A New Data-Drive Approach for Seismic-Reservoir Characterization and Comparison with Traditional Algorithm

Summary We have evaluated a data-drive algorithm based on backprogation (BP) neural network for seismic-reservoir characterization. The applicability and reliability of this method are assessed by comparison with the traditional, linear rock-physics model (RPM) based inversion algorithm, and Bayesian algorithm. First, elastic parameters are estimated from seismic data by conventional inversion method. Then, we take advantage of the three methods to estimate the petrophysical properties of interest and facies from elastic parameters. However, the main difference among the three methods remains the different assumptions. Some limitations and assumptions can be overcome by the data-drive algorithm based on BP neural network. Among the above three methods, the proposed method is the most effective to predict petrophysical properties and facies. Numerical experiment on the real data, confirms the validity of the proposed approach.