Reduction of uncertainties in neural network prediction of oil well logs

The ability to accurately predict the location of remaining oil in the neighborhood of existing production wells is of vital economic importance to the petroleum industry. A methodology to determine the confidence limits of results obtained by neural network models is formulated. This methodology consistently combines experimental data (e.g., sensor measurements) with model-predicted results. Best estimates for the network model parameters are obtained, and uncertainties underlying modeling processes based on learning are reduced. Preliminary results show the promise of this methodology for petroleum reservoir characterization.