This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 201693, “Subsurface Analytics Case Study: Reservoir Simulation and Modeling of a Highly Complex Offshore Field in Malaysia Using Artificial Intelligence and Machine Learning,” by Rahim Masoudi, SPE, Petronas; Shahab D. Mohaghegh, SPE, West Virginia University; and Daniel Yingling, Intelligent Solutions, et al., prepared for the 2020 SPE Annual Technical Conference and Exhibition, originally scheduled to be held in Denver, 5–7 October. The paper has not been peer reviewed.
Using commercial numerical reservoir simulators to build a full-field reservoir model and simultaneously history matching multiple dynamic variables for a highly complex offshore mature field in Malaysia had proved challenging. In the complete paper, the authors demonstrate how artificial intelligence (AI) and machine learning can be used to build a purely data-driven reservoir simulation model that successfully history matches all dynamic variables for wells in this field and subsequently can be used for production forecasting. This synopsis concentrates on the process used, while the complete paper provides results of the fully automated history matching.
Subsurface Analytics
In the presented technique, which the authors call subsurface analytics, data-driven pattern-recognition technologies are used to embed the physics of the fluid flow through porous media and to create a model through discovering the best, most-appropriate relationships between all measured data in each reservoir. This is an alternative to starting with the construction of mathematical equations to model the physics of the fluid flow through porous media, followed by modification of geological models in order to achieve history match.
The key characteristics of subsurface analytics are that no interpretations, assumptions, or complex initial geological models (and thus no upscaling) exist. Furthermore, the main series of dynamic variables used to build this model is measured on the surface, while other major static, and sometimes even dynamic, characteristics are based on subsurface measurements, thereby making this approach a combination of reservoir and wellbore-simulation models rather than merely a reservoir model. The history-matching process of the subsurface analytics process is completely automated.
Top-Down Modeling (TDM)
TDM is a data-driven reservoir modeling approach under the realm of subsurface analytics technology that uses AI and machine learning to develop full-field reservoir models based on measurements rather than solutions of governing equations. TDM integrates all available field measurements into a full-field reservoir model and matches the historical production of all individual wells in a mature field with a single AI-based model. The model is validated through blind history matching. The approach then can forecast a field’s behavior on a well-by-well basis.
TDM is a data-driven approach; thus, the quality assurance/quality control (QA/QC) of the data input is para-mount before embarking on the modeling process to ensure that the artificial neural network (ANN) is taught properly with reliable training of the data set. This includes the understanding of data availability and magnitude, analysis of well-by-well production performance trends, and identification of data anomalies.
Spectral methods for full-field micromechanical modelling of polycrystalline materials
