Research On Pulsed Neutron Gamma Density Logging Method

Formation bulk density is an important parameter for formation evaluation and plays an important role in calculating porosity and identifying gas-bearing layers.Current nuclear logging methods to obtain bulk density include gamma-gamma density logging（GGD）with a ¹³⁷Cs chemical source and neutron-gamma density logging（NGD）with a pulsed neutron source.Compared with the traditional GGD,NGD is safer and more environmentally friendly,and has gradually become a substitute for GGD and a future development trend.Both GGD and NGD obtain the formation density based on the attenuation law of gamma rays,but NGD uses secondary gamma rays produced by inelastic scattering of fast neutrons with the formation.On the one hand,the initial energy of the inelastic gamma is at the Me V level,leading to the effect of the electron pair effect cannot be ignored,and the initial energy,source intensity,generation location and spatial distribution of the secondary inelastic gamma source also vary with the formation,making the influence factors of NGD much more complicated than GGD,which in turn makes the density prediction accuracy of NGD lower than that of GGD.On the other hand,the influence of borehole factors on NGD is significant,which restricts the popularization and application of NGD.Therefore,it is necessary to carry out research on methods to correct the borehole influence factors and improve the accuracy of density prediction in NGD.In this paper,the theoretical equation between bulk density,inelastic gamma flux and fast neutron flux is derived based on the simple diffusion theory,and the intrinsic relationship between them is investigated.The conventional density calculation models are optimized to obtain the optimal polynomial model（OPM）,and then the conventional borehole correction method based on the OPM model is investigated.Since the mathematical nature of the density calculation problem in NGD is a regression problem,a machine learning regression algorithm is introduced instead of the conventional density calculation model to improve the density prediction accuracy.In addition,because the machine learning regression algorithm can directly input borehole parameters when building the density prediction model,a new method of using machine learning regression algorithm to achieve borehole correction is proposed.In addition to the introduction of machine learning algorithms with better learning ability,we also start from reducing the influence of interference factors to further improve the density prediction accuracy.To this end,eight fundamental problems in the study of NGD influence factors are analyzed in depth,the misconceptions in the current study of electron pair effect influence are clarified,the nature of electron pair effect influence is elucidated,and the integrated use of the inelastic gamma counting ratio,fast neutron counting ratio and gamma energy spectrum is proposed to reduce the influence of interference factors such as electron pair effect.And on this basis,a new method of pulsed neutron density（NGSD）based on gamma energy spectrum and machine learning regression algorithm is proposed,and the intrinsic relationship between elemental content and bulk density is deeply analyzed from the perspective of MCNP matter card production,which lays the theoretical foundation of NGSD.It is shown that the introduction of the machine learning regression algorithm significantly reduces the density prediction error and can achieve density prediction and borehole correction in one step,which greatly simplifies the borehole correction process and avoids the problems such as accuracy loss in the conventional borehole correction method.In the absence of instrument offset information,the machine learning algorithm can also achieve good results,greatly expanding the application scenario.In addition,the introduction of gamma energy spectrum reduces the density prediction error to the level of 0.01,which greatly improves the density prediction accuracy.The new method NGSD can achieve the lowest training set error and test set error of（0.005,0.009）on the simple borehole dataset and（0.009,0.013）on the complex borehole dataset,which is much lower than that of（0.035,0.034）for the conventional NGD method OPM on the simple borehole dataset.When the instrument uses only two gamma detectors,the best performance of NGSD on the simple and complex borehole datasets reach（0.000,0.016）and（0.011,0.017）,respectively;even when the instrument uses only one gamma detector,the best performance of NGSD on the simple and complex borehole dataset reach（0.001,0.018）and（0.013,0.022）,both much lower than the conventional NGD method using four detectors.This means that NGSD can use two or even one gamma-ray detector to achieve the high-precision formation density measurement,simplifying the instrument structure,saving instrument development costs,and significantly improving the formation density prediction accuracy,while significantly improving the resolution in the axis direction of the instrument.