Automatic Facies Classification from Well Logs

Facies grouping is a vital assignment which can improve the odds of accomplishment of a well altogether. The pertinent characterization calculations accept well logs as data sources and group the development into unmistakable bunches or electrofacies. Incorporating the electrofacies with center estimations can prompt a comprehension of the topographical facies. We build up a broadly useful work process for solo electrofacies order, which accepts well logs as sources of info and can be utilized for various application situations time connection. The bunching is performed utilizing the Gaussian blend model methodology. The ideal number of groups is consequently decided guaranteeing repeatable bunching results from different acknowledge of the characterization work process. The work process was applied on field information from seaward Norway. We watch high closeness in the subsequent facies with the ones decided outwardly by the field geologist from center information, by looking at their penetrability porosity connections. This new methodology expels the client mediation in the work process and gives a vigorous answer for mechanizing the electrofacies order handling.

Facies arrangement is a key component in the assessment of petrophysical developments and in store portrayal. Electrofacies are characterized as groups of comparable log reactions in a well or a lot of wells and their mix with center estimations can prompt topographical facies, which can speak to arrangement of petrophysical properties. There has been huge advancement towards creating robotized work processes for facies characterization (Busch et al, 1987; Lim et al. 1997; Rabaute, 1998; Qi and Carr, 2005; Skalinski et al., 2006; Tang et al., 2011).

There are three principle challenges in electrofacies characterization. To begin with, the way that the greater part of the occasions there are no marked information requires to utilize an unaided arrangement strategy. There are different unaided learning calculations like the k-implies (Lloyd, 1982) or the various leveled grouping calculation (Ward, 1963) to perform order. Notwithstanding, these calculations perform “hard” task of information focuses to groups, in which every information point is related particularly with one bunch (Bishop, 2006) and they don’t consider the way that field information can have some vulnerability over the groups they are doled out. Second, the ideal number of groups is normally obscure and in this manner is required to be an info given by the client. Different methodologies have been created to maintain a strategic distance from the client’s subjectivity in the decision of the ideal number of bunches and robotize the procedure. Probably the most widely recognized ones are the Bayesian Information Criterion (Schwarz, 1978) and the Cross-Entropy Clustering (Tabor and Spurek, 2014), which anyway don’t give a generally powerful arrangement. At last, it is normal various acknowledge of the grouping calculation to give diverse bunching results, regardless of whether the information logs and the calculation boundaries are saved the equivalent for all acknowledge. This is on the grounds that every one of the information boundaries of the calculation is introduced arbitrarily for every acknowledgment and thus the calculation meets to an alternate estimation of loglikelihood and the grouping is distinctive each time subsequently.

Leave a Reply

Your email address will not be published. Required fields are marked *