Spatial prediction of oil and gas distribution using Tree Augmented Bayesian network

Highlights

• A Bayesian network(BN) is used to predict the spatial distribution of oil and gas.

• BN is convenient to calculate the posterior probability of oil and gas.

• The probability map can be obtained from the calculation results of BN model.

• The map intuitively reflects the change of oil and gas spatial distribution.

Abstract

Accurate prediction of the spatial distribution of hydrocarbon accumulations underlies risk reduction and improvement of exploration work. Determining the location of oil and gas reserves is a comprehensive and complex problem riddled with uncertainty. This paper applies Bayesian network, which is an effective tool for uncertainty knowledge expression and reasoning, to predict the spatial distribution of oil and gas resources, which is the first attempt of its kind. Taking the first member of Dongying Formation (Ed₁) of Nanpu Depression in Bohai Bay Basin for example, 222 exploratory wells, basin simulation, seismic interpretation, and other data were synergized to train the Tree Augmented Bayesian network (TAN) structure, then the hydrocarbon-bearing posterior probability of the target layer is calculated based on the trained TAN model, and ultimately the probability map of hydrocarbon spatial distribution prediction in Ed₁ was obtained. Prediction results indicate that oilfields are found mostly in areas with high resource potential and high hydrocarbon-bearing probability. Notably, the northern part of No.2 structural belt has a high petroleum bearing probability and no oilfields have ever been found there, which can be considered as the next exploration direction. The application results show that Bayesian network serves to capture essential spatial characteristics of hydrocarbon accumulations and accurately predict the spatial distribution of oil and gas resources, which can help to visualize the geological risk, optimize the drilling strategy, and thus contribute to exploration success.

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.