Using an Artificial-Neural-Network Method To Predict Carbonate Well Log Facies Successfully

Abstract

Summary The Maastrichtian (Upper Cretaceous) reservoir is one of five prolific oil reservoirs in the giant Wafra oil field. The Maastrichtian oil production is largely from subtidal dolomites at an average depth of 2,500 ft. Carbonate deposition occurred on a very gently dipping, shallow, arid, and restricted ramp setting that transitioned between normal marine conditions to restricted lagoonal environments. The average porosity of the reservoir interval is approximately 15%, although productive zones have porosity values up to 30–40%. The average permeability of the reservoir interval is approximately 30 md. Individual core plugs have measured permeability up to 1,200 md. Efforts to predict sedimentary facies from well logs in carbonate reservoirs is difficult because of the complex carbonate sedimentary facies structures, strong diagenetic overprint, and challenging log analysis in part owing to the presence of vugs and fractures. In the study, a workflow including (1) core description preprocessing, (2) log- and core-data cleanup, and (3) probabilistic-neural-network (PNN) facies analysis was used to predict facies from log data accurately. After evaluation of a variety of statistical approaches, a PNN-based approach was used to predict facies from well-log data. The PNN was selected as a tool because it has the capability to delineate complex nonlinear relationships between facies and log data. The PNN method was shown to outperform multivariate statistical algorithms and, in this study, gave good prediction accuracy (above 70%). The prediction uncertainty was quantified by two probabilistic logs—discriminant ability and overall confidence. These probabilistic logs can be used to evaluate the prediction uncertainty during interpretation. Lithofacies were predicted for 15 key wells in the Wafra Maastrichtian reservoir and were effectively used to extend the understanding of the Maastrichtian stratigraphy, depositional setting, and facies distribution.

Practical implementation of knowledge-based approaches for steam-assisted gravity drainage production analysis

“…Applications of neural network can also be found in petroleum engineering. ANN has achieved significant popularity in areas such as production prediction (Al-Fattah & Startzman, 2003), reservoir characterization or properties prediction (An & Moon, 1993;Gharbi & Elsharkawy, 1999;Tang, Meddaugh, & Toomey, 2011), history matching (Ramgulam, 2006), classification (Stundner & Al-Thuwaini, 2001), proxy for prediction of recovery performance (Lechner & Zangl, 2005), production operation optimization and well design (Yeten, Durlofsky, & Aziz, 2002). In recent years, the neural network has also been utilized to evaluate enhanced oil recovery projects (Parada & Ertekin, 2012;Zerafat, Ayatollahi, Mehranbod, & Barzegari, 2011) and assess CO 2 sequestration process (Mohammadpoor, Firouz, Reza, & Torabi, 2012).…”

Cognitive Data-Driven Proxy Modeling for Performance Forecasting of Waterflooding Process

“…In petroleum engineering an extensive variety of neural network applications can be found [16][17][18][19], particularly in the areas of: reservoir characterization or property prediction [20][21][22][23], classification [19], proxy for recovery performance prediction [24,25], history matching [26], and design or optimization of production operations and well trajectory [27][28][29][30][31][32][33]. In particular, neural networks have been utilized in recent years as a proxy model to predict heavy oil recoveries [34][35][36][37][38][39], to perform EOR (enhanced oil recovery) screening [40][41][42]to characterize reservoir properties in unconventional plays [43], and to evaluate performance of a CO 2 sequestration process [44].…”

An Integrated Application of Cluster Analysis and Artificial Neural Networks for SAGD Recovery Performance Prediction in Heterogeneous Reservoirs

“…A wide variety of neural network applications can be found in petroleum engineering (Mohaghegh 2002;Bravo et al 2012;Saputelli et al 2002;Stundner 2001), particularly in the areas of: classification (Stundner 2001), reservoir characterization or property prediction (Tang et al 2011;Raeesi et al 2012), proxy for recovery performance prediction (Awoleke & Lane 2011;Lechner & Zangl 2005), history matching (Ramagulam et al 2007), and design or optimization of production operations and well trajectory (Stoisits et al 1999;Luis et al 2007;Artun et al 2012;Yeten & Durlofsky 2003;Oberwinkler et al 2004;Malallah & Sami Nashawi 2005;Zangl et al 2006). In particular, neural networks have been utilized in recent years as a proxy model to predict heavy oil recoveries (Queipo et al 2002;Popa et al 2011;Popa & Patel 2012;Ahmadloo et al 2010;Aminian et al 2003); to perform EOR (enhanced oil recovery) screening (Zerafat et al 2011;Karambeigi et al 2011;Parada & Ertekin 2012); to characterize reservoir properties in unconventional plays (Holdaway 2012); and to evaluate performance of CO 2 sequestration process (Mohammadpoor et al 2012).…”

“…Numerous works also highlight techniques for pre-processing the input and target data to be used, which includes normalization and outlier detection (Tang et al 2011). Furthermore, in applications where responses from detailed flow simulations are used to train a network that would serve as a proxy for reservoir performance prediction, experimental design is often performed to reduce redundancy in training data and to minimize computational time (Queipo et al 2002).…”