ESTIMATION OF SHEAR WAVE VIA FEED-FORWARD NEURAL NETWORK AND PETROPHYSICAL BASED THIN BED AVO MODELING OF ZAMZAMA GAS FIELD

Abstract

Geoscientists continue to be challenged by the complexities of petroleum reservoir systems. Due to a lack of high-quality data, there is a lack of understanding of reservoir dynamics, resulting in inaccurate reservoir potential estimation. .Shear wave velocities have a wide range of applications in petrophysical analysis and in accurate reservoir characterization. Shear wave velocity helps to improve the reservoir and fluid prediction accuracy by decreasing seismic-based uncertainty. However, acquiring S-wave velocity is difficult and costly, resulting in the lack of S-wave velocity data for some gas and oil reservoirs. Therefore, estimation of precise S-wave velocity has become a serious challenge. In this study, three-layer Feed Forward Neural network model (FFNN) in combination with feature importance has been utilized to estimate the shear wave log (DT4S). Prior to neural network application, Feature importance has been to analyze the input data to get optimum results. Out of available 46 logs, 6 logs (DT, LLS, LLD, NPHI, GR, PEFZ) have been chosen for the development of FFNN. The model contains 1 hidden layer containing 16 neurons. 70% data has been used to train the FFNN and remaining 30% data was used for testing. The obtained results (Zamzama-01 well) shows 87% correlation with the measured log. To further check the credibility of this designed network, it was applied on another area where input logs cover the early cretaceous rock. The trained network provides excellent matching (84%) with the measured log. After checking the credibility of designed neural network, DT4S log was also estimated in Zamzam-03, and 05 well where it does not acquired. Estimates DT4S log in combination with other elastic and reservoir parameters were crossplotted in order to distinguish the hydrocarbon bearing zone. The crossplot clearly separate the gas bearing and non-gas bearing zones. Further petrophysical investigation of Zamzama-01, clearly indicate that Pab-sandtone formation comprises of thin sand beds that serve as suitable reservoirs. These beds have good porosity values (11%), low shale volume (12%) and low water saturation (25%). To study these thin beds, three-layer forward AVO model, and wedge modeling have been used. Wedge modeling results indicate that bed having thickness of 26m, 22m, and 19m can be identified if seismic data have dominant frequencies of 25Hz, 30Hz and 35HZ, respectively. Moreover, AVO forward modeling shows Class IV AVO response from the top layer of gas-bearing sand