Latent Space Classification of Seismic Facies

Abstract

Supervised and unsupervised seismic facies classification methods are slowly gaining popularity in hydrocarbon exploration and production workflows. Unsupervised clustering is data driven, unbiased by the interpreter beyond the choice of input data and brings out the natural clusters present in the data. There are several competing unsupervised clustering techniques, each with advantages and disadvantages. In this dissertation, I demonstrate the use of various classification techniques on real 3D seismic data from various depositional environments. Initially, I use the popular unsupervised Kohonen self-organizing maps (SOMs) algorithms and apply it to a deep-water Gulf of Mexico 3D dataset to identify various deep-water depositional facies including basin floor fans, mass transport complexes and feeder channels. I then extend this algorithm to characterize a heterogeneous Mississippian Chert reservoir from Oklahoma and map the locations of the tight/non-porous chert and limestone vs. more prospective porous tripolitic chert and fractured chert zones. The tight chert and dense limestone can be highly fractured, giving rise to an additional seismic facies. In both the case studies, a large number of potential classes are fed into the SOM algorithm. These "prototype vectors" are clustered and colors are assigned to them using a 2D gradational RGB color-scale for visual aid in interpretation.

Kohonen SOM suffers from the absence of any proper convergence criterion and rules for parameter selection. These shortcomings are addressed by the more recent development of generative topographic mapping (GTM) algorithm. GTM is based on a probabilistic unsupervised classification technique and "generates" a PDF to map the data about a lower dimensional "topographic" surface residing in high dimensional attribute space. GTM predicts not only which cluster best represents the data, but how well it is predicted by all other clusters. For this reason, GTM interfaces neatly with modern risk analysis wokflows. I apply the GTM technique to classify 15 sets of horizontal well parameters in one of the recent unconventional shale plays, correlating the results with normalized estimated ultimate recovery (EURs), allowing an estimation of EUR based on the most relevant parameters.

I extend the GTM workflow to consider multi-attribute inversion volumes and do seismic facies classification for a Barnett shale survey. With the aid of microseismic data, the clusters from GTM analysis are interpreted as brittle or ductile. I also apply the GTM technique to the P-impedance, lambda-rho, mu-rho and the VP/VS volumes from a Veracruz Basin survey in Southern Mexico that was acquired over a heterogeneous conglomerate reservoir.

Finally, I introduce limited supervision into both the SOM and GTM algorithms. The target vectors for both SOM and GTM are the average attribute vector about the different facies identified from the well logs. This supervision introduced user-defined clusters. In the preliminary supervision, I use multiattribute minimum Euclidean distance measures, comparing the results with the unsupervised SOM results. For GTM, I calculate the probability of occurrence of the well facies in the survey.

Given the appropriate 3D seismic attribute volumes, SOM and GTM workflows will not only accelerate seismic facies identification, but also with GTM, quantify the identification of different petrotypes or heterogeneities present in the reservoir zone. The final product of my dissertation is a suite of algorithms, workflows, user interfaces and user documentation allowing others to build upon and extend this research.