Application Of Smooth Neural Networks For Inter-Well Estimation Of Porosity From Seismic Data

Abstract

We apply an approach based on smooth neural networks to a 3D seismic survey in the Shedgum area of the Ghawar Field to estimate the reservoir porosity distribution of the Arab-D Member. We conducted numerous systematic cross-validation tests to assess the accuracy of the method and to compare it to that of traditional back-propagation networks. The results obtained from these tests indicate that the regularized back-propagation network can be quite adept at estimating the porosity distribution of the reservoir in the inter-well regions from seismic data. The accuracy remained consistent as the network parameters (size and training length) were varied. On the other hand, the traditional back-propagation network gave acceptable results only when the optimal network parameters were used, and the accuracy deteriorated significantly as soon as deviations from these optimal parameters occurred. Moreover, utilizing smooth networks, the final porosity volume corroborates our existing understanding of the reservoir and shows substantial similarity to the simple geologic model constructed by interpolating the well information, while adding significant detail and enhanced resolution to that model. We also scrutinize multi-attribute analysis, analyze how attributes can be both constructive and damaging to the prediction of the reservoir properties, and evaluate their effectiveness in enhancing the accuracy of the solution.