Delineation of the Runaway Sequence, Boonsville Field, Texas, Combining Similarity Analysis and Neural Networks

Abstract

In order to reduce the uncertainty level derived from analyzing only the seismic attribute maps, two types of analysis were integrated to characterize, seismically, the Runaway Sequence of the Boonsville Field (Texas, USA): a similarity analysis and an unsupervised neural network. From the observation and comparison of both techniques, patterns in the seismic data, which could be associated with prospective areas, were identified and delimited. As input data, maps of seismic interval attributes were generated and wells with production logs were located within the Runaway Sequence. To perform the Similarity Analysis, a MATLAB routine was modified and adapted to generate similarity maps relative to wells with different production characteristics. The maps obtained were contrasted with each other to determine differences between the regions of seismic similarity associated with the presence or absence of hydrocarbons. With the Unsupervised Vector Quantizer (UVQ) network, a classification of seismic attributes was carried out, selecting, as training data, values extracted, at random points, from the maps of interval attributes of the Sequence Runaway, and a number of 10 groups or facies. The results indicate that the similarity analysis could provide information regarding the presence and type of fluid (gas or oil), while the neural network maps show good facies discrimination, which could be associated with the existence of a lithological factor influenced by porosity in the map responses obtained.