神经网络在地震资料处理中的应用

Neural Network Approach to Seismic Signal Processing

基于人工神经网络的非线性表达能力和信号在神经网络中传播的机理,提出采用多层前馈神经网络模型实现多维数据的非线性滤波及降维。考虑到大地介质高度的非均质性和各向异性引起的地球物理信号的非线性,将这种神经网络应用到了地震资料处理之中。采用人工合成的多道地震记录试算,结果证明了这种方法的有效性。

This paper is based on the capability of artificial neural network in expressing nonlinearity inherent in input data and the mechanism to propagate signals through neural network. Two multi-layer feedforward neural 'network models for filtering and dimensionality-reduction of multidimensional data and their implementation are proposed. In consideration of the nonlinearity inherent in geophysical signals resulted from the high heterogeneity and anistropy of the earth medium, these models are used in processing multichannel seismic signals. The results from the experiments with synthetic seismic records have verified the effectiveness of this neural network approach.