%D 2020
%R 10.1016/j.jappgeo.2020.104161
%O cited By 0
%J Journal of Applied Geophysics
%L scholars12582
%K Elastic moduli; Hydrocarbons; Learning systems; Offshore gas fields; Oil bearing formations; Oil well logging; Petroleum reservoir engineering; Petroleum reservoirs; Regression analysis; Rocks; Seismic waves; Shear flow; Well logging, Correlation coefficient; Effective approaches; Elastic properties; Empirical relationships; Forward modelling; Hydrocarbon predictions; Hydrocarbon reservoir; Indicator methods, Deep learning, computer simulation; detection method; empirical analysis; forward modeling; hydrocarbon reservoir; machine learning; Monte Carlo analysis; numerical model; P-wave; regression analysis; reservoir characterization; shear modulus; uncertainty analysis; well logging, Malay Basin; Pacific Ocean; South China Sea
%X Hydrocarbon reservoirs are commonly identified based on their elastic properties, with high success rates in many cases. However, the porosity and rock matrix may interfere with the fluid effect. The brine-saturated rock may have a similar response with the hydrocarbon reservoir, which cause misinterpretation. A new method, the �advanced fluid indicator� method, is proposed to reduce the uncertainty in reservoir detection and characterisation by the rotation of the P-wave and shear moduli. This advanced fluid indicator exhibits the ability to identify the fluid type. The local empirical relationships are extracted from the well log data. Then the Monte Carlo forward modelling is constructed using the relationships. The advanced fluid indicator is calculated using the relative parameters. The correlation coefficient between the advanced fluid indicator and bulk modulus of the pore fluid reaches 0.9935 in the forward modelling, which indicates that the new method can be used to detect the fluid type directly. Deep learning is then applied to train a regression model using the rock physics dataset derived in the forward modelling. The advanced fluid indicator is calculated by the trained regression model using well log data. This method has been applied in S field in Malay Basin, South China Sea. It successfully detects all the reservoirs that are identified in the well log section. Furthermore, the advanced fluid indicator values indicate the fluid type of each reservoir. While the three currently used crossplot methods can also detect the three gas reservoirs, they are unable to separate the two oil reservoirs from non-hydrocarbon-bearing rocks clearly. Therefore, the proposed advanced fluid indicator method outlines an effective approach that combines hydrocarbon prediction and deep learning to detect and characterise hydrocarbon reservoirs with higher accuracy and lower ambiguity than current methods. © 2020 Elsevier B.V.
%T Advanced fluid indicator based on numerical simulation and deep learning
%I Elsevier B.V.
%A C. Liu
%A D.P. Ghosh
%A A.M.A. Salim
%V 182