神经网络反演方法在地震储层上的应用

Neural network inversion on application of seismic reservoir

基于神经网络对地球物理资料进行反演,其原理是把希望输出与实际输出之差偏差归结为连接权的"过错",通过把输入层单元的误差,逐层向输入层逆向回转,分摊给各层单元,从而获得各层单元的参考误差,以调整相应的连接权;并讨论了神经网络反演和基于模型反演在少井的情况下对反演效果的影响,以及神经网络反演中的各参数对分辨率的影响,对神经网络反演的应用及推广提供了相关参考.

The principle of neural network inversion of geophysical data is that differences between desired output and actual output are summed up as the ＂fault＂ of connection weights,and then,the error of input layer unit is reversed rotation to the input layer,layer by layer,and allocates to the layers of units in order to obtain a reference error of layers units and to adjust the connection weights of the corresponding.The influences of neural network inversion and model based inversion in the case of less well effect inversion and the various parameters of neural network inversion on resolution ratio are discussed,and provide relevant information to application for neural network inversion.