It is well known that the inelastic count rates from Pulsed Neutron Capture (PNC tools are dependent upon formation bulk density and hydrogen index. Recently a method has been developed from Monte-Carlo Modeling which utilizes the PNC capture count rates to correct the inelastic count rate ratio for hydrogen index effects. The corrected inelastic count rate ratio, RINC, is then calibrated to predict bulk density by applying regression techniques. The effects of borehole size and salinity are also incorporated into the regression parameters. We call this approach the RINC method.
An alternative technique also recently developed is to obtain a transform of the PNC measurements directly into formation density by applying a neural network ensemble (NNE. In this case, a well (the training well having both a density log and a PNC log must be available. The effects of the borehole environment are implicitly accounted for by some of the PNC curves input to the neural network during the training process. Predictions of formation density can then be obtained on neighboring wells (the application wells by normalization of the input logs from the application well to the training well and applying the neural network to the normalized logs. We call this approach the NNE method.
This paper compares the two methods applied to actual field examples. The effects of variations in the formation density, hydrogen index, borehole size, borehole fluid, and formation fluid are investigated. Also investigated are the number and type of PNC curves (including the corrected inelastic count rate ratio RINC input to the neural network ensemble. In addition, results from the Monte Carlo modeling and an overview of the RINC method will also be presented.
It is expected that prediction error in either case will be related to the borehole size (caliper. Thus the prediction of the open hole caliper curve from PNC measurements with the NNE will also be discussed and evaluated.