Predicting Vertical Resistivity By Machine Learning

Recently we presented results of compilation and formation-level up-scaling of all triaxial electric resistivity data from exploration wells drilled at the Norwegian continental shelf available in DISKOS database. The purpose for that work was to analyze the formation electrical anisotropy using the results of direct measurements. The triaxial database was assembled and analysis demonstrated that subsurface is, and should be treated as mostly anisotropic, as far as the direct measurements are available. In particular, the median formation anisotropy was found to be around 2.5, and no formations were found to have anisotropy less than 1.25. In this work we extend the study by applying machine learning (ML), based (trained and tested) on the existing triaxial database. The triaxial database also got extended by recently published wells. We discuss some details of the ML-model. We confirm previously obtained results, strengthen the statistics, and make further observations. The anisotropy, as well as the vertical resistivity are surveyed, in particular, in controlled source electromagnetic (CSEM) imaging, and anomalies in relevant distributions commonly serve as direct hydrocarbon indicators. Thus, we also show comparison of the ML-prediction to the CSEM inversion results available to us. We further discuss the potential application of this ML-predictor.