et al., 1998). Since traps are predominantly structural in the Red River petroleum system (Husinec, 2016), R_r and C_r were adopted as the criteria to assess the potential of traps formed by faults and anticlinal folds, respectively.

The cap rock is a low-permeable or impermeable lithologic unit that prevents the escape of hydrocarbons from the underground reservoir. Good cap rocks should be impervious, forming a barrier around and above the reservoir rock. Extension and thickness of cap rock (D_c) are other aspects to be taken into account in seal evaluation (Gudmestad et al., 2010). The low-permeable Lake Alma anhydrite extends across the entire Red River petroleum system and its sealing ability mainly depends on the thickness parameter (Khan et al., 2006). Therefore, D_c was taken into consideration in analyzing the cap rock potential.

Data preparation

The Williston Basin Targeted Geoscience Initiative database (TGI II Williston Basin Project Working Group, 2008) and the Geoscience Data Repository of Earth Sciences Sector of Natural Resources Canada are the sources of all data used in this study. The latest date of the data for Manitoba and Saskatchewan provinces of Canada at the time of compilation is 2008. Datasets required for the purpose of modeling petroleum potential were selected from the databases on the basis of the criteria outlined in the previous section. All datasets were reprojected to the Universal Transverse Mercator projection (zone 13 N and datum NAD83). Spatial operations such as clip, merge, join, etc. were subsequently used to prepare vector datasets. In the spatial domain, FP and OFP can be applied at a pixel or voxel level. Petroleum potential of the Red River petroleum system was modeled in two-dimensional space in this study, so all datasets were converted to raster data model with spatial resolution of 1000 meters which is considered to be appropriate for the scale of the source data (Hengl, 2006).

TOC, T_max and HI values were determined by Rock-Eval pyrolysis on 183 core samples of the source rock across the study area and interpolated by the tension spline method (Figure 4(a) to (c)).

Figure 4.

The spatial data layers used in this study: TOC (a), T_max (b), HI (c), thickness of the source rock (d), thickness of the reservoir rock (e), thickness of the cap rock (f), roughness index (g) and curvature (h) maps.

Isopach data of the source, reservoir and cap rocks, and structure data of the reservoir rock were collected from exploratory and production wells in the northern part of the Williston Basin. The ordinary kriging method was used to rasterize the isopach and structure data. The resulting isopach grids of the Upper Yeoman, the reservoir zone of the lowest cycle of the Red River strata and the Lake Alma anhydrite were used in as D_s, D_r and D_c datasets, respectively (Figure 4(d) to (f)).

Surface roughness is the amount of elevation variability of a particular area. It can be estimated using techniques such as roughness index (RI), variograms of elevation, standard deviation of elevation, chain set measurements and fractal geometry (Stambaugh and Guyette, 2008). RI was applied to structure grid of the reservoir rock to create R_r dataset (Figure 4(g)). It simply quantifies surface heterogeneity by the following equation (Riley et al., 1999)

RI = Y [{(\sum X_{i j} - X_{c})^{2}]}^{\frac{1}{2}}

(16)

where

X_{c}

is the elevation value of central cell and

X_{i j}

is the elevation value of its neighboring cells.

Furthermore, curvature of the structure grid of the reservoir rock was calculated in order to obtain C_r dataset (Figure 4(h)). Curvature is the second derivative of a surface fitted to elevation and measures the concavity or convexity of the surface at a particular pixel (Winstral et al., 2002). Thus, the curvature of structure grids can be utilized in evaluating the trapping potential.

Application and results

In order to apply FP, cell values of the eight raster datasets described in the previous section were extracted into a database table with eight fields as criteria and about 110,000 records as alternatives. In the first step, for each criterion, a generalized preference function $p_{k} (d)$ was specified. A vector containing the triangular fuzzy weights must then be defined. To this end, the exploration experts participating in the present study were asked to assign one of the seven linguistic variables shown in Figure 5 to each criterion. This linguistic variable represents the relative contribution of each criterion to the overall petroleum potential as a triangular fuzzy number (a special case of the TF interval). The variables assigned by different experts were combined by fuzzy simple additive weighting (Chou et al., 2008) to obtain a triangular fuzzy weight for each criterion. The aggregate weights of the criteria are illustrated as triangular fuzzy numbers in Figure 6. Finally, ${\tilde{ϕ}}^{+} (a_{i})$ and ${\tilde{ϕ}}^{-} (a_{i})$ were calculated by means of formulae (6) to (9).

Figure 5.

Representation of the linguistic variables as triangular fuzzy numbers.

Figure 6.

The graphical representation of aggregate weights of the criteria.

Numerous approaches have been proposed in the literature for ranking fuzzy numbers (Detyniecki and Yager, 2000; Facchinetti and Ricci, 2004; Gu and Xuan, 2017; Lee et al., 1994; Wang et al., 2006), most of which transform the fuzzy number into a real number. Among the available methods for defuzzification, the center of area (COA) method yields more reasonable and reliable results. It also makes possible a consistent evaluation of triangular/trapezoidal fuzzy data and of crisp data (Geldermann et al., 2000; Zhang et al., 2009). Therefore, the method was used to defuzzify the resulting fuzzy positive and negative outranking flows. The COA defuzzification method can be expressed mathematically as follows (Geldermann et al., 2000)

x_{defuzzified} = \frac{\int x μ (x) d x}{\int μ (x) d x} = \frac{m_{u}^{2} - m_{ℓ}^{2} + α m_{ℓ} + β m_{u} + (β^{2} - α^{2}) / 3}{α + β + 2 m_{u} - 2 m_{ℓ}}

(17)

where

μ

is the membership function for a TF number.

After defuzzification through the use of the COA method, net outranking flow was calculated as ${ϕ^{net} = ϕ^{+} - ϕ}^{-}$ and represents the overall outranking characteristic of each alternative. The greater the positive outranking flow and the lower the negative outranking flow, the greater the net outranking flow, and accordingly, the higher the overall preference intensity of an alternative. The $ϕ^{net}$ values were then normalized to between 0 and 1 for easy comparison and classification using $ϕ_{norm}^{net} = ϕ_{i}^{net} - ϕ_{\min}^{net} / ϕ_{\max}^{net} - ϕ_{\min}^{net}$ . The $ϕ_{norm}^{net}$ values were converted to the raster data model. The resulting petroleum potential map that was smoothed by a $3 \times 3$ pixel median filter is shown in Figure 7. The higher the pixel value, the more potential the pixel has for petroleum occurrence.

Figure 7.

The potential map of the petroleum system under study obtained from FP with experts-assigned weights.

The accuracy in determining the parameters of knowledge-driven models such as FP by the experts is dependent on many factors including complexity of the research object, expert qualification, the scale of evaluation, the number of criteria to be evaluated, etc. For example, complicated models with a large number of criteria may suffer from expert-elicited values of parameters. For experts who contributed to this study, there was noticeably higher ambiguity in determining weights of the criteria than others. Therefore, the experts-assigned weights were optimized using the CS algorithm to obtain more accurate results. In order to determine the optimal values of parameters, the optimization objective function needs to be first defined. The objective function, in fact, is an indicator of agreement between actual and calculated values of the variable of interest. In this paper, RMSE was selected as the objective function to be minimized in the optimization process. The formula to calculate RMSE is

RMSE = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} (ϕ_{norm, cal, i}^{net} - ϕ_{norm, act, i}^{net})^{2}}

(18)

where

ϕ_{norm, cal}^{net}

is the model-calculated value for pixels (alternatives) that lie inside the area of the discovered oil pools,

ϕ_{norm, act}^{net}

is the actual value of

ϕ_{norm}^{net}

for the same pixels, and

i = 1, 2, \dots, n

, where

n

is the total number of pairs of calculated and actual data. Ideally,

ϕ_{norm, act}^{net}

should be equal to 1, the highest value a pixel can have.

Considering that the weights of the eight criteria were defined as triangular fuzzy numbers, 24 parameters should be optimized. The algorithm parameters, including the number of host nests and the maximum number of iterations, were set to 50 and 3000, respectively, to ensure a fair evaluation. Following Yang and Deb (2009), the mutation probability value $p_{a}$ was also set to 0.25. In addition, in the continuous search space, the following constraints were taken into account when determining the optimal values of the weights: $0 \leq m \leq 1$ , $0 \leq m - α \leq 1$ , $0 \leq m + β \leq 1$ , $m - α < m + β$ , $m - α < m < m + β$ , where $m = m_{ℓ} = m_{u}$ because a triangular fuzzy weight is a special case of the TF interval with $m_{ℓ} = m_{u}$ . Figure 8 depicts the number of iterations versus the values returned by the objective function. The triangular fuzzy numbers estimated by the CS algorithm (Figure 9) were eventually used as weights of the criteria in FP to construct an OFP approach. The potential map of the petroleum system under study obtained from OFP is illustrated in Figure 10.

Figure 8.

The fitness convergence of the CS algorithm.

Figure 9.

Optimized triangular fuzzy weights of the criteria.

Figure 10.

The petroleum potential map of the Red River petroleum system produced by OFP.

In order to make comparison of the results easier, $ϕ_{norm}^{net}$ values resulting from each method were reclassified into five classes ranging from no to high potential (Figure 11). A vector layer of the discovered oil pools of the Red River petroleum system was then overlaid on the reclassified maps (black-outlined polygons in Figure 11) to evaluate the predictive accuracy of FP and OFP.

Figure 11.

Vector layer of the discovered oil pools overlaid on the petroleum potential map obtained from FP (a) and OFP (b).

Discussion

Petroleum exploration is an expensive, difficult and risky, but necessary, operation. A robust and efficient approach is thus required to achieve a high degree of accuracy in modeling petroleum potential, which will ultimately lead to lower costs and risks in exploration. In this regard, one of the important contributions of this study is to use the optimized triangular fuzzy weights instead of the experts-assigned ones. A visual comparison of the two sets of weights is displayed in Figure 12. It can be noticed that the experts have undervalued the relative importance of the geochemical criteria including TOC, T_max and HI. They have assigned higher values for weights of the other five criteria than that computed by the CS algorithm. Another remarkable point is that there is a significant difference between the optimized and experts-assigned weights for T_max, HI, R_r and D_r in contrast to the other criteria.

Figure 12.

Experts-assigned triangular fuzzy weights of the evaluation criteria (non-hatched triangles) compared to optimized weights (hatched triangles).

In this study, the predictive accuracy of the models was evaluated by comparing the high-potential class of the reclassified maps to the discovered oil pools of the Red River petroleum system. As indicated in Table 2, FP and OFP have successfully predicted 69.96% and 83.81% of the discovered pools area, respectively. In further detail, 49.26% and 14.57% of the area of the pools have been located in the interval $ϕ_{norm}^{net} =$ [0.8, 0.9] of FP and OFP, respectively, while these percentages are 20.7% and 69.24% for the interval $ϕ_{norm}^{net} =$ [0.9, 1]. It is clear that OFP has led to more accurate results than FP. The use of the CS algorithm to optimize the triangular fuzzy weights of the criteria has improved predictive accuracy by approximately 14%. This implies a relatively higher success rate, lower cost and risk in petroleum exploration.

Table 2.

Attributes of petroleum potential classes.

Petroleum potential class	$ϕ_{norm}^{net}$	Area (%)		Area of discovered pools located in class (%)
Petroleum potential class	$ϕ_{norm}^{net}$	FP	OFP	FP	OFP
No-potential	0–0.1	14.99	13.96	0	0
No-potential	0.1–0.2	16.39	10.72	0	0
Very low	0.2–0.3	11.36	12.09	0	0
Very low	0.3–0.4	10.46	14.53	0	0
Low	0.4–0.5	14.19	13.95	0.16	0
Low	0.5–0.6	11.32	10.49	9.19	4.6
Moderate	0.6–0.7	11.61	11.18	0.11	4.11
Moderate	0.7–0.8	8.19	11.86	20.58	7.48
High-potential	0.8–0.9	1.28	0.83	49.26	14.57
High-potential	0.9–1	0.21	0.39	20.7	69.24

The next important point to be considered is the screening capability of the model, which has an inverse relationship with area ratio of the high-potential class. The lower the area ratio of the high-potential class, the smaller the area for more detailed investigation, and accordingly, the lower the cost of exploration. Therefore, the screening capability of a model is an important issue in petroleum potential assessment. FP and OFP resulted in the ratio values of 0.015 and 0.012, respectively (Figure 13), both of which are considerably lower than the values obtained in the studies discussed earlier.

Figure 13.

Pie charts showing the ratio of the high-potential class area to the total area under study for FP and OFP.

Conclusions

This study resulted in the following conclusions:

The proposed OFP approach has the ability to model petroleum potential under uncertainty and has identified the location of most of the oil accumulations discovered to date in the Canadian side of the Red River petroleum system.

There is a significant increase in predictive accuracy of the model with the optimized weights compared to the use of the experts-assigned or non-optimized weights, which means that the CS algorithm has successfully estimated the triangular fuzzy weights of the criteria.

Areas with the highest potential and the lowest risk have been located in the central part of the Canadian Williston Basin which can be considered as future exploration targets in the Red River petroleum system.

We plan to continue this research by improving the OFP approach as well as by including additional input datasets into the model. The former will be accomplished by optimizing all parameters of the model and by using other types of preference functions. For the latter, a petroleum system with relatively complete datasets will be selected to be evaluated by the proposed approach in the future study. Datasets pertaining to the permeability and porosity of reservoir rock were inaccessible in the present study. It is expected that using such datasets can provide more accurate results.

Optimal well placement is a challenging problem in oil field development since it depends on reservoir, petrophysical and fluid properties on the one hand, and on economic parameters on the other, all of which are highly uncertain. It requires an efficient and robust ranking algorithm that its input parameters are properly tuned and has the capability to appropriately deal with uncertain data and subjective preferences of decision makers. We suggest that studies be performed to determine optimal well locations in the petroleum industry using OFP.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Mohamad Hamzeh

References

Abedi

Torabi

Norouzi

, et al. (2012a) ELECTRE III: A knowledge-driven method for integration of geophysical data with geological and geochemical data in mineral prospectivity mapping. Journal of Applied Geophysics 87: 9–18.

Abedi

Torabi

Norouzi

, et al. (2012b) PROMETHEE II: A knowledge-driven method for copper exploration. Computers & Geosciences 46: 255–263.

Al-Shemmeri

Al-Kloub

Pearman

(1997) Model choice in multicriteria decision aid. European Journal of Operational Research 97: 550–560.

Albadvi

(2004) Formulating national information technology strategies: A preference ranking model using PROMETHEE method. European Journal of Operational Research 153: 290–296.

Arab Amiri

Karimi

Alimohammadi Sarab

(2015a) Hydrocarbon resources potential mapping using evidential belief functions and frequency ratio approaches, southeastern Saskatchewan, Canada. Canadian Journal of Earth Sciences 52: 182–195.

Arab Amiri

Karimi

Alimohammadi Sarab

(2015b) Hydrocarbon resources potential mapping using the evidential belief functions and GIS, Ahvaz/Khuzestan Province, southwest Iran. Arabian Journal of Geosciences 8: 3929–3941.

Bingham

Zurita-Milla

Escalona

(2012) Geographic information system-based fuzzy-logic analysis for petroleum exploration with a case study of northern South America. AAPG Bulletin 96: 2121–2142.

Blewitt

Coolbaugh

Holt

, et al. (2002) Targeting of potential geothermal resources in the Great Basin from regional relationships between geodetic strain and geological structures. Geothermal Resources Council Transactions 26: 523–526.

Bradley

Powley

(1994) Pressure compartments in sedimentary basins: A review. In: Ortoleva

(ed.) Basin Compartments and Seals. Tulsa: The American Association of Petroleum Geologists, pp.3–26.

10.

Brajevic

Tuba

(2014) Cuckoo search and firefly algorithm applied to multilevel image thresholding. In: Yang

(ed.) Cuckoo Search and Firefly Algorithm: Theory and Applications. Studies in Computational Intelligence, vol. 516. Cham: Springer, pp.115–139.

11.

Brans

Mareschal

(1990) The Promethee methods for MCDM; The Promcalc, Gaia and Bankadviser software. In: Costa

CABe

(ed.) Readings in Multiple Criteria Decision Aid. Berlin, Heidelberg: Springer, pp.216–252.

12.

Brans

Mareschal

(1994) The PROMCALC & GAIA decision support system for multicriteria decision aid. Decision Support Systems 12: 297–310.

13.

Brans

Mareschal

Vincke

(1984) PROMETHEE: A new family of outranking methods in multicriteria analysis. In: Brans

(ed.) Operational Research’84 (IFORS’84). Washington, DC, USA, pp.477–490.

14.

Brans

Vincke

(1985) A preference ranking organization method (the PROMETHEE method for multiple criteria decision-making). Management Science 31: 647–656.

15.

Brans

Vincke

Mareschal

(1986) How to select and how to rank projects: The Promethee method. European Journal of Operational Research 24: 228–238.

16.

Carranza

EJM

Hale

(2002) Spatial association of mineral occurrences and curvilinear geological features. Mathematical Geology 34: 203–221.

17.

Carranza

EJM

Hale

Faassen

(2008a) Selection of coherent deposit-type locations and their application in data-driven mineral prospectivity mapping. Ore Geology Reviews 33: 536–558.

18.

Carranza

EJM

Wibowo

Barritt

, et al. (2008b) Spatial data analysis and integration for regional-scale geothermal potential mapping, West Java, Indonesia. Geothermics 37: 267–299.

19.

Chen

Fang

(1993) A new method for prospect appraisal. AAPG Bulletin 77: 9–18.

20.

Chen

Jin

, et al. (2008) Exploration risk evaluation using object-based modeling, an example from the Tertiary fractured play, western Qaidam Basin of China. Petroleum Science 5: 195–202.

21.

Chen

Osadetz

(2005) Simulating the spatial distribution of undiscovered petroleum accumulations. Journal of Canadian Petroleum Technology 44: 24–30.

22.

Chen

Osadetz

Gao

, et al. (2000) Characterizing the spatial distribution of an undiscovered hydrocarbon resource: The Keg River Reef Play, Western Canada Sedimentary Basin. Bulletin of Canadian Petroleum Geology 48: 150–162.

23.

Chen

Osadetz

(2006a) Geological risk mapping and prospect evaluation using multivariate and Bayesian statistical methods, western Sverdrup Basin of Canada. AAPG Bulletin 90: 859–872.

24.

Chen

Osadetz

(2006b) Undiscovered petroleum accumulation mapping using model-based stochastic simulation. Mathematical Geology 38: 1–16.

25.

Chou

S-Y

Chang

Y-H

Shen

C-Y

(2008) A fuzzy simple additive weighting system under group decision-making for facility location selection with objective/subjective attributes. European Journal of Operational Research 189: 132–145.

26.

Civicioglu

BesdokEmail

(2013) A conceptual comparison of the Cuckoo-search, particle swarm optimization, differential evolution and artificial bee colony algorithms. Artificial Intelligence Review 39: 315–346.

27.

Coolbaugh

Taranik

Raines

, et al. (2002) A geothermal GIS for Nevada: Defining regional controls and favorable exploration terrains for extensional geothermal systems. Geothermal Resources Council Transactions 26: 485–490.

28.

Crovell

Balay

(1994) Geologic model, probabilistic methodology and computer programs for petroleum resource assessment. In: Teleki

Mattick

Kókai

(eds) Basin Analysis in Petroleum Exploration: A Case Study from the Bekes Basin, Hungary. Netherlands: Springer Science & Business Media Dordrecht, pp.295–304.

29.

Derby

Kilpatrick

(1985) Ordovician Red River dolomite reservoirs, Killdeer Field, North Dakota. In: Roehl PO and Choquette PW (eds) Carbonate Petroleum Reservoirs. New York: Springer, pp.59–69.

30.

Detyniecki

Yager

(2000) Ranking fuzzy numbers using a weighted valuations. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 8: 573–591.

31.

Dolson

(2016) Understanding Oil and Gas Shows and Seals in the Search for Hydrocarbons. Switzerland: Springer International Publishing.

32.

Dow

(1974) Application of oil-correlation and source-rock data to exploration in Williston Basin. AAPG Bulletin 58: 1253–1262.

33.

Dubois

Prade

(1980) Fuzzy Sets and Systems. New York: Academic Press.

34.

El Taki

Pratt

(2009) Synsedimentary deformation in laminated dolostones and evaporates of the Herald Formation (Red River): Signature of Late Ordovician tectonic activity in southern Saskatchewan. In: Summary of Investigations 2009. Saskatchewan, Canada: Saskatchewan Geological Survey, pp.1–10.

35.

Ericsson

McKean

Hooper

(1998) Facies and curvature controlled 3D fracture models in a Cretaceous carbonate reservoir, Arabian Gulf. Geological Society, London, Special Publications 147: 299–312.

36.

Espitalié

(1986) Use of T_max as a maturation index for different types of organic matter. Comparison with vitrinite reflectance. In: Burrus

(ed.) Thermal modeling in sedimentary basins: 1st IFP exploration research conference, 3–7 June 1985, pp.475–496. Carcans, France: Editions TECHNIP.

37.

Facchinetti

Ricci

(2004) A characterization of a general class of ranking functions on triangular fuzzy numbers. Fuzzy Sets and Systems 146: 297–312.

38.

Fister

Yang

Fister

, et al. (2014) Cuckoo search: A brief literature review. In: Yang

(ed.) Cuckoo Search and Firefly Algorithm: Theory and Applications. 1 ed. Cham, Switzerland: Springer, pp.49–62.

39.

Flannery

(2006) Integrated analysis of the Bakken petroleum system, US Williston Basin. In: Gilboy

Whittaker

(eds) Saskatchewan and Northern Plains Oil & Gas Symposium. Canada: Saskatchewan Geological Society Special Publication 19, pp.138–145.

40.

Gandomi

Yang

Alavi

(2013) Cuckoo search algorithm: A metaheuristic approach to solve structural optimization problems. Engineering with Computers 29: 17–35.

41.

Gao

Chen

Osadetz

, et al. (2000) A pool-based model of the spatial distribution of undiscovered petroleum resources. Mathematical Geology 32: 725–749.

42.

Geldermann

Spengler

Rentz

(2000) Fuzzy outranking for environmental assessment. Case study: Iron and steel making industry. Fuzzy Sets and Systems 115: 45–65.

43.

Gerhard

Anderson

Fischer

, (1990) Petroleum geology of the Williston Basin. In: Leighton

Kolata

Oltz

, et al. (eds) Interior Cratonic Basins. Tulsa: The American Association of Petroleum Geologists, pp.507–559.

44.

Goumas

Lygerou

(2000) An extension of the PROMETHEE method for decision making in fuzzy environment: Ranking of alternative energy exploitation projects. European Journal of Operational Research 123: 606–613.

45.

Grant

Milton

Thompson

(1996) Play fairway analysis and risk mapping: An example using the Middle Jurassic Brent Group in the northern North Sea. NPF Specical Publications 6: 167–181.

46.

Xuan

(2017) A new approach for ranking fuzzy numbers based on possibility theory. Journal of Computational and Applied Mathematics 309: 674–682.

47.

Gudmestad

Zolotukhin

Jarlsby

(2010) Petroleum Resources with Emphasis on Offshore Fields. UK: WIT Press.

48.

Hallett

(2002) Petroleum Geology of Libya. Amsterdam, The Netherlands: Elsevier.

49.

Hanoun

Nahavandi

Creighton

, et al. (2012) Solving a multiobjective job shop scheduling problem using Pareto Archived Cuckoo Search. In: 17th international conference on emerging technologies & factory automation (ETFA 2012), pp. 1–8. Kraków, Poland: IEEE.

50.

Heck

LeFever

Fischer

, et al. (2002) Overview of the Petroleum Geology of the North Dakota Williston Basin. Bismarck, North Dakota, USA: North Dakota Geological Survey.

51.

Hengl

(2006) Finding the right pixel size. Computers & Geosciences 32: 1283–1298.

52.

Hitchon

(1996) Rapid evaluation of the hydrochemistry of a sedimentary basin using only standard formation water analyses: Example from the Canadian portion of the Williston basin. Applied Geochemistry 11: 789–795.

53.

Husinec

(2016) Sequence stratigraphy of the Red River Formation, Williston Basin, USA: Stratigraphic signature of the Ordovician Katian greenhouse to icehouse transition. Marine and Petroleum Geology 77: 487–506.

54.

Hyne

(2001) Nontechnical Guide to Petroleum Geology, Exploration, Drilling and Production. USA: Pennwell Corp.

55.

Jahn

Cook

Graham

(2008) Hydrocarbon Exploration and Production. Amsterdam, The Netherlands: Elsevier.

56.

Kendall

(1976) The Ordovician Carbonate Succession (Bighorn Group) of Southeastern Saskatchewan. Saskatchewan, Canada: Saskatchewan Department of Mineral Resources, p.185.

57.

Khan

Rostron

Margitai

, et al. (2006) Hydrodynamics and petroleum migration in the Upper Ordovician Red River Formation of the Williston Basin. Journal of Geochemical Exploration 89: 179–182.

58.

Kraipeerapun

Fung

Brown

(2005) Assessment of uncertainty in mineral prospectivity prediction using interval neutrosophic set. In: International conference on computational and information science, pp.1074–1079. Xi’an, China: Springer, Berlin, Heidelberg.

59.

Kreis

Kent

(2000) Basement controls on red river sedimentation and hydrocarbon production in southeastern Saskatchewan. In: Christopher JE, Gilboy CF, Haidl FM, Harper CT (eds) Summary of Investigations 2000. Saskatchewan, Canada: Saskatchewan Geological Survey, pp.21–42.

60.

Kumar

Chakarverty

(2011) Design optimization using genetic algorithm and cuckoo search. In: IEEE international conference on electro/information technology (EIT), pp.1–5. Mankato, MN, USA: IEEE.

61.

Lee

K-M

Cho

C-H

Lee-Kwang

(1994) Ranking fuzzy values with satisfaction function. Fuzzy Sets and Systems 64: 295–309.

62.

Lillis

(2013) Review of oil families and their petroleum systems of the Williston Basin. Mountain Geologist 50: 5–31.

63.

Linkov

Varghese

Jamil

, et al. (2004) Multi-criteria decision analysis: A framework for structuring remedial decisions at contaminated sites. In: Linkov

Ramadan

(eds) Comparative Risk Assessment and Environmental Decision Making. Dordrecht: Springer, pp.15–54.

64.

Ting

Man

, et al. (2013) Parameter estimation of photovoltaic models via cuckoo search. Journal of Applied Mathematics, 1–8.

65.

Magalhães

Filho

(2012) Targeting of gold deposits in Amazonian exploration frontiers using knowledge- and data-driven spatial modeling of geophysical, geochemical, and geological data. Surveys in Geophysics 33: 211–241.

66.

Magoon

(1997) The petroleum system – An exploratory tool to find oil and gas and to assist in risk management. In: Howes

JVC

Noble

(eds) Proceedings of the conference on petroleum systems of SE Asia and Australasia. Jakarta, Indonesia: Indonesian Petroleum Association, pp.25–36.

67.

Magoon

(1999) Petroleum systems. In: Beaumont

Foster

(eds) Exploring for Oil and Gas Traps: American Association of Petroleum Geologists Treatise of Petroleum Geology, Handbook of Petroleum Geology. Tulsa: AAPG, pp.3–34.

68.

Magoon

Dow

(1991) The petroleum system – From source to trap. Tulsa, Oklahoma, USA: The American Association of Petroleum Geologists.

69.

Mantegna

(1994) Fast, accurate algorithm for numerical simulation of Lévy stable stochastic processes. Physical Review E, Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics 49: 4677–4683.

70.

Martinelli

(2012) Petroleum exploration with Bayesian Networks: from prospect risk assessment to optimal exploration. Mathematical Sciences. Oslo: Norwegian University of Science and Technology.

71.

Max

Clifford

Johnson

(2013) Hydrocarbon system analysis for methane hydrate exploration on Mars. In: Ambrose

Reilly

II and Peters

(eds) Energy Resources for Human Settlement in the Solar System and Earth’s Future in Space: AAPG Memoir 101. Tulsa: The American Association of Petroleum Geologists, pp.99–114.

72.

Nesheim

(2017) Stratigraphic and geochemical investigation of Kukersites (petroleum source beds) within the Ordovician Red River Formation, Williston Basin. AAPG Bulletin 101: 1445–1471.

73.

Obermajer

Osadetz

Fowler

, et al. (2000) Light hydrocarbon (gasoline range) parameter refinement of biomarker-based oil ± oil correlation studies: an example from Williston Basin. Organic Geochemistry 31: 959–976.

74.

Pedersen

Shimmield

Price

(1992) Lack of enhanced preservation of organic matter in sediments under the oxygen minimum on the Oman Margin. Geochimica et Cosmochimica Acta 56: 545–551.

75.

Perrodon

(1992) Petroleum systems: Models and applications. Journal of Petroleum Geology 15: 319–325.

76.

Peters

Cassa

(1994) Applied source rock geochemistry. In: Magoon

Dow

(eds) The Petroleum System – From Source to Trap. Tulsa, Oklahoma, USA: The American Association of Petroleum Geologists, pp.93–120.

77.

Pitman

Price

LeFever

(2001) Diagenesis and Fracture Development in the Bakken Formation, Williston Basin: Implications for Reservoir Quality in the Middle Member. USA: US Geological Survey.

78.

Pollastro

Roberts

LNR

Cook

(2013) Geologic assessment of technically recoverable oil in the Devonian and Mississippian Bakken formation. In: Assessment of Undicsovered Oil and Gas Resources of the Williston Basin Province of North Dakota, Montana and South Dakota. Reston, Virginia, USA: U.S. Geological Survey. Available at: https://pubs.usgs.gov/dds/dds-069/dds-069-w

79.

Kent

Urban

(2003) Characterization of Ordovician Midale field: Implication for Red River play in northern Williston Basin, southeastern Saskatchewan. AAPG Bulletin 87: 1699–1715.

80.

Riley

DeGloria

Elliot

(1999) A terrain ruggedness index that quantifies topographic heterogeneity. Intermountain Journal of Sciences 5: 23–27.

81.

Roisenberg

Schoeninger

da Silva

(2009) A hybrid fuzzy-probabilistic system for risk analysis in petroleum exploration prospects. Expert Systems with Applications 36: 6282–6294.

82.

Rommelfanger

(1988) Entscheiden Bei Unschärfe – Fuzzy Decision Support-Systeme. Berlin: Springer-Verlag.

83.

Rommelfanger

(1996) Fuzzy linear programming and applications. European Journal of Operational Research 92: 512–527.

84.

Ruffo

Bazzana

Consonni

, et al. (2006) Hydrocarbon exploration risk evaluation through uncertainty and sensitivity analyses techniques. Reliability Engineering & System Safety 91: 1155–1162.

85.

Ruzyla

Friedman

(1985) Factors controlling porosity in dolomite reservoirs of the Ordovician Red River formation, Cabin Creek Field, Montana. In: Roehl PO and Choquette PW (eds) Carbonate Petroleum Reservoirs. New York: Springer, pp.39–58.

86.

Sonnenberg

(2011) TOC and pyrolysis data for the Bakken Shales, Williston Basin, North Dakota and Montana. In: Robinson

LeFever

Gaswirth

(eds) The Bakken-Three Forks Petroleum System in the Williston Basin. Denver, Colorado, USA: The Rocky Mountain Association of Geologists, pp.308–331.

87.

Speight

(2013) Heavy Oil Production Processes. Oxford, UK: Gulf Professional Publishing.

88.

Stambaugh

Guyette

(2008) Predicting spatio-temporal variability in fire return intervals using a topographic roughness index. Forest Ecology and Management 254: 463–473.

89.

TGI II Williston Basin Project Working Group (2008) TGI II Williston Basin database. Canada: Manitoba Geological Survey.

90.

Tounsi

(2005) An approximate reasoning based technique for oil assessment. Expert Systems with Applications 29: 485–491.

91.

Tuba

Subotic

Stanarevic

(2011) Modified cuckoo search algorithm for unconstrained optimization problems. In: Leandre

Demiralp

Tuba

, et al. (eds) 5th European conference on European computing conference (ECC’11), pp.263–268. Paris, France: World Scientific and Engineering Academy and Society (WSEAS).

92.

Ufuk Bilsel

Büyüközkan

Rua

(2006) A fuzzy preference‐ranking model for a quality evaluation of hospital web sites. International Journal of Intelligent Systems 21: 1181–1197.

93.

Wang

Y-M

Yang

J-B

D-L

, et al. (2006) On the centroids of fuzzy numbers. Fuzzy Sets and Systems 157: 919–926.

94.

Williams

(1974) Characterization of oil types in Williston Basin. AAPG Bulletin 58: 1243–1252.

95.

Winstral

Elder

Davis

(2002) Spatial snow modeling of wind-redistributed snow using terrain-based parameters. Journal of Hydrometeorology 3: 524–538.

96.

Xie

Guo

, et al. (2011) Prediction of petroleum exploration risk and subterranean spatial distribution of hydrocarbon accumulations. Petroleum Science 8: 17–23.

97.

Yang

X and

Deb

(2009) Cuckoo search via Lévy flights. In: Abraham

Carvalho

Herrera

, et al. (eds) Proceedings of the world congress on nature and biologically inspired computing (NaBIC 2009), Coimbatore, India, pp.210–214. USA: IEEE Publications.

98.

Yang

Deb

(2010) Engineering optimization by cuckoo search. International Journal of Mathematical Modelling and Numerical Optimisation 1: 330–343.

99.

Yang

Deb

(2014) Cuckoo search: Recent advances and applications. Neural Computing and Applications 24: 169–174.

100.

Yang

Deb

Karamanoglu

, et al. (2012) Cuckoo search for business optimization applications. In: 2012 national conference on computing and communication systems (NCCCS), Durgapur, India, pp. 1–5. USA: IEEE.

101.

Zadeh

(1965) Fuzzy sets. Information and Control 8: 338–353.

102.

Zargani

Vaughan

Missallati

(2003) Spatial integration of geological datasets for predictive hydrocarbon studies in Murzuq Basin, SW Libya. In: Geoscience and remote sensing symposium. Toulouse, France, pp.991–993. USA: IEEE.

103.

Zhang

Kluck

Achari

(2009) A comparative approach for ranking contaminated sites based on the risk assessment paradigm using fuzzy PROMETHEE. Environmental Management 44: 952–967.

104.

Zhao

Zhu

(2019) 2-Dimension linguistic PROMETHEE methods for multiple attribute decision making. Expert Systems with Applications 127: 97–108.

105.

Ziyong

Hangyu

Xiaodan

(2017) Fuzzy fusion of geological and geophysical data for mapping hydrocarbon potential based on GIS. Petroleum Geoscience 24(1): 131–141.