Abstract
Artificial neural network models are increasingly prevalent in predicting various reservoir engineering parameters. The petroleum engineering community at large is witnessing a surge in the use of expert systems and artificial neural networks related technologies, within the broader domain of artificial intelligence. These relatively new computer based tools provide answers to a number of engineering problems which are usually expensive to quantify experimentally, and at the same time reduce the uncertainties inherent in many of the measurement techniques.
Kalam and Al-Alawi1-4 have successfully applied artificial neural network (ANN) derived models to assess formation damage in limestone formations, and prediction of wettabilities of cores from diverse producing carbonate reservoirs in the Sultanate of Oman. The ANN predictions were then compared with measured data, and a number of established empirically derived correlations (commonly used in typical reservoir simulation studies).
The developed ANN models provide quick and fairly accurate predictions of the end-point relative permeabilities of reservoir cores from naturally fractured heterogeneous reservoirs, using simple log derived input parameters. Errors in estimating in these important reservoir characterisation properties from a number of widely used correlations are significantly higher than the experimental measurements. The relative permeability curves from ANN models predictions are consistent with those determined from the tedious and rather expensive SCAL programmes, and as such reduce the uncertainties in reservoir simulations.
