Resolution analysis in well log estimation by using neural networks: Eastern Venezuela

Summary Neural networks have been used successfully in many geophysical applications for pattern recognition and property estimation. In this study, we take advantage of the strength of neural networks to predict well logs using seismic attributes at an Eastern Venezuelan field. Due to the depth of the target (more than 16000 ft.), the frequency content of the seismic data is low, so a resolution study was required to quantify the quality of the estimations. A neural network is trained at eighteen well locations with sonic velocities and seismic attributes to build a nonlinear mathematical model. This model is applied to the whole 3D seismic data to generate a pseudo sonic velocity cube. The results shown that, the neural network estimations improved the resolution in the pseudo logs compared to those obtained directly from seismic inversion.