THE APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN LOGGING INTERPRETATION

Artificial neural network (ANN) theory has been widely applied in several aspects of logging interpretation in the petroleum industry, such as lithofacies identification, logging parameter prediction as well as other model recognition problems. Among several neural network models and algorithms applied in petroleum industry, the feed-forward neural network model and back-propagation algorithm (BPA) are most widely used for their easy implementation and high robustness. Although the BPA shows strong capability on solving model recognition, it has several problems, such as local optimum and low rate of convergence, when dealing with parameter prediction.

After comparing several kinds of neural network models and algorithms, the self-organization neural network model and Kohonen optimal algorithm are chosen to solve the lithofacies identification problem in the CaiNan oilfield in China. The feed-forward neural network model and Levenberg-Marquardt (L-M) optimal algorithm are selected to predict logging parameters, such as porosity, permeability and water saturation, in the same oilfield. Based on Matlab tools, an artificial neural network logging interpretation program is compiled and applied to the logging evaluation of this oilfield. Simulation results show that the relative prediction error of porosity, permeability and water saturation can respectively reach 6%, 30% and 30% which is higher than normal methods for parameter prediction. The results prove that the self-organization neural network model and Kohonen optimal algorithm are suitable for lithofacies identification and that the feed-forward neural network model and the L-M optimal algorithm are suitable for logging parameter prediction in the oilfield.