Clustering as an efficient tool for assessing fluid content and movability by resistivity logs

Pedram Masoudi (1, 2) , Babak Nadjar Araabi (3) , Tahar Aïfa (2) , Hossein Memarian (1)

1 School of Mining Engineering, University College of Engineering
2 GR - Géosciences Rennes
3 Department of Electrical and Computer Engineering

Neither core measurements nor well tests provide precise measurement of fluid contents; so, at the moment there is no validated saturation measurement in the oil industry. However, resistivity logs contain some valuable information about reservoir fluids, and classes of saturation based on the behaviour of resistivity data. In order to extract saturation value, clustering algorithms are proposed and tested here. Clustering is an unsupervised categorization method, which relies on natural groupings of real data, instead of predefined labels. Distance or dissimilarity plays an important role in the formation of clusters in an algorithm. The application of three clustering algorithms on prediction of water saturation is discussed here. It is shown that Fuzzy C-Means clustering divides data only according to saturation property; while the Gustafson-Kessel algorithm considers not only saturation but also permeability. The reason is that Gustafson-Kessel can detect linear patterns. This algorithm is introduced as the most appropriate clustering method for predicting permeability, and for understanding the reservoir quality of the formation under examination. Gath-Geva clustering did not provide as much information as Gustafson-Kessel. In addition to these three clustering algorithms, two other methods were likewise checked, but they did not provide acceptable interpretations, so not reported. Another achievement of this study is the introduction of a cluster label instead of a single value which is very unreliable for expressing saturation and permeability simultaneously. Predictions of these two petrophysical properties are provided merely by two conventional resistivity well-logs: deep and shallow. The output of cluster analysis is much closer to well-scale reservoir properties as compared with that of core-scale properties, since clustering algorithms are applied to logs that are volumetric recordings. The applicability and efficiency of the proposed methods are examined and verified through the use of well-logs of Sarvak Formation in an anticlinal oil field in the Abadan Plain, Iran.

Mots clés: Gustafson-Kessel clustering; water saturation in carbonate reservoirs; clustering permeable zones; clustering saturated zone; application of unsupervised learning

Domaines: Géologie appliquée

Pedram Masoudi, Babak Nadjar Araabi, Tahar Aïfa, Hossein Memarian. Clustering as an efficient tool for assessing fluid content and movability by resistivity logs. The Fourth International Mine and Mining Industries Congress & The Sixth Iranian Mining Engineering Conference, Nov 2016, Téhéran, Iran. 9 p. ⟨insu-01396454v2⟩