Modelling a fluvial reservoir with multipoint statistics and principal components

Traditional reservoir modelling techniques use oversimplified two-point statistics to represent geological phenomena which are typically curvilinear and have other complex geometrical configurations. Use of multipoint statistics has shown some improvement in recent years to reduce such limitations. This paper compares the performance of the use of conventional and multipoint data for estimating porosity from seismic attributes in a fluvial reservoir using neural networks. According to the results of the study, the neural network trained on multipoint data gave smaller error and higher correlation coefficient of porosity in a blind test. Further improvement is also obtained by reducing the dimensionality of the input space using principal components. This study shows a successful integration of neural networks and principal components for modelling multipoint data in practical reservoir studies.