Spatial prediction of oil and gas distribution using Tree Augmented Bayesian network
Introduction
It is of great significance to study the spatial distribution law and predict the geographic locations of oil and gas resources with the aim to minimize exploration risk, optimize exploration targets and improve exploration efficiency. For decades, with the development of the industry, numerous efforts have been made to satisfy the requirements of resource management and improve exploration efficiency. Spatial prediction of hydrocarbons distribution, as a supplement and extension of the conventional assessment methods, has increasingly captured the attention of oil companies and researchers (Xie et al., 2011). Many experts and scholars have proposed methods to describe the characteristics of the spatial distribution of hydrocarbons. These methods are mainly divided into three categories: genetic method (White, 1993, 1988; Grant et al., 1996), stochastic simulation method (Kaufman and Lee, 1992; Chen et al., 2000, 2004a; 2004b; Gao et al., 2000; Chen and Osadetz, 2006b), and statistical method (Harff et al., 1992; Pan, 1997; Chen and Osadetz, 2006a; Xie et al., 2011; Zhu et al., 2018; Handhal et al., 2019). In terms of genetic method, experts assign different probability values to different geological factors that control hydrocarbon formation, and superimpose them on a map to select favorable hydrocarbon accumulation areas, mostly the base of their own professional judgment. This method predicts the location of oil and gas reserves solely based on favorable geological conditions and the probability value assigned by different people may vary considerably. Apparently a consistent and repeatable geological evaluation method is more desirable for reliable exploration strategy analysis. As to stochastic simulation method, it simulates the prospective locations of undiscovered petroleum accumulation considering both geological reasoning and the spatial correlation between petroleum accumulations in the area. However, the accuracy of this method is highly dependent on the results of exploration and drilling, and information integration is not fully utilized. As to statistical method, it mainly involves a data integration using Mahalanobis distance, a probabilistic classification employing Bayesian statistics, and the estimation of probability of hydrocarbon occurrence at untested locations utilizing the established classification model, and ultimately it forms a hydrocarbon-bearing probability map that reflects the synthesis of a variety of factors spanning geological data, exploration and drilling results, geophysical data and other information. At present, although the third kind of method is widely applied, there is still great room for improvement in improving the accuracy of model prediction (Chen and Osadetz, 2006a; Hu et al., 2009; Xie et al., 2011; Zhu et al., 2018).
In this paper, a prediction method of oil and gas spatial distribution based on Tree Augmented Bayesian network (TAN) is proposed. Compared with the previous methods, it has two advantages: (1) The relationship between geological variables can be visible and interpretable through the network topology structure; (2) Bayesian Network has a solid foundation in mathematical theory. Its topology structure is mainly determined through data learning. Its data learning network structure serves to minimize subjective intervention to the greatest extent and objectively and effectively solve the uncertainty in spatial distribution of oil and gas resources (Kahneman et al., 2011; Milkov, 2015). Taking the Nanpu Depression of the Bohai Bay Basin in China as an example, its TAN topological structure is first obtained through seven kinds of geoscientific information, and then the hydrocarbon-bearing probability of the target layer is predicted based on the topological structure. Ultimately, a spatial probability map of oil and gas distribution is formed. The application results show that the proposed TAN method serves to visualize the relationship between various geological factors, predict the spatial distribution of hydrocarbons, minimize exploration risk, optimize exploration targets, and provide theoretical basis for drilling decision-making.
This paper is organized as follows. Section 2 presents the methodology of the TAN method. Section 3 presents the application of the method, including geological setting, data processing, and model development. Section 4 describes the results and discussion. Section 5 gives the conclusions.
Section snippets
Methodology
The purpose of studying the spatial distribution of oil and gas resources is to determine whether a potential drilling target is productive or not based on our understanding of the basin and other available data. In other words, it aims to explore the possibility of unearthing oil and gas resources in an underground location prior to drilling. In this study, this problem is based on a two-class classification with uncertainty in a multidimensional space. Assuming that n exploration wells have
Geological setting
Nanpu Depression is located in the northeast of Huanghua Depression, Bohai Bay Basin, China (Fig. 2(a)–2(b)). The northwest boundary of Nanpu Depression is Xinanzhuang fault, separated from Xinanzhuang uplift and Laowangzhuang uplift. The northeast boundary is Baigezhuang fault, which is adjacent to Baigezhuang uplift and Matouying uplift, and the southern boundary overlaps with Shaleitian uplift (Wei and Sun, 2017). The depression can be divided into 8 structural units from north to south,
Results and discussion
According to expert opinions, the buried depth of the target layer generally affects the distribution of oil and gas, so we take the geological variable ST as the root node. Fig. 6 presents the topological structure of a TAN model based on 222 exploration wells and seven related geological data. By virtue of this model, prediction was performed on 222 exploration wells. The prediction results were compared with the actual exploration results. The number of correctly predicted wells are 184,
Conclusions
In this paper, TAN Bayesian network, a new technology for predicting the spatial distribution of hydrocarbon occurrence, is proposed. It was applied to predict the distribution of oil and gas in Nanpu Depression of Bohai Bay Basin. In addition, the probability map of oil and gas was obtained. Results indicate that oil wells and oilfields are generally located in the high probability area of oil and gas (>50%), while dry wells are located in the low probability area (<50%). The results
Computer code availability
The Python source codes of TAN and case study are available on GitHub at https://github.com/RenHongJia/codes-of-TAN. More information about the codes can be found in the file “README.md.” For any question please contact at email address of corresponding author of the current manuscript.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors would like to thank Research Institute of Petroleum Exploration Development, PetroChina for providing the data and assistance during the study. The authors also acknowledge the support from "Natural Science Foundation of Jilin Province (No. 20190201273JC)".
References (38)
- et al.
An object-based model for predicting the locations of undiscovered oil and gas resources, western sverdrup basin, Canada
Mar. Petrol. Geol.
(2004) - et al.
Supersd: an object-based stochastic simulation program for modeling the locations of undiscovered petroleum accumulations
Comput. Geosci.
(2004) - et al.
Control of strike-slip faulting on reservoir formation of oil and gas in nanpu sag
Petrol. Explor. Dev.
(2008) - et al.
Play fairway analysis and risk mapping: an example using the middle jurassic brent group in the northern north sea
(1996) - et al.
Gis-based machine learning models for mapping tar mat zones in upper part (dj unit) of zubair formation in north rumaila supergiant oil field, southern Iraq
J. Petrol. Sci. Eng.
(2019) - et al.
Improving tree augmented naive bayes for class probability estimation
Knowl. Base Syst.
(2012) Risk tables for less biased and more consistent estimation of probability of geological success (pos) for segments with conventional oil and gas prospective resources
Earth Sci. Rev.
(2015)Fusion, propagation, and structuring in belief networks
Artif. Intell.
(1986)- et al.
Evaluation and oil-source contributions for dongying formation hydrocarbon source rocks of nanpu sag
Pet. Geol. Oilfield Dev. Daqing
(2017) - et al.
Natural gas resources of the sedimentary basins in China
Mar. Petrol. Geol.
(2008)
Evaluation of geological risk and hydrocarbon favorability using logistic regression model with case study
Mar. Petrol. Geol.
Dynamics of fluids in porous media
Soil Sci.
Bayesian network structure learning using quantum annealing
Eur. Phys. J. Spec. Top.
Exploration risk evaluation using object-based modeling, an example from the tertiary fractured play, western qaidam basin of China
Petrol. Sci.
Characterizing the spatial distribution of an undiscovered hydrocarbon resource: the keg river reef play, western Canada sedimentary basin
Bull. Can. Petrol. Geol.
Geological risk mapping and prospect evaluation using multivariate and bayesian statistical methods, western sverdrup basin of Canada
AAPG Bull.
Undiscovered petroleum accumulation mapping using model-based stochastic simulation
Math. Geol.
Elements of Information Theory
Modeling and Reasoning with Bayesian Networks
Cited by (0)
- 1
X.-C. Wang developed the concept for this study. H.-J. Ren designed the study. X.-X. Guo wrote the algorithm. Q.-L. Guo provided the experimental data. R. Zhang guided the way of writing.