Intelligent Production Modeling Using Full Field Pattern Recognition

Abstract

Production-data analysis has been applied extensively to predicting future production performance and field recovery. These applications operate mostly on a single-well basis. This paper presents a new approach to production-data analysis using artificial-intelligence (AI) techniques in which production history is used to build a fieldwide performance-prediction model. In this work, AI and data-driven modeling are used to predict future production of both synthetic-(for validation purposes) and real-field cases.In the approach presented in this article, production history is paired with field geological information to build data sets containing the spatio-temporal dependencies among different wells. These dependencies are addressed by compiling information from closest offset wells (COWs) that includes their geological and reservoir characteristics (spatial data) as well as their production history (temporal data).Once the data set is assembled, a series of neural networks are trained using a back-propagation algorithm. These networks are then fused together to form the intelligent time-successive production-modeling (ITSPM) system. This technique uses only the widely available measured data such as well logs and production history of existing wells to predict future performance of the existing wells and production performance of the new (infill) wells. To demonstrate the applicability of this method, a synthetic oil reservoir is modeled using a commercial simulator. Production and well-log data are extracted into an all-inclusive data set. Several neural networks are trained and validated to predict different stages of the production. The ITSPM method is used to estimate the production profile for nine new wells in the reservoir. Furthermore, ITSPM is also applied to two giant oil fields in the Middle East. The first one has more than 200 wells and 40 years of production history. ITSPM's production predictions of the four newest wells in this reservoir are compared with their actual production. The second real field has hundreds of wells producing from multiple layers. The field has undergone waterflooding for almost its whole life. This case also shows the capabilities of this technique in more-complex scenarios and especially multiphase systems. Production-Data-Analysis BackgroundMature fields that have been producing for a long time do not provide much reservoir-characterization data. This is because of their age and the lack of historical data from their old wells. In the time that oil price might not be in its highest range, infill-drilling decisions should be made with a better degree of understanding from such fields in order to have an economically better decision.Using production data as a tool in decision making has been an interest of petroleum engineers for many years. Different methods

Geomechanical Properties of Unconventional Shale Reservoirs

“…Once the network is trained and calibrated, then the final model is applied to the verification set. If the results are satisfactory then the neural network is accepted as part of the entire prediction system [15,16]. Figures 4 to 8 show the actual well logs and generated logs for 5 blind wells shown as black circles in Figure 3.…”

A systematic review of data science and machine learning applications to the oil and gas industry

“…Using ANN to forecast the production of several wells using limited production history can potentially help identify the expected productivity of new wells and therefore optimize the field development. Khazaeni and Mohaghegh (2011) developed production data analysis method with AI techniques using production history data to build a field-wide performance prediction model. In their work, production history is paired with field geological information to build datasets containing the spatiotemporal dependencies among different wells.…”

A Connectionist Model for Dynamic Economic Risk Analysis of Hydrocarbons Production Systems

“…The ANN model has an intrinsic/remarkable ability to mimic complex system behaviors and replicate complicated process behaviors in various engineering and science systems irrespective of the process dimensionality and nonlinearity natures (Kazatchenko, Markov, Mousatov, & Pervago, 2006;Moussa, Elkatatny, Mahmoud, & Abdulraheem, 2018;Nikravesh & Aminzadeh, 2003;Rajabi & Tingay, 2013;Rajabzadeh, Ruzich, Zendehboudi, & Rahbari, 2012;Rezaee, Kadkhodaie Ilkhchi, & Barabadi, 2007;Ridha & Gharbi, 2003;Tariq, Elkatatny, Mahmoud, & Abdulraheem, 2016). The black-box model is characterized by its intrinsic potency of reasoning attributes such as perception, discovery, grouping, and generalization (Aleardi, 2015;Cranganu & Bautu, 2002;Gharbi, 2003;Kalantari Dahaghi & Mohaghegh, 2009;Khazaeni & Mohaghegh, 2011;Maleki, Moradzadeh, Riabi, Gholami, & Sadeghzadeh, 2014;Mohaghegh, 2000;Mojtaba, Nasser, Davoud, & Mostafa, 2010;Onalo, Adedigba, Khan, James, & Butt, 2018;Zendehboudi et al, 2018). Zendehboudi et al (2018) described ANN as a strong mathematical black-box model/tool characterized by functions identical to the human neural systems.…”

“…Petroleum reservoirs are complex process systems defined by intrinsically uncertain data and a distinct pressure gradient. The upstream sector's assets include huge uncertainties/high risks (Asheim, 1988; Bittencourt & Horne, 1997; Khazaeni & Mohaghegh, 2011; Wang, Chen, & Chen, 2019; Zhao, Luo, & Xia, 2012). Hence, the investments in these complex underground systems often suffer significant economic risks due to the process complex dynamics, environmental factors, process data uncertainties, and human errors.…”

“…Nicotra, Godi, Cominelli, and Christie (2005) implemented a similar method with an in‐built stochastic feature. Some other researchers employed artificial neural network (ANN) models to forecast the amount of production (Augusto et al., 2014a; Khazaeni & Mohaghegh, 2011; Lechner, Ag, & Zangl, 2005; Shahkarami, Mohaghegh, Gholami, & Haghighat, 2014; Wang et al., 2019; Zhao et al., 2012). Maschio, Schiozer, and Carlo (2014) used Markov chain Monte Carlo (MCMC) for oil production prediction.…”