Application of neural networks to identify lithofacies from well logs

Neural networks are physical cellular systems that can acquire, store, and utilise knowledge obtained by experience. They have a number of performance characteristics in common with biological neural networks or the human brain. They can simulate the nervous systems of living animals that work differently from conventional computing and analyse, compute and solve complex practical problems using computers. Studies show that neural networks can be used to solve a great number of practical problems that occur in modelling, predictions, assessments, recognition, and image processing. Especially, neural networks are suitable for application to problems where some results are known but the manner in which these results can be achieved is not known or is difficult to implement; or where the results themselves are not known. Lithofacies are, by definition, determined directly from the rock body, either an outcrop or conventional core. However, in subsurface studies it is important to be able to estimate the lithofacies of uncored wells or sections of wells. This is commonly done in a somewhat subjective manner using a wireline-log suite, and often using only the gamma-ray log. This paper presents an application of neural networking to identify lithofacies from wireline logs in a lithologically complex formation. The methodology provides a considerably more thorough, consistent, and less subjective means by which to estimate lithofacies, and indeed depositional facies, than is currently in use. Only conventional wireline-logs were available in this study. Nonetheless, it was possible to establish an identification network of lithofacies by integrating error back propagation and self-organising mapping algorithms. This network was applied to identify lithofacies in six wells. The results indicate that, given a reliable training set which is provided from a detailed lithofacies analysis of conventional cores within the formation under study, it is possible to estimate lithofacies from conventional well logs by means of neural network techniques.