地震波形反演中的梯度正则化

Gradient regularization in seismic waveform inversion

当初始模型不够精确,以及当地震记录或者在物理场的近似模拟过程中存在噪音时,地震反演算法往往不可避免的收敛于目标函数的局部极小值。为了提高波形反演的稳定性,在计算模型修正量的过程中,增加梯度正则化是有必要的。该策略是在常规多尺度反演的基础上,根据在反演迭代期间目标函数的梯度变化趋势得到实现。

When the initial model is not accurate enough,and there is noise in the seismic record or the approximate simulation of the physical field,the seismic inversion algorithm tends to converge to the local minimum of the objective function.In order to improve the stability of waveform inversion,it is necessary to add gradient regularization in the process of calculating model correction.Based on the conventional multi-scale inversion,the strategy is realized according to the gradient trend of the objective function during the inversion iteration.