Time-Lapse Seismic Inversion for Predicting Reservoir Parameters Based on Dual-Network

Abstract

Summary Prediction of reservoir dynamic parameters is an important technology for understanding reservoir dynamic changes and guiding resource development. Time-lapse seismic exploration provides an effective means for predicting reservoir dynamic parameters; however, the lack of time-lapse logging curves and prediction accuracy limits its application. To alleviate this problem, we propose a two-stage dual-network (DN) prediction method. The first stage establishes the relationship between the baseline logging curve and seismic attributes through improved deep feedforward neural network (TP-DFNN) in order to predict the baseline reservoir parameters. The second stage establishes the relationship between the baseline seismic data and reservoir parameters through a fully convolutional neural network (FCN) to realize reservoir dynamic parameter prediction. The proposed method can calculate both reservoir elastic parameters (P- and S- wave velocities, density) and reservoir physical parameters (porosity, permeability, Poisson’s ratio, resistivity, etc.). Only the baseline logging curve is needed, and there is no need to set up an initial model in the DN inversion. The test results of the field data show that the proposed method can efficiently and accurately predict the time-lapse changes in reservoir parameters.