Development of ANN-based predictive model for miscible CO2 flooding in sandstone reservoir

Nur Iman Khamidy, Zeeshan Tariq, Zuher Syihab

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

11 Scopus citations

Abstract

Up to 2010, 44.55% of 312 EOR's project for light oil implemented around the world in sandstone reservoirs were come from continuous miscible gas CO2 injection which contributed to an incremental Recovery Factor (RF) of about 34.5% for less than 10 years of production period. This fact has triggered many oil industries to apply this potential and proven technogy for their assets. This potential comes with the needs of having a robust tool to forecast additional recovery due to CO2 injection. This work focuses to development of predictive model using artificial neural network (ANN). More than 6000 series of input-output parameters for ANN training and validation/testing data are extracted from numerical reservoir simulator of 1/8 of five-spot pattern models. The models are set as combination of reservoir geometry, rock, fluid and well operating condition parameters within the range of CO2 EOR screening criteria. The main objective of this work is to find the best ANN architecture/model which accurately matches reservoir simulation results, especially the relationship of RF, total volume of injected CO2 (GI) and the reservoir characteristics and well operating conditions. Trial and error of ANN architectures and parameters are done on number of hidden layers, number of neurons for each hidden layer, learning rate (LR) value, and momentum constant (MC) with minimization algorithm (Lavenberg-Marquardt) in Feed-Forward Back Propagation (FFBP) schemes under log-sigmoid transfer function. An optimum result of ANN model is achieved with an architecture of 18-26-11-2. The relative error of RF and GI of the ANN model are within range of 3 to 10% respectively. A better average relative error of RF and GI of 2.8% and 4.15% respectively are obtained after removing the outliers (unrealistic combinations of input data) from training process of the ANN model. Furthermore, it is clearly found that oil viscosity plays the most the important factor in CO2 EOR method.

Original languageEnglish (US)
Title of host publicationSociety of Petroleum Engineers - SPE Middle East Oil and Gas Show and Conference 2019, MEOS 2019
PublisherSociety of Petroleum Engineers (SPE)
ISBN (Electronic)9781613996393
StatePublished - 2019
EventSPE Middle East Oil and Gas Show and Conference 2019, MEOS 2019 - Manama, Bahrain
Duration: Mar 18 2019Mar 21 2019

Publication series

NameSPE Middle East Oil and Gas Show and Conference, MEOS, Proceedings
Volume2019-March

Conference

ConferenceSPE Middle East Oil and Gas Show and Conference 2019, MEOS 2019
Country/TerritoryBahrain
CityManama
Period03/18/1903/21/19

Bibliographical note

Publisher Copyright:
© 2019, Society of Petroleum Engineers.

Keywords

  • ANN
  • Artificial Neural Network
  • Continuous miscible gas CO2 flooding
  • Five-spots pattern
  • Homogeneous
  • Predictive model
  • Sandstone reservoirs

ASJC Scopus subject areas

  • Energy Engineering and Power Technology
  • Fuel Technology

Fingerprint

Dive into the research topics of 'Development of ANN-based predictive model for miscible CO2 flooding in sandstone reservoir'. Together they form a unique fingerprint.

Cite this