Comparison of quantitative attribute analysis methods for forecasting reservoir thicknesses based on seismic data

Abstract

Three main methods for predicting effective thicknesses using seismic attributes are analyzed: linear regression based on full-wave seismic modeling data, cokriging of “attribute-well” values, neural network forecast based on a group of attributes “trained” on wells. The weathering crust in the Pre-Jurassic complex of Western Siberia was chosen as a methodological example, since with its relatively small capacity (from 0 to 50 meters) reflections from the roof and the sole are not separated. The amplitude-frequency response of this interference reflection depends on the thickness, and therefore some seismic attributes may react to this.

The analysis of independent forecast results by all three methods, their advantages and disadvantages is given. For the first time, all the basic calculations were performed using domestic software.