We develop an attitude correction for the marine controlled-source electromagnetic method transmitter (ACMT), a deep-learning algorithm based on transformer modeling and modal fusion. This model uses motion sensor data to correct attitude effects in towed marine controlled-source electromagnetic method (MCSEM) transmitters. To train ACMT, we compute forward responses for 2000 combinations of randomly modeled reservoirs and transmitter attitudes, using the corresponding attitude-free forward responses as training labels. ACMT demonstrates robust performance on the test set, significantly outperforming the wavelet analysis method and fully connected network algorithm in terms of the mean-square error (MSE). This superior performance is evidenced by the attitude-corrected data, which indicated a reduction in MSE of more than 98% compared with data with uncorrected attitudes, closely aligning with theoretical predictions. We also apply ACMT to MCSEM field data, wherein the apparent resistivity pseudosection from the corrected data more closely matches the actual subsurface conditions than the precorrection data. Experimental results indicate that ACMT can effectively capture and correct the nonlinear relationship between motion attitude and electric field variations, with the corrected data discriminating the electrical structures better than the uncorrected data.

You do not have access to this content, please speak to your institutional administrator if you feel you should have access.