用模拟退火法作AVA反演

AVA INVERSION BY USE OF SIMULATED ANNEALING

地矿部石油物探研究所

Research Institute of Geophysical Prospecting for Petroleum

模拟退火法是一种求解最优化问题的Monte Carlo算法,近年来开始应用于地球物理中。本文介绍了模拟退火法的基本思想及方法,其具体实现主要有两种算法,即Metropolis算法和热处理算法。笔者将这两种算法应用于地震叠前振幅反演,即反射振幅随入射角变化(AVA)反演中。一个合成数据集和两个超声物理模拟数据集的反演结果表明,模拟退火法是一种适于求解非线性最优化问题的方法,它的思路新颖,方法简单,易于实现,特别是易于实现参数和控制条件的约束,但这种方法的应用条件是正演计算速度要快。另外,尽管AVA中的强等值性使得AVA反演存在严重的多解性,但反演结果仍揭示了介质的弹性特征,说明用AVA反演来估计地层岩性和含油气性是可行的。

As a Monte Carlo technique for optimization problems, Simulated Annealing has been recently introdued into geophysics. This paper describes its foundamcntal concepts and practical schemes. Simulated Annealing can be implemented in two versions, namely the Metropolis and heatlth algorithms. Both of them are used by the author in prcstack amplitude inversion, i. c, the inversion of amplitude versus incidence (AVA) data. The results from inversion on a synthetic dataset and two ultrasonic modeling datasets show that Simulated Annealing is appropriate for solving nonlinear optimization problems. The method is independent of the initial guess model and especially easy to accept contraints on cither model parameters or con-troling conditions. The difficicncy is that the method calls for fast forward computations. In spite of the severe ambiguity of the solution in AVA inversion because of strong AVA equivalence, the resalts still reveal the elastic features for the media. This demonstrates the applicability of AVA inversion in estimating the lithology and hydrocarbon occurrence for strata.