Summary
Seismic facies analysis provides an effective way to estimate the properties of reservoir and characterize its heterogeneity. The conventional approach used to generate a map of seismic facies utilizes the poststack seismic attributes. In order to find more complex and atypical reservoirs, it is increasing essential to analyse the reservoir based on prestack seismic data which carries abundant stratigraphic and depositional information. In this paper, we develop a prestack reflection pattern based seismic facies analysis methodology. We extract an augmented Gabor feature derived from the Gabor wavelet representation of prestack data. Then, combining the feature with pattern recognition techniques, seismic facies analysis can be achieved through unsupervised or supervised learning. We tested the effectiveness of our method with application to the wide-azimuth data form LZB region and the CMP gather data from Sulige region, The results are geologically intriguing, and they demonstrate the vast superiority over the conventional approach.
Introduction
Seismic facies analysis is the term used for techniques that depict enough variability of seismic data to reveal details of the underlying geologic features. Therefore, the use of seismic facies classification techniques has been steadily increasing in hydrocarbon prediction process over the past 20 years(Coléou, Poupon, and Azbel 2003).
The object of seismic facies analysis can be achieved through the use of pattern recognition algorithm to seismic attribute volumes. The effectiveness of seismic facies map greatly depends on the proposed attributes. There are several hundred surface and volume attributes based on descriptions of seismic stratigraphic reflection patterns. Seismic trace amplitude is the most commonly used attribute for seismic facies analysis(Matos et al. 2006). In order to characterize both intratrace and intertrace relationships of amplitudes, coherence and gray-level cooccurrence matrix based attributes were adopted for identify certain geological feature(Chopra and Alexeev 2006, Chopra and Marfurt 2005, Gao 2003). West et al.(2002) used gray level co-occurrence matrix features and neural network to generate seismic facies. There is no rule for the selection which attribute can best illustrate rock property changes. Roy(2013) reported on applying a combination of different attributes with self-organizing neural network(SOM) and generative topographic mapping(GTM) for seismic facies analysis. Matos et al.(2006) used wavelet transform to identify seismic trace singularities which are less sensitive to horizon interpretation errors, thus resulting in an improved facies analysis.
