Prediction of the electromagnetic responses of geological bodies based on a temporal convolutional network model

The transient electromagnetic method employed in aeromagnetic surveys has been widely used for geophysical, petroleum, and engineering exploration because geophysical characteristics can be predicted as an inversion problem based on measured electromagnetic response data. However, this process requires uniformly and densely distributed electromagnetic response data, which are typically unavailable in actual TEM applications due to the high cost of the aeromagnetic surveys, which necessitates the use of large grid patterns to effectively map large areas. Therefore, developing methods for predicting missing electromagnetic response data based on the available data is essential for ensuring the accurate characterization of geological bodies. The present work addresses this issue by establishing an electromagnetic response curve prediction model based on a temporal convolutional network (TCN) architecture. Firstly, the electromagnetic response data is subjected to grey relational analysis to obtain correlations and reduce the data dimension. Secondly, the response data with correlation degrees greater than a threshold are selected as TCN model input. Finally, the TCN model establishes the nonlinear relationship between the electromagnetic response parameter sequence and its output sequence. The proposed model and other existing state-of-the-art prediction models are applied to actual electromagnetic prospecting data, and the results demonstrate that the proposed TCN model provides higher prediction accuracy and stronger robustness than the other models considered. Moreover, the proposed model is suitable for processing multiple series of related data, such as electromagnetic response prediction models. Therefore, the proposed model has good application prospects in electromagnetic response prediction and electromagnetic response recovery research.