Methodology-selection by Fuzzy Analytic Hierarchy Process for Studying Net Pays
Résumé
Net pay intervals are conventionally determined, applying cutoff values on geological well-logs. Recently developed methodologies utilize more complicated algorithms: Bayesian classifier, artificial neural network and Dempster-Shafer theory. The outputs of these methodologies are not completely compatible. It rises this question that which method should be recommended in each situation: (i) industrial use, (ii) research goal and (iii) general situations. Fuzzy analytical hierarchy process is used here to compare effectiveness of the four net pay determination methods in each three situations. Six criteria were defined: precision, generalization ability, fuzziness, simplicity of methodological concepts, user-friendly and speed of the algorithm. For the precision and generalization ability, mean squared error of training and generalization data are used, respectively. Mean squared error and speed (inverse of time) of the algorithm are continues variables, and provide quantitative comparison. While qualitative comparison is done for the criteria of simplicity and user-friendly. For the criterion of the fuzziness, a ranking, i.e. categorical variable, is used based on the number of classes that the classifier could provide. The comparison is done based on the results of net pay determination (the four methods) on sandy Burgan and carbonated Mishrif reservoirs, Iranian offshore oilfields. The results show that from viewpoint of general situation, Bayesian and Dempster-Shafer-based methods are the best. Artificial neural network and Dempster-Shafer are the most suitable methods for industrial mode, and Bayesian and artificial neural network are the best methods for research applications. Finally, cutoff methodology is never prioritized.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...