Inversion based on deep learning of logging-while-drilling directional resistivity measurements

Abstract Because of the response complexity and real-time interpretation required by geosteering, an efficient and reliable inversion method for logging-while-drilling (LWD) directional resistivity measurement is important. A three-layer parametric inversion method based on deep learning has been developed. The inversion process includes two steps. The first step is the inversion of formation parameters, such as the resistivity and distance to boundary, and the second step is the uncertainty estimation of the inversed parameters, to remove the low-quality inversed boundaries. The input of the first part of the inversion is the directional resistivity logging data. The input of the second step is the output of the first step. The second step is based on the relationship between the output error and the value of each output parameter of the first step. A synthetic example shows the original results of the first step of the inversion have some low-quality false boundaries, such as some of the ones in thick layers. By estimating the uncertainty and distinguishing the low-quality boundary in the second step, these false boundaries can be effectively removed.