Active fuzzy modeling for estimating problems in hydrocarbon reservoirs. Issue 7 (October 2016)

• Record Type: Journal Article
• Authors: Fasanghari, Mehdi; Bahrpeyma, Fouad; Jolai, Fariborz
• Is Part Of: Neural computing & applications. Volume 27:Issue 7(2016)
• Page Start: 1981
• Page End: 1992
• Publication Date: 2016-10
• Subjects: Active fuzzy modeling -- Active learning method -- Ink drop spread -- Hydrocarbon reservoir; Neural networks (Computer science) -- Periodicals; Neural circuitry -- Periodicals; Artificial intelligence -- Periodicals; Neural Networks (Computer) -- Periodicals; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux nerveux -- Périodiques; Intelligence artificielle -- Périodiques; 006.32
• DOI: 10.1007/s00521-015-1992-y
• Languages: English
• ISSNs: 0941-0643

Abstract

Abstract While active learning method (ALM) uses error as the learning parameter, selection of the validation data is still challenging. In this paper, to prevent form encountering with sample size problem, we applied an error-independent version of ALM that we call the active fuzzy modeling (AFM) with a distance threshold to model parameters of hydrocarbon reservoirs. In this paper, we demonstrate that measuring the generalization error is a vital factor in the process of ALM. Regression (R ) and mean squared error (MSE) for estimating RHOB by AFM were 0.96 and 0.0032, respectively. On the other hand, R of 0.91, 0.89 and 0.92 and MSE of 0.0051, 0.0067 and 0.0047 for ANN, TS-FIS and NF, respectively, illustrate that AFM performs much better in comparison with conventional modeling approaches and produces more reliable results. Comparing the results of the presented method with ANN, TS-FIS and NF in aspect of rapidity, robustness, storage, complexity and acceptability in estimating RHOB reports the accuracy and high-performance behavior of AFM. This method is illustrated by an example of an oil field at NW Persian Gulf.