Artificial neural network modelling as an aid to source rock characterization

Abstract

A data-based approach is used to establish detailed and accurate geochemical characterization in oil source rock intervals, using well logs and noisy information from cuttings. The method starts by extracting examples from the intervals of interest, using generalized relationships between total organic content (TOC) and well log responses. The examples were used to train an artificial neural network (ANN) that extracted more detailed and accurate relationships between TOC and well-log responses specific to the study area. A combination of the ‘quickprop’ algorithm and ‘Dynamic Node Creation’ scheme was utilized to facilitate efficient training. The trained ANN is useful for mapping source rock intervals in the area of interest. This method performs satisfactorily when applied to the Egret source rock of the Jeanne d'Arc Basin, offshore eastern Canada.