A novel inversion method of logging while drilling azimuthal electromagnetic data in anisotropic formations based on BiLSTM

Abstract With the widespread application of the logging-while-drilling (LWD) azimuthal electromagnetic (EM) tool, the inversion of formation resistivity and boundaries has become a significant concern. However, conventional inversion methods face practical challenges, as they are often time-consuming, nonlinear, and ill-posed. To address these challenges, we designed a deep learning model based on a Bidirectional Long Short-Term Memory (BiLSTM) network to invert LWD azimuthal EM data in anisotropic formations. Initially, an anisotropic formation model of horizontal stratigraphy was established, with formation parameters (such as resistivity) assigned random values. A fast and efficient analytical method was then employed to calculate the logging response. These steps were repeated to generate a substantial number of samples. Subsequently, each sample was divided into two segments—deviate and horizontal—based on the inclination angle of the tool during drilling, resulting in two distinct sample sets. The BiLSTM network with varying hyperparameters was then trained and tested using these two sample sets, resulting in the development of two deep learning inversion models. Finally, the inversion performance of the two inversion models was analyzed. The experimental results demonstrated that the two inversion models could not only accurately invert the formation resistivity and the positions of layers boundaries, but also exhibited a rapid inversion speed, with a single-point inversion time of only 0.4 ms. This high inversion performance is crucial for reservoir detection, precise instrument targeting, and effective drilling within the reservoir. Moreover, we showcased the robustness of the inversion method by artificially introducing noise into the logging data. These results underscore the considerable potential of the intelligent inversion approach for LWD azimuthal EM inversion applications.