Machine learning modelling in service of geophysics analysis: estimation of oil industry parameters

Abstract

CSEM (Controlled Source Electromagnetics) has been used in parallel with seismic surveys for the past decade by the oil industry and is a development of older geophysical techniques that aim at the "mapping" of electrical properties, especially electrical resistance, of the subsoil. This thesis aims to study statistical well log data analysis methods, to accelerate the solution of the inverse problem which is required during CSEM data processing. The inverse problem determines the distribution of the electrical resistance of the subsoil, which would generate signals equal to the actual received signals at the receivers. Solving the inverse problem is a computationally demanding and time-consuming process (it takes several hours to days for a complete solution by computing clusters). In particular, the possibility of providing reliable estimations of the ground electrical resistivity using sonic well-log data will be studied. These estimations will be used as initial conditions in the process of solving the inverse problem which, in turn, can significantly reduce the duration of the inversion while increasing its accuracy.