Wan Zhonghong

Reservoir property prediction using the dynamic radial basis function network

Summary The nonlinear relationship between seismic attributes and reservoir property can be estimated by many statistical methods, such as machine learning and neural networks. However, an adaptive process for parameters learning cannot be implemented in most statistical methods. In this paper, we propose a dynamic radial basis function (D-RBF) network method to predict the reservoir property from seismic attributes. The main feature of this method is that the nonlinear relationship between the reservoir property and seismic attributes can be determined adaptively without interpreter intervention. In fact, if the current RBF network cannot reduce the prediction errors in the system, the proposed method will increase or decrease the hidden units in the RBF network to re-estimate the relationship between the reservoir property and the seismic attributes. We illustrate the proposed method using two real 3D seismic data sets. The results show that models based on D-RBF networks can predict reservoir property with high accuracy.

Detecting carbonate‐karst reservoirs using the directional amplitude gradient difference technique

This kind of carbonate reservoirs is abundant, they consist mainly of tectonic fractures, diagenetic fractures, stylolite, matrix porosity, dissolution porosity, and karst caves. Among them, the highly heterogeneous karst cave is the main reservoir space for hydrocarbon accumulation. In seismic section, it exhibits a strong heterogeneity no matter horizontally or vertically and shows as “a string of beads” (Figure 1). Moreover, unlike carbonate reservoirs in other areas, Ordovician carbonate-karst reservoirs in the Tarim Basin are deeply buried, and their depth can reach 6 to 7 km.

Spectral decomposition and derived techniques for clastic reservoir identification and its application

Summary Spectral decomposition can be an effective seismic data analysis method in the frequency domain. It can be used to identify reservoirs and, in favorable conditions, can help in the detection of hydrocarbons. Many techniques derived from spectral decomposition have been developed in recent years. In this paper, we examine four representative techniques and discuss the advantages and shortcomings of each. The mono-frequency analysis method is found to be the simplest but does not use fullspectrum information; the multi-frequency analysis and spectrum characteristic analysis methods are intelligent and complex in some sort. Compared with other methods, the strata absorption coefficient analysis method is complex but more robust, practical and effective；Its application to the TM3D survey provides an example of its success.

Joint poststack P- and PS-wave impedance inversion and an example from northern China

The main structure of the SLG gas field, in Ordos Basin in northern China, is a mild monocline with a dip angle of less than 1°. Tectonic movement is minimal with some nose-like structures having relief of less than 20 m (Figure 1).

Automatic Geological Body Identification Using the Modified Rival Penalized Competitive Learning Clustering Algorithm

In the past few years, the use of multi-attribute seismic volume classification has been widely studied in reservoir analysis and interpretation workflows to identify abnormal geologic characteristics. Generally speaking, the identifycation process can be very difficult and time-consuming. In this paper, we present an unsupervised approach based on the modified rival penalized competitive learning (MRPCL) theory to identify 3D geologic characteristics automatically. In the proposed algorithm, a new cost function and some parameter learning methods will be introduced to operate the identification of geologic characteristics effectively and to determine the number of seismic facies automatically. By clustering the multiattribute seismic data, different geologic bodies can be extracted directly from the 3D seismic data volume. Finally, this proposed approach is demonstrated on a simple 2D synthetic data set and on two real 3D seismic data sets.