Research on Intelligent Generation Method of Logging Curve Based on Deep Learning

In well logging evaluation, there are often cases where some well logging curves are distorted or missing, and other known information needs to be used to generate the required well logging curves. The traditional method of multiple regression fitting to generate curves has low accuracy. Based on the structural characteristics of logging data, the logging data is processed into layer samples, and four suitable deep learning algorithms, including three types of recurrent neural networks and onedimensional convolutional neural networks, are applied to intelligently generate logging curves in public datasets and compare their effects. The one-dimensional convolutional neural network model produces the highest accuracy of the curves, but there are still some segments that differ significantly from the measured curves. To further improve accuracy, a sample set was established under similar geological background conditions. Taking C8-1-2 of the XH105 field in the southern Ordos Basin as an example, a one-dimensional convolutional neural network model was used to intelligently generate acoustic time difference curves from other curves. Blind well testing showed good consistency between the generated curves and the measured curves, confirming that the intelligent curve generation method can achieve better results under similar geological backgrounds.