An efficient forward modeling method of electromagnetic response of multiscale hydraulic fracture based on deep learning

Although deep learning has been applied in various fields, its usage in solving electromagnetic (EM) fields is limited. In this letter, we propose a novelty scheme to simulate the EM response of hydraulic fractures rapidly based on Long Short-Term Memory (LSTM). The calculation speed of the traditional finite element method (FEM) is extremely slow due to the multiscale characteristic of fracture formation. It is difficult to provide sufficient samples for training of the deep learning. To improve the calculation efficiency, we first employed a combination of transitional boundary conditions (TBC) and FEM to establish the dataset. Then, it is divided into two parts for training and testing the deep learning model. Finally, the model is applied to calculate the EM response of the hydraulic fracture formations. The results indicate that the method has a high accuracy and strong applicability to predict the EM response of hydraulic fractures. Furthermore, we utilize transfer-learning(T-L) technology to improve the generalization of the method. This enables us to quickly model the EM response of hydraulic fractures at various frequencies and spaces of the tool.