Abstract: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 The paper has not been peer reviewed.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.

On application of machine learning method for history matching and forecasting of times series data from hydrocarbon recovery process using water flooding

Data-driven approach for hydrocarbon production forecasting using machine learning techniques

Reinforcement learning based automated history matching for improved hydrocarbon production forecast

The Power of Predictive Analytics in Oil Field Development: Integrating Machine Learning with Reservoir Hydrocarbon Data to Enable Enhanced Oil Recovery of Hugin Formation within the Theta Vest Structure

Machine learning for recovery factor estimation of an oil reservoir: A tool for de-risking at a hydrocarbon asset evaluation

Insights into the Application of Machine Learning in Reservoir Engineering: Current Developments and Future Trends

Machine Learning and the Internet of Things Enable Steam Flood Optimization for Improved Oil Production

Advanced Deep Regression Models for Forecasting Time Series Oil Production

Time-series well performance prediction based on Long Short-Term Memory (LSTM) neural network model

An Artificial Intelligence Neural Networks Driven Approach to Forecast Production in Unconventional Reservoirs – Comparative Analysis with Decline Curve

Application of convolutional long short-term memory for spatio-temporal forecastings of hydrocarbon saturations and pressure in oil fields

Real-Time Liquid Rate and Water Cut Prediction From the Electrical Submersible Pump Sensors Data Using Machine-Learning Algorithms

Machine learning based decline curve analysis for short-term oil production forecast

Reservoir Simulation, Modeling of a Complex Offshore Field Uses AI and Machine Learning

A novel hybrid recurrent convolutional network for surrogate modeling of history matching and uncertainty quantification

Application of Data Analytic Techniques and Monte-Carlo Simulation for Forecasting and Optimizing Oil Production from Tight Reservoirs

Prediction of single well production rate in water-flooding oil fields driven by the fusion of static, temporal and spatial information

Hybrid Machine Learning for Modeling the Relative Permeability Changes in Carbonate Reservoirs under Engineered Water Injection

Data-Driven Models to Predict Hydrocarbon Production From Unconventional Reservoirs by Thermal Recovery

A Machine Learning Approach for Virtual Flow Metering and Forecasting

Improving Short-range Reservoir Inflow Forecasts with Machine Learning Model Combination