This paper describes a methodology for the integration of well logs and a series of grid-based attributes extracted from interpreted seismic data for prediction of porosity distributions. The studied area is located in the southwest Iran. Before finding a relationship between the target logs and predicted logs from 3D seismic data, we have interpreted the 3D seismic data in the studied area. Also we matched and combined well data with seismic for forward modeling and seismic inversion. Inversion produces a full band acoustic impedance model of earth which improves the vertical resolution. Then we have checked other different inversion methods such as spare spike and model based. Since the model based method resulted with a better resolution outcome, therefore we decided to apply model based inversion method in the reservoir level. In the next step we applied a linear and a non linear transforms between a group of seismic attributes and porosity logs. Then we obtained a relationship for estimating the porosity at all locations of the seismic volumetric data. Finally we found an improvement in the porosity prediction from linear multi attribute transforms when using neural network methods.
Introduction: Nowadays integration of different sources of data and building static and dynamic reservoir models has increased efficiency of exploration and production activities. Seismic reservoir characterization has crucial importance in those activities attributes, their analysis and study form are major field of study in reservoir characterization.
Acoustic impedance is one of the attributes that from some years ago has been used extensively for determination of reservoir properties, especially porosity, fluid saturation and clay content. The idea of using multi attribute seismic analysis to estimate log properties was first introduced (Schultz et al., 1994). The integration of well-log and seismic data has been a consistent aim of geoscientists. One type of integration is forward modeling of synthetic seismic data from the logs. A second type of integration is inversion of the logs from the seismic data which is called seismic inversion. The method for finding the statistical relationship between seismic data and well logs has been described by Russell et al. (1997).
We used this method in an oil field in the southwest of Iran. The data have been supplied by the National Iranian Oil Company (NIOC) and consist of a 3-D seismic data, which ties 6 wells. In this paper we applied model based seismic inversion method.
The Main Description: We interpreted the 3-D seismic data in the studied area by formal Software then depth map were prepared and also appropriate velocity field. Synthetic seismogram method is used for horizon identification. Understanding of reflection characters is essential for tracking of target horizons in the whole area. The 3-D seismic volume was interpreted at different horizons and was reproduced by the model based inversion method.
The training data set for each well consist of the real porosity, a single composites seismic trace that was extract from the 3-D seismic volume around the well and the impedance log predicated with the seismic inversion method. The attributes available for the multi attribute analysis was 25 plus external of seismic inversion.
The cross plot of the target porosity value against predicted porosity from five seismic attributes (Cosine Instantaneous Phase, 1/Inversion-Result, Filter 25/30-35/40, Derivative Instantaneous Amplitude, Amplitude Weighted Phase) are shown in Figure 5. The normalized correlation is now 0.74 For improving the result we used the radial basis function neural network (RBF) method, the RBFN, described by Powell,1987, and first applied by Ronen et al, 1994, is a feed-forward network where the Gussian bell curve is the basis function.
Conclusions: In this study we have focused mainly on 3D seismic interpretation of gas bearing layer by using different seismic inversion and multi attribute transform methods, the acquired results are as follow:
1. The optimum number of attributes in this study was five (Cosine Instantaneous Phase, 1/Inversion-Result, Filter 25/30-35/40 Hz, Derivative Instantaneous Amplitude, and Amplitude Weighted Phase).
2. The quality of inverted data (as one of the main input) is very important in multi attribute analysis.
3. We found that correlation coefficient for porosity prediction is improved in RBFN method with respect to multi attribute transform.
Acknowledgments: The authors wish to thank NIOC, Exploration Directorate, and Geophysics Department for supporting this study. We are grateful to Research Council of the University of Tehran for providing full support of this study.