Improving the accuracy of flow units prediction through two committee machine models: An example from the South Pars Gas Field, Persian Gulf Basin, Iran

  • Authors:

  • Javad Ghiasi-Freez;Ali Kadkhodaie-Ilkhchi;Mansur Ziaii

  • Affiliations:

  • Faculty of Mining, Petroleum and Geophysics, Shahrood University of Technology, Shahrood, Iran;Geology Department, Faculty of Natural Science, University of Tabriz, Tabriz, Iran;Faculty of Mining, Petroleum and Geophysics, Shahrood University of Technology, Shahrood, Iran

  • Venue:
  • Computers & Geosciences
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Intelligent reservoir characterization is a prerequisite study for development of oil and gas fields. Hydraulic flow units are mappable portions of hydrocarbon-bearing rocks that control fluid flow, and their modeling allows an accurate understanding of reservoir quality within a hydrocarbon field. The current study presents two committee machines based on intelligent models to make a quantitative formulation between flow units and conventional log responses in the South Pars Gas Field, Iran. First, a committee machine with intelligent systems (CMIS) is constructed using a genetic algorithm-pattern search technique. The overall mean squared error and coefficient of determination between the measured and predicted flow zone indicators (FZI) using the CMIS for the test data are 0.1468 and 0.775, respectively. Afterwards, a committee fuzzy inference system (CFIS) is constructed. For this purpose, the training data are divided into two individual clusters based on the FZI values. The two FZI clusters are trained with the individual fuzzy inference systems, and a classifier network assigns appropriate weights to each cluster. The MSE and coefficient of determination of the CFIS are 0.1233 and 0.812, respectively. The CFIS shows some improvement in the accuracy of predictions in comparison with the CMIS. The results of this study demonstrate the higher performance of the committee machines compared to the individual expert systems for estimating reservoir properties. Moreover, a primary clustering of the model data and their training with the individual expert systems can lead to a considerable improvement in the accuracy of committee machines.