基于BP神经网络的高密度电阻率数据反演方法

High-Density Resistivity Data Inversion Method Based on BP Neural Network

目的对高密度电阻率数据反演计算,解决传统局部线性化迭代反演时求解容易陷入局部模型极小解和依赖初始模型的选择等问题.方法采用二维改进有限元法计算不同地电模型的视电阻率,利用所得结果训练BP神经网络,神经网络的自学习能力将使最终输出为全局最优解.结果将神经网络与最小二乘法的反演结果作以比较,BP神经网络能够得到全局最优解,因此能较好地反映地电模型,并且计算过程中不需要选择初始模型.结论神经网络训练成功后,能够克服传统反演方法的不足,因此适合解决复杂的非线性反演问题.

In order to solve the problem that the solution of the traditional local linear iterative inversion is likely to fall into the local minimum solution and rely on the choice of initial model this paper applies BP neural network to calculate high-density electrical resistivity.The improved two-dimensional finite element method is used to calculate the apparent resistivity of different geoelectric models and the result is applied to train BP neural network and the self-learning ability of the neural network can output the global optimal solution.This paper compares the result of the neural network with the least square method.The result shows that the neural network can gain the global optimal solution,so it can better reflect the geoelectric model and the initial model is not required in the process of the calculation.Once the neural network is trained successfully,it is able to overcome the deficiencies of the traditional inversion method.Thus it is suitable to solve complicated non-linear inversion problems.