拟神经网络反演在煤田中应用

Parker快速富氏变换反演单一密度界面存在两方面问题：一是求解目标界面重力异常难度较大;二是受正演速度及反演参量维数的限制,不能对界面进行精细划分.拟神经网络BP算法的引入,首先解决了快速三维正演问题,又突破了反演参量维数的限制,实现快速收敛,有效解决两个或多个密度界面的反演问题.在实际应用中,先用密度“补偿法”正演求取剩余生力异常,然后利用拟神经网络BP算法同时反演两个二维密度界面,拟合求得两个界面的深度异常,在此基础上预测煤田.

Application of Quasi-Neural Network Inversion in Coalfield

There are problems in two aspects of the Parker fast Fourier transform for inversion of unitary density interface: the first is rather difficult to solve gravity anomaly on target interface;while the second is limitation from forward problem solving rate and inversion parameter dimensionality,the interface cannot be divided precisely.The introduction of quasi-neural network BP algorithm,firstly solved the fast 3D forward problem,then the breakthrough of limitation from inversion parameter dimensionality,realized the fast convergence,so that effectively solved the inverse problem with two or multiple density interfaces.In practical application,to use density compensation method impetrated forwardly residual gravity anomaly,then to use the quasi-neural network BP algorithm invert two 2D density interfaces simultaneously,to fit up depth anomaly of two interfaces,and predicting coalfield on this basis.