Reservoir Characterization Using Seismic Event Attributes and Neural Networks

This paper presents an approach to map reservoir physical properties using event-based seismic attributes and neural networks. The technique uses single trace windowed seismic attributes and well log reservoir physical properties to train the neural network. After the training, the response of the neural network over all of the windowed seismic data set is a map with the reservoir physical property. The technique applied on actual data shows excellent results. The neural nets have attracted considerable interest in the recent years, due to the acknowledgment of its potential on doing a variety of routine and exhausting jobs of classification. This kind of job is found in the most diversified fields of activity, such as the processing of signals and the definition of patterns, where the neural networks appear as potential alternative to the numerical computation. One of its best characteristics is the versatility in learning a mapping unknown function within the input data and the expected result, making the correlation possible. In the seismic exploration, many references about the subject are found, some of which, are hereby mentioned. Kou- Yuang Huang et al (1990) used neural networks to pick seismic horizons. J. Schmidt and F. A. Hadsell (1992) used neural networks for seismic stacking velocity picking and D. H. Johnston (1993) for seismic attributes calibration. C. T. Kalkomey (1997) pointed out the potential risks when using seismic attributes as predictors of reservoir properties. C. Gastaldi et al (1997) shown an example of the application of seismic attributes for reservoir characterization.