摘要:
瞬变电磁法(transient electromagnetic method,TEM)目前常用的解释方法是采用全区视电阻率参数,其涉及的公式复杂,迭代过程繁琐耗时。本文分析TEM数据特征,引入人工神经网络(ANN),实现了瞬变电磁拟电阻率成像。设计多隐层BP神经网络,利用瞬变电磁解析方法计算出响应幅值,作为拟电阻率的映射参数参与网络训练,再使用训练集外的新数据来测试训练后的网络。构建了均匀半空间、一维层状模型,验证神经网络的正确性和适应性,对三维地电模型进行了成像。结果表明:神经网络计算的拟电阻率能够反映出地电模型的目标体异常,网络成像结果准确度较高。最后,利用神经网络算法处理实测数据,进一步说明神经网络成像可以作为资料解释的依据。该研究证明了人工神经网络在瞬变电磁成像上的可行性,为瞬变电磁成像提供了一种新的思路。
Abstract:
The transient electromagnetic method (TEM) commonly uses the all-time apparent resistivity parameter for interpretation, which involves complex formulas and time-consuming iterative processes. Based on the characteristics of TEM data, this study employed the artificial neural network (ANN) for TEM pseudo-resistivity imaging. First, this study designed a multi-hidden-layer BP neural network and calculated a response amplitude through TEM analysis. The response amplitude, as the mapping parameter of pseudo resistivity, was used for network training. Then new data outside the training set were used to test the trained network. A homogeneous half-space and one-dimensional layered model was built to verify the correctness and adaptability of the neural network. The imaging of the three-dimensional geoelectric model was performed. As revealed by the results, the pseudo resistivity calculated based on the neural network can reflect the target anomalies of the geoelectric model, with highly accurate network imaging results. Finally, the measured data were processed using the neural network algorithm, further indicating that the neural network-based imaging can serve as a basis for data interpretation. This study verified the feasibility of the ANN in TEM imaging, thus providing a new approach for TEM imaging.