Oil Reservoir Porosity Prediction Using a Neural Network Ensemble Approach

Abstract

The problem of parameter prediction in a hydrocarbon reservoir is typically accomplished by an interpreter using sparse well information and seismic data. The resulting maps may contain varying levels of uncertainty depending on the experience of the interpreter and the availability and quality of seismic and well data.

This chapter describes a neural network ensemble approach to the interpolation problem. An interval of seismic data representing the zone of interest is extracted from a three-dimensional (3-D) data volume. Combinations of seismic data, complex trace attributes, and geometric data are used as inputs to a multilayer perceptron (MLP) neural network. A series of networks is trained, and results are used to produce final ensembles.

Our results show that realistic prediction maps can be generated using a neural network ensemble approach with input data consisting of raw seismic amplitudes, instantaneous amplitude, and coordinate geometry. Ensemble maps created for subintervals in the zone of interest show consistent features. Also, network testing errors at finer levels of resolution are not unusually large compared to errors for a single layer prediction.