Generalized data mining analysis on well logs for facies classification

Abstract EN

Well logs are petro-physical data recording properties of fluid and rocks in oil & gas wells.

One important aspect of well log interpretation is to understand downhole lithology across drilling interval.

This paper is based on the work of the “Facies Classification Using Machine Learning” authored by Brendon Hall [1] in different software environment.

The dataset was from [1], which was originally taken from a University of Kansas class exercise on the Hugoton and Panoma gas fields.

First, we examined a set of well-logs for a completeness of petrophysical data.

Then, we applied different classification algorithms which is not widely spread between different models and capture the different dependencies between the well logs and wireline information in the lithology and facies classification.

Finally, we proposed an algorithm which has the higher precision and more generalized compared with training and testing data set, as the best suitable for the facies classification.

In our case, we arrived at a conclusion of Bayesian algorithms which gave us a simple model with more common and narrow results.