用模拟退火神经网络技术进行波阻抗反演

IMPEDANCE INVERSION BY USING ANNEALING NEURAL NETWORK

利用基于模拟退火算法的神经网络技术进行测井约束的波阻抗反演，可根据数据本身之间的内在联系建立一个自适应非线性认知系统，只要在输入端输入特征数据，便能在输出端得到期望输出值，而不必关心系统本身的内部机理。在反演前，从测井资料中整理出地层波阻抗参数，用神经网络建立起地震波特征和地层波阻抗参数的映射关系，然后再利用这种映射关系进行外推，得到其它地震道所对应的波阻抗参数。在训练过程中，引入了模拟退火算法，使网络能有效地避开局部极小，这样可以提高收敛速度和拟合精度。

Impedance inversion is carried out by using simulated annealing neural network technology. An nonlinear system can be establishied based on the numerical relationship. At first, the impedance parameters are gotten from well log data, and then the mapping of seismic characteristics to acoustic impedance is constructed by neural network.The corresponding information of other traces can be extrapolated by using the mapping.In the process of training, simulated annealing algorithm is introduced, and local minima are effectively avoided. The speed of convergence and the simulation procision can be improved.