Fluid typing in tight sandstone from wireline logs using classification committee machine

Abstract Owing to the low porosity and low permeability of tight sandstone reservoirs, the geophysical log responses to different fluids are unclear, and it is difficult to identify the fluid types of some layers from wireline logs. The committee machine is a recently developed hybrid intelligent algorithm that combines expert networks and makes the final judgment through a decision mechanism. Fluid identification by log interpretation is a classification problem in machine learning; thus, we construct a classification committee machine (CCM). It is composed of a back-propagation neural network, a probabilistic neural network, and a decision tree classifier, and the final result is decided by voting. The processing flow of fluid identification from wireline logs is summarized, and the expert networks are optimized. Fluid typing for a gas reservoir and an oil reservoir in the Ordos Basin is conducted and analyzed in detail, and logs that are sensitive to the fluids in layers are chosen as input for the CCM. The prediction results from the CCM are compared with those of the three expert networks. Case studies from the gas field and oilfield show that the CCM can effectively combine intelligent algorithms through a decision-making mechanism and provide more accurate results.