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

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.