Quantifying Sand Fraction from Seismic Attributes using Modular Artificial Neural Network

The goal of this paper is the blind prediction of reservoir property from seismic attributes for well tops guided zones. Diverse variation in the reservoir properties, vertically and laterally, shows the nonlinear and complex nature of the reservoir system. In this context, use of a single network for the prediction of reservoir characteristic for a complete well may not be good in achieving target property. This study implements a modular neural network to predict sand fraction between the well tops where three seismic attributes (Amplitude, frequency and inverted impedance) are taken as input characteristics. Total depth range of the wells has been divided into three different units separated by well tops which are analyzed from the log data. Eight wells are used to model sand fraction from seismic attributes. In the process of modeling, the connecting pattern between inputs and target are trained with three different networks from seven wells and then trained networks are applied to blindly predict reservoir characteristic for remaining well which was not used in the training process. Target characteristic has been obtained from three different networks and in three depth ranges; and then the results obtained from three different networks can be merged to represent the predicted log for a complete well. It is envisaged that horizon or well tops based prediction of reservoir characteristics has enhanced the prediction accuracy and hence modular based neural networks can be applied to characterizes reservoir parameters where input are seismic attributes.