A Method for Correction of Shoulder-Bed Effect on Resistivity Logs Based on a Convolutional Neural Network

Abstract

Abstract —Shoulder beds may have a significant effect on the resistivity log responses. This problem is especially acute in studies of complex strata composed of thin beds with contrasting properties. Different approaches to taking account of the shoulder-bed effect on logging signals are known, such as correction charts, deconvolution operations, and using advanced algorithms of numerical data inversion, which allow one to consider the vertical inhomogeneity of the section. The best result is achieved using the inversion toolkit, but the high labor- and resource-intensiveness of the approach limits its widespread use. The deconvolution approach does not have these disadvantages, but it does not take into account the influence of radial changes in the medium properties on the shapes of measured signals.The possibility of using artificial neural networks (ANN) to increase the vertical resolution of the measured logging data is explored. We assume the existence of a deconvolution-like transformation in which change in the medium properties in the radial direction is also considered. In this case, we can find its approximation using a neural network. The approach is demonstrated by creating a transformation algorithm for the high-frequency electromagnetic logging (VIKIZ) sounding tool, which is widely used in the CIS countries for petroleum exploration. The developed algorithm has been tested on the VIKIZ logs from the Fedorovskoe oilfield (Latitudinal Ob’ region).