At Dagger Draw it is difficult to arrive at reliable estimates of total pore volume in the vuggy dolomitic reservoir. Correlating wire-line logs to core porosity using a multi-layer perceptron neural network (MLP) generated a "ground truth" porosity estimator for the reservoir. Well logs in the field include CAL, DRHO, GR, LLD, LLS, MSFL, NPHI, PEF, RHOB, and values of DPHI, PHI-L (liquid phase porosity) and PHI-G (gas phase porosity) were calculated. Log values were aligned with the center point of laboratory analyzed core samples in five cored wells with a total of 21 cored sections. Neural network solutions generated using four of the five wells to blindly predict the fifth well validated results. Each well was excluded in this fashion, for a total of five tests.
The resulting correlation tool, if applied at hundreds of wells within the field without core data, would provide the basis for geostatistically derived pore volume maps.
Fuzzy ranking was used to evaluate the usefulness of individual logs for correlating with core porosity, and also indicated that the problem was complex, and guided architecture design for the MLP. The seven best logs were used as inputs for the MLP, and results show that the MLP can describe the complex relationships between the wire-line data and core porosity. Regression relationship were able to predict the excluded wells at up to CC=0.80. These results are a great improvement on the linear equations previously used for estimating porosity at the Dagger Draw and better predictions are expected, as the process is refined and neural network algorithms are improved.
The use of artificial intelligence to generate core porosity data using only wire-line log data can provide data that could only be attained previously using expensive cores.