This case study compares core photographs, core analysis, a geologist’s facies analysis, sequence stratigraphie analysis, and a description of cores to facies identified by a Simulated Neural Network (SNN). The SNN operates by mimicking human-like reasoning-under-uncertainty from incomplete and partial inputs. The facies identified by the neural network matched those in the geologist’s facies description. The well from which these data were taken penetrates intertidal, lower shoreface, and shelf-to-lower shoreface sequences containing sandstones, siltstones, shales, mudstones, coal, and minor sandy-dolomite beds. Siltstones and mudstones are at times highly interbedded, burrowed, calcareous, and contain organic matter. The log database’s 1/4-foot depth increment currently limits the neural network’s resolution. However, the network’s resolution will be enhanced if dipmeter data is also input. In this case study, the network rapidly and accurately identified 11 lithofacies in the test interval.
Operation of this neural network does not require knowledge of geology or experience as a geologist, although assigning a name to the facies identified does require knowledge of the local geology. For this case study, a petrophysicist unskilled in geology performed all parameter selections.
Log analysts routinely adjust their log correlations and descriptions to incorporate the fact that log data are sometimes missing. In this study, acoustic shear waves could not be measured in facies composed of very soft rock. The neural network’s method of operation automatically accounted for the fact that shear data are missing across some intervals.
The SNN program used for this case study, as implemented on an affordable, highly accessible PC platform (but not restricted thereto), is suitable for depositional environment studies, facies identification, sedimentation studies, and well-to-well facies correlation. Additional applications include identification of productive intervals, and fracture identification.