Real time quasi–2-D inversion of array resistivity logging data using neural network

Abstract

We present a quasi-2-D real-time inversion algorithm for a modern galvanic array tool via dimensional reduction and neural network simulation. Using reciprocity and superposition, we apply a numerical focusing technique to the unfocused data. The numerically focused data are much less subject to 2-D and layering effects and can be approximated as from a cylindrical 1-D earth. We then perform 1-D inversion on the focused data to provide approximate information about the 2-D resistivity structure. A neural network is used to perform forward modeling in the 1-D inversion, which is several hundred times faster than conventional numerical forward solutions. Testing our inversion algorithm on both synthetic and field data shows that this fast inversion algorithm is useful for providing formation resistivity information at a well site.