Supervised decent learning for LWD inverse problems

Figure Caption

Inversion results of the five-layer continuous model from different methods. (a) True model. (b) Reconstructed model from SDM. (c) Reconstructed model from Occam’s inversion.

Research Summary

We have studied the application of an efficient and robust supervised descent method (SDM) for solving directional electromagnetic LWD inverse problems. Further improvement is also explored according to the characteristics of the LWD and geosteering applications. Considering the essence of geophysical inversion is to reconstruct the subsurface structures with not only the measurements but also the geological/geophysical and prior knowledge; the SDM combines the machine learning scheme and the advantages of the conventional methods.