Application of Multilayer Feed Forward Neural Network - Porosity Mapping in Gas Hydrate Sediment of K-G Basin, India

Abstract

SUMMARY Porosity has been estimated in the gas hydrate bearing sediments in Krishna-Godavari (K-G) basin at site National Gas Hydrate Programme (NGHP) 01-10 using multilayer feed forward neural network (MLFN). A series of elastic parameters namely P-wave velocity (Vp), S-wave velocity (Vs), density (Dn), Vp/Vs, P impedance (Zp) and S impedance (Zs) have been derived from seismic data using pre-stack inversion. The porosity (O_D) estimated from the density log for the depth interval 1024-1224m is ranging from 51.6 to 82.6 %. The estimated porosity has been treated as target log and Zp, Zs, Vp/Vs and Dn have been used as input parameters during the training of MLFN. The trained network is then used to train the network to generate subsurface porosity along a 2D multi channel seismic section (line-16) at site NGHP01-10 within the time interval of 1.433 – 1.631s corresponding to the depth interval of 1024-1224m. The predicted porosity is decreasing in gas hydrate stability zone (GHSZ) and is varying laterally. Porosity is increased in the free gas bearing zone below the base of GHSZ