Reservoir characterization from seismic waveform using forward modeling and pattern recognition

摘要

Suppose we are given several wells in a seismic survey area. Each has a litho-logical column (or facies log), and sonic and density logs. The lithological section defines a geologic section type―a specific set of target layers with defined thickness, lithology, porosity, presence of hydrocarbons, and so on. The problem is now to predict the geology away from these wells. This is a problem in seismic pattern recognition, which has been studied for more than 30 years (e.g., Mathieu and Rice, 1969). Given several wells in the seismic survey area, the prediction process follows these steps: 1. Select an informative part of the seismic record and a set of seismic attributes. 2. Select reference traces around the wells. 3. Using the seismic attributes, reference traces, and a pattern recognition program, classify all seismic traces. 4. Ascribe the geologic properties from each well to all zones in which the seismic waveform is similar to the one found in the well's vicinity.