基于神经网络的视电阻率快速算法

本文从瞬变电磁均匀半空间二次磁场响应公式出发，提出了一种基于神经网络的视电阻率快速计算方法.以中心回线为例，根据瞬变响应公式的特点，简化网络结构，选用三层BP神经网络和误差训练算法，用均匀半空间样本数据进行训练，确定了收敛快、误差小的一步正割法和隐含单元数，得到基于不同采样时窗的一组网络参数.用本文方法与二分法、牛顿迭代法做模型计算比较，及最后的实验计算，说明算法的快速，准确.本文方法不依赖初始模型，避开了复杂的电磁场数值计算，实现了视电阻率的快速计算，对瞬变电磁法资料的快速解释有一定的参考价值.

A fast algorithm of apparent resistivity based on neural networks

According to expression of the secondary field in half infinite homogeneous medium space, this paper proposed an algorithm for calculating apparent resistivity based on neural networks. Taking center-loop-line set for example, after analyzing the characteristic of the expression in transient, a predigest structure of network is obtained. Then, a three-layer BP neural network is trained by using numerical calculation methods and sample data in homogeneous half-space, and its number in the hidden layer is obtained on basis of rapidly convergent and little error. Naturally, we got a series of parameter based on different time windows in network. Finally, compared to the traditional numerical calculation methods, such as dichotomy, Newton with the model and experiment, BP neural network method does prove its fast and accurate feature. The method proposed in this paper, which does not depend on the initial model and avoids a complex electromagnetic field numerical calculation, achieves a rapid apparent resistivity of calculation and will be useful with reference to the explanation of the transient electromagnetic method.