Efficiency Gains in Inversion-Based Interpretation Through Computer-Driven Classification

Abstract

I have applied a Kohonen self-organized mapping algorithm (SOM) to increase the efficiency of inversion-based interpretation by introducing non-specialists to the interpretation of AVO inversion products. Specifically, the utilization of acoustic impedance (AI) and gradient impedance (GI) volumes, calculated from a colored inversion, with a SOM, will allow any geoscientist to easily identify areas of interest that are traditionally only interpreted by specialists. This application of a SOM can shorten the common bottleneck between interpreters and AVO experts. Following the identification of these seismically anomalous regions, appropriate specialists are more effective at discerning the subtle geophysical meaning localized around these SOM-based anomalies. The result is a minimization of the time a specialist is required and maximization of the engagement of the general interpreter. Progressive seismic interpretation, an integrated specialty consisting of quantitative seismic interpretation (QI), seismic attribute analysis, and associated workflows (e.g. machine learning). By using an ever-evolving and often state-of-the-art (or progressive) methodology, the practitioner can now understand the contents of a seismic dataset more fully. Therefore, employing these workflows is often a required step in the evaluation of a prospect in the oil and gas industry. The largest drawbacks of these methods are directly associated with the cost, both in monetary terms and in time, of the specialists that are required to play a heavy role in the entire evaluation process. From the initial screening of leads to the assessment of a prospect before drilling, these seismic specialists are often in short supply (or simply non-existent in some instances), which is sometimes a painfully obvious inefficiency. The result is an increase the time required to evaluate a prospect, increase of the cost, and possibly lower quality interpretation results. By employing a computer-based classification algorithm from a progressive interpretation perspective, it is possible to classify a seismic volume without direct human interaction. The resultant classes from such a method are composed of groups of data that either represents different parts of the seismic background (i.e. the common background geology, Figure 1), or anomalous regions in the dataset (Figure 2). Since most targets in exploration geophysics described as rare occurrences, these anomalous regions are of significance.

AAPG Datapages/Search and Discovery Article #90291 ©2017 AAPG Annual Convention and Exhibition, Houston, Texas, April 2-5, 2017