A Physics-Driven Deep-Learning Inverse Solver for Subsurface Sensing

Solving inverse problems accurately and efficiently has always been an important issue in subsurface sensing. Pure data-driven machine learning methods have achieved great success in the past few years, but these methods still face questions about reliability. At the same time, extremely massive data without any physical guidance may lead to missing opportunities for breakthroughs. In this paper, we propose a physics-driven deep learning framework for providing a fast and accurate surrogate to solve non-linear inverse problems. Particularly, leveraged by the forward physical model and 1D Convolutional Neural Network (CNN), the proposed method provides more reliable solutions to the inverse problem with improved performance. Applications for magnetotelluric data inversion demonstrate the effectiveness of our method.