论文部分内容阅读
常规频域预测滤波方法是建立在自回归(autoregressive,AR)模型基础上的,这导致滤波过程中前后假设的不一致,即首先利用源噪声的假设计算误差剖面,却又将其作为可加噪声而从原始剖面中减去来得到有效信号。本文通过建立自回归-滑动平均(autoregres sive/moving-average,ARMA)模型,首先求解非因果预测误差滤波算子,然后利用自反褶积形式投影滤波过程估计可加噪声,进而达到去除随机噪声目的。此过程有效避免了基于AR模型产生的不一致性。在此基础上,将一维ARMA模型扩展到二维空间域,实现了基于二维ARMA模型频域非因果空间预测滤波在三维地震资料随机噪声衰减中的应用。模型试验与实际资料处理表明该方法在很好保留反射信息同时,压制随机噪声更加彻底,明显优于常规频域预测去噪方法。
The conventional frequency-domain prediction filtering method is based on the autoregressive (AR) model, which leads to inconsistent assumptions in the filtering process, that is, the error profile is first calculated using the assumption of the source noise, Subtracting from the original profile yields a valid signal. In this paper, we establish the autoregress sive / moving-average (ARMA) model to solve the non-causal prediction error filter operator, and then estimate the additive noise using the reflexive convolution form of the projection filtering process to achieve the removal of random noise purpose. This process effectively avoids inconsistencies arising from the AR model. Based on this, the one-dimensional ARMA model is extended to the two-dimensional space domain, and the application of two-dimensional ARMA model based on frequency-domain non-causal spatial prediction filtering in the random noise attenuation of 3D seismic data is realized. The model test and the actual data processing show that this method preserves the reflection information well and suppresses the random noise more completely, which is obviously better than the conventional frequency domain prediction denoising method.