Neural network applications of seismic attributes for predicting porosity and production and mapping fault zones

Seismic attribute analysis using neural network applications has become an important tool for understanding the physical features and internal structure of reservoirs. Understanding such features is vital to providing effective strategies for exploration and exploitation in both new and old survey areas.;The primary objective of this thesis is neural network applications of multi-attribute analysis to reveal reservoir features. Two neural network approaches were used for attribute analysis. First, unsupervised network training was applied to geometric attributes in order to map possible fault and fracture zones in the Parkman Sandstone Member of the Upper Cretaceous Mesaverde Formation. Second, supervised neural network analysis was applied to instantaneous attributes derived from pre-stack migrated seismic data complemented by well logs.;The data set used in this study is proprietary and was collected in eastern Wyoming by WesternGeco. The data set available for this study comprises 3D seismic data, sets of well logs, and base maps. The data set became available to Montana Tech in 2011 under a license agreement.;The outcome of this thesis is an enhanced subsurface display of the geology in order to identify reservoir properties. Fault identification and porosity and production predictions were goals for the target zones in order to grasp a better understanding for the reservoir facies. Unsupervised network training revealed no major fault zones that could potentially jeopardize horizontal well drilling in the Parkman Member. Furthermore, predicted porosity and production maps for the Parkman Member obtained from a supervised neural network analysis showed that production wells were penetrating medium and high porosity sections. Moreover, the predicted porosity map for the Sussex Member revealed that two available production wells were penetrating into high porosity sections in the Sussex.