Abstract
Formation bulk density is vital for reservoir evaluation in oil and gas geophysical exploration. The pulsed neutron-gamma density (NGD) logging method, utilizing a pulsed neutron source, provides a safer and eco-friendlier option for formation density measurement compared to conventional gamma-gamma density (GGD) logging, which employs a (137)Cs chemical gamma-ray source. However, the accuracy of NGD is compromised by complex interferences like pair production and fast neutron transport, with borehole conditions further influencing its performance. Using intuitive path diagrams instead of complex mathematical derivations, we thoroughly analyze various interference factors and find that the effects of pair production represent merely a minor aspect of the broader impacts resulting from changes in the formation's chemical composition. Like using fast neutron terms to characterize neutron transport's influence on inelastic gamma-ray production, we integrate gamma-ray spectra, critical indicators of formation chemistry, into our density calculation model to account for pair production and other interferences. We apply advanced machine learning regression algorithms to handle the increased input features and mitigate traditional borehole correction methods' accuracy loss and complexity. Integrating gamma-ray spectra with machine-learning regressors significantly improves density prediction accuracy, reducing root-mean-square errors from over 0.03 g/cm(3) to below 0.01 g/cm(3) in both training and test sets. This method outperforms the conventional four-detector NGD method even with a single gamma-ray detector, saving costs, enhancing resolution, and demonstrating considerable practical potential. Moreover, machine learning enables density prediction and borehole correction in a single step, streamlining the workflow, reducing accuracy loss, and achieving impressive results even without tool standoff information, broadening its applicability.