Gradient Boosting Decision Tree-Based PMM Model Integrated Into FDTD Method for Solving Subsurface Sensing Problems

It is known that both the physical domain in size and problem complexity should also be considered to decrease more and more computational resources, especially for a large-scale complicated problem, besides taking into account numerical accuracy and efficiency for extremely low-frequency subsurface sensing application in the airborne transient electromagnetics (ATEMs). For better satisfying with the above conditions, an alternative perfectly matched monolayer (PMM) model is proposed based on the gradient boosting decision tree (GBDT) for the finite-difference time-domain (FDTD) implementation, which is incorporated to further enhance the performance. In the GBDT-based PMM model, it is not only higher accuracy that could be obtained through the ensemble learning method of feature attention but also less memory and time consumption that are required. In terms of model training stability and its lightweight, moreover, the proposed model has significant merits because of its dependence on the characteristics of traditional machine learning (ML) models. Eventually, numerical cases of low-frequency ATEM applications are considered to validate the performance of the proposed algorithm. It is observed from the results that we can not only gain advantages in accuracy, efficiency, and problem complexity but also succeed in integrating this model into the FDTD solver.