Method for deep learning inversion of transient electromagnetism based on fassa-CNN-lstm

Abstract

Provided is a FASSA-CNN-LSTM based transient electromagnetic deep learning inversion method, including: on the basis of survey data of a survey area, determining underground resistivity and a formation thickness range, building a layered model, and generating transient electromagnetic response data (sampling time-decay voltage) corresponding to the layered model through forward modeling; building a FASSA-CNN-LSTM deep learning model, performing model training with the electromagnetic response data as input data, and dynamically adjusting a learning rate via an optimizer to finally obtain a corresponding depth-resistivity output data; and inverting the collected true transient electromagnetic response data by using the trained FASSA-CNN-LSTM inversion model to obtain a true depth-resistivity geoelectric model. Compared with conventional time series deep learning inversion models, such as LSTM and GRU, the FASSA-CNN-LSTM inversion model can more effectively retain important information in sequences and reduce the time required for training when processing long transient electromagnetic sequences, thereby improving the accuracy of transient electromagnetic inversion and the efficiency of model training.

Images