Exploring AI Models for Forward Logging While Drilling from MLP to LSTM-Transformer Networks

Computational Electromagnetics (CEM) plays a crucial role in modeling electromagnetic wave interactions in subsurface environments, particularly in Logging While Drilling (LWD) for oil and gas exploration. Traditional numerical methods, such as the coefficient propagator approach, are highly accurate and efficient. However, their limited parallelization capabilities pose challenges in leveraging modern parallel computing architectures, making it difficult to efficiently handle large-scale datasets and complex inverse problems. This paper explores the application of deep learning models, including MLP, U-Net, and LSTM-Transformer, to forward modeling of LWD responses. Using a synthetic dataset generated via a Fortran-based electromagnetic simulation tool, we demonstrate that the LSTM-Transformer model achieves highest accuracy and computational efficiency, offering a scalable alternative for forward modeling in LWD.