In-situ recovery of heavy oil remains a significant challenge due to the high viscosity of heavy crude, which impedes both flow and extraction efficiency. Existing industrial practices predominantly rely on thermal recovery techniques such as steam injection, which are energy-intensive and environmentally unsustainable. Electromagnetic heating presents a promising alternative, offering advantages in efficiency, controllability, and environmental sustainability.

This project focuses on inverse modeling of the electromagnetic heating process to facilitate real-time monitoring, adaptive control, and predictive optimization of in-situ heavy oil recovery. To support this goal, machine learning will be incorporated as an auxiliary module within conventional full-wave inversion workflows, combining the computational speed of machine learning with the physical consistency of deterministic solvers. A bounded inversion strategy based on L-BFGS-B will be employed, in which physics-informed constraints derived from machine learning models are embedded into the optimization process. In addition, GPU-accelerated operator-level designs will be investigated to improve computational throughput and facilitate large-scale, real-time deployment.