论文部分内容阅读
本文阐述了一种基于全局优化的算法——平均场遇上算法(简称MFA),及其在地震数据处理中的两个基本问题地震道反褶积和与表面有关的多次波衰减中的应用。基于平均场近似,MFA法用一套确定性的修正规则代替了模拟运火的随机性,且这些规则不是作用在变量自身上,而是作用在变量的均值上。当温度降低时,MFA规则根据前一温度上各个变量的值修正其变量。通过最小化平均值,MFA较常规模拟退火法能更快地收敛到平衡状态。修正规则依赖于价格函数的形式,在价格函数类似于Hopfield网络的能量函数时,很容易获得。将一个问题映射成Hopfield网络不是使用MFA的前提条件,但映射能使解析计算更为简单。通过将震源子波和反射系数用二值神经元表示,地震反褶积问题可映射到一个Hopfield网络。在本文中,当网络的神经元到达平衡状态时,问题就得到了解决。通过最小化MFA网络的价格函数并使用一个恰当的冷却进程表,就能使系统脱离局部最小值。假设一个一维的地质模型,用同样的思路来设计算子,此算子可用来衰减由平面波转化的地震记录中与表面有关的多次反射波。消除多次波问题的价格函数是基于被多次压制地震数据的最小能量标准的。
In this paper, we present an algorithm based on global optimization (MFA) and its two basic problems in seismic data processing: deconvolution of traces and surface-related multiple attenuation application. Based on the mean-field approximation, the MFA method replaces the randomness of simulated fire with a set of deterministic correction rules, and these rules act not on the variables themselves but on the mean of the variables. As the temperature decreases, the MFA rules correct its variables based on the values of the individual variables at the previous temperature. By minimizing the average, MFA converges more quickly to equilibrium than the conventional simulated annealing method. The correction rules depend on the form of the price function, which is easily obtained when the price function is similar to that of the Hopfield network. Mapping a problem to a Hopfield network is not a prerequisite for using MFA, but mappings make parsing calculations much easier. The seismic deconvolution problem can be mapped to a Hopfield network by representing the source wavelets and reflection coefficients in binary neurons. In this article, the problem is solved when neurons in the network reach equilibrium. By minimizing the price function of the MFA network and using an appropriate cooling schedule, it is possible to get the system out of the local minimum. Assuming a one-dimensional geological model, we use the same idea to design operators that can be used to attenuate surface-related multiple reflections in seismic records transformed from plane waves. The price function to eliminate the multiple problem is based on the minimum energy standard that has been suppressed by seismic data multiple times.