论文部分内容阅读
把神经网络与重磁异常反演理论相结合，提出了用于重磁反演的一种拟BP神经网络方法．基于3层神经网络结构，把隐含层神经元设定为三维空间物性（磁化强度或密度）单元．对实测与理论重磁异常经S型函数变换，采用自动修改物性单元物性值的拟BP算法，反演三维空间的物性分布．利用该网络对理论模型数据和内蒙古某花岗岩体上的航磁资料进行了反演计算，取得了满意的反演效果．
Combining neural network with gravity anomaly inversion theory, a kind of quasi-BP neural network method for gravity and magnetic inversion is proposed. Based on the 3-layer neural network structure, hidden layer neurons are set as three-dimensional physical property (magnetization or density) units. The measured and theoretical gravity and magnetic anomalies are transformed by S-shaped functions, and the pseudo-BP algorithm that automatically modifies the physical properties of physical units is used to inverse the physical distribution in the three-dimensional space. Using this network, the theoretical model data and the aeromagnetic data of a granite body in Inner Mongolia were calculated and the satisfactory inversion results were obtained.