ScholarMate
WeChat
Sina
Facebook
Share link
Summary
Due to the Induced Polarization (IP) effect, the sign reversal occurs frequently in the late-time channels of Airborne Electromagnetic (AEM) signal for central loop configuration. Since the EM signal is related to multiple IP parameters, and the sensitivity of each parameter varies, serious non-uniqueness can be observed when inverting the data. In this paper, we present an AEM inversion scheme in time-domain for IP parameters based on Pearson correlation constraint and deep -learning algorithm. The inversion first predicts the IP parameters in time-domain based on deep learning. After that, it gives a small range of constraints on the time constant and frequency exponent and then inverts the resistivity and chargeability. This can largely reduce the uniqueness of the solution. For the inversion of resistivity and chargeability, we use the Pearson correlation coefficients in statistics to construct the constraints of these two physical parameters, so that we can further reduce the non-uniqueness of solutions. To verify the effectiveness of our inversion scheme, we carry out the experiments on the synthetic models of double prisms or an arch. It is shown that the inversion results of resistivity and chargeability based on the Pearson correlation constraint are closer to the true model than the traditional Gaussian Newton method, the inverted resistivity and chargeability based on the predicted time constant and frequency exponent by deep learning are equivalent to those when their true values are given in the inversions. Finally, we invert an AEM survey dataset from Australia with and without IP effect, respectively. The results show that the data fitting and the continuity of geoelectrical section are both largely improved when the IP effect is considered.
-
Institution桂林理工大学; 吉林大学; 中国科学院
