Prediction of Petroelastic Properties of Rocks Based on Machine Learning

The research is related to the need for seismic modeling of Western Siberia oil deposits in order to clarify geological cross-sections and structures of oil deposits. In order to qualitatively match seismic data and a geological section and then interpret them, it is necessary to have acoustic and density log data in all wells. The presence of these data and their quality are of particular importance in solving problems of seismic inversions. But in some wells, these data are missing or of very poor quality. Thus, we need to restore the acoustic and density logs using other well logging data. Paper presents the results obtained by using neural networks. Neural networks are an effective data processing mechanism for predicting results of various kinds. This technology was applied in this research to predict the petrophysical properties of the formation. The prediction of interval time and density was made. The results agree with the initial data for the cross-section. The elastic parameters are calculated for seismic modeling. Their comparison made it possible to separate the rocks according to lithotypes into reservoirs and non-reservoirs and to construct structural maps taking into account the refinement of the distribution of reservoirs over the field area.