Abstract

A novel approach based on the concept of Bayesian neural network learning theory is developed and applied to German Continental Deep Drilling Program (KTB) well log signal for classification of lithofacies boundaries. We parameterized different combination of synthetic model to match with the published histogram lithofacies succession. A Multi Layer Perceptron (MLP) network model (with four layers e.g. input, output and two hidden layers) is found suitable for the present pattern recognition problem. Objective function is optimized following scaled conjugate gradient optimization technique and uncertainty about the relationship between input and output domain is appropriately taken care of by assuming Gaussian prior distribution of networks parameters. Posterior distribution of network parameter is estimated following the Bayesian probability theory. The stability and effectiveness of the new approach is also tested on