COMPARING BAYESIAN AND NEURAL NETWORK SUPPORTED LITHO-FLUID PREDICTION FROM SEISMIC DATA OF MEHAR BLOCK, LOWER INDUS BASIN, PAKISTAN.

Abstract

The Mehar gas field lies in the lower Indus basin is a newly developed gas field. The goal of the study was to determine the subsurface structure and characterize the reservoir potential of the Ranikot Formation and Pab Sandstone, which are main producing reservoirs of Mehar area. Seismic interpretation of 3D seismic cube data is used for delineation of subsurface structure which show the existence of a huge, bounded anticline and north to south trending thrust fault. Petrophysical study is made on Mehar well to know reservoir properties of Ranikot and Pab sandstone formations. Seismic inversion is a technique for making a connection between seismic data and interpretative elastic physical characteristics of potential reservoirs. Post-stack seismic inversion is used to estimate reservoir characteristics like as porosity and acoustic impedance in the calculation of reservoir characterization. The reservoir parameters, as well as fundamental variables such as acoustic impedance and porosity of the target zone, are delineated using post-stack time migrated seismic data (POSTM) and log data in this study. Seismic inversion and geostatistical methods were employed to finish this task. The method for inverting seismic data into acoustic impedance is essential to the study's main findings. Furthermore, a good wavelet representative of the given conditions is necessary for a positive outcome. Then, applying probabilistic neural network (PNN) techniques, geostatistical inversion is used to estimate the porosity, volume of shale and water saturation of Ranikot and Pab sandstone using well Mehar- 01,02 and 03. The impedance volume is converted to volume of shale, porosity and water saturation volume using PNN, and the results are compared using the petrophysical parameters of wells. Bayesian classification is run in 3D data cube to predict the lithologies and fluid present in the reservoir. For Bayesian classification two type of cross plot are used for prediction of lithology and fluid, P-impedance vs Vp/Vs for fluid prediction and Lambda-Rho vs Vp/Vs for lithology prediction with density plotted on Z axis. The cross-plots clearly separated and delineated the lithofluid classes (wet sand, gas sand, shale, and limestone) with specific orientation/patterns.