The thick loess sediments produce a lot of noise during seismic acquisition, typically surface wave. Conventional methods are difficult while suppressing 3D surface waves since they are hyperbolic in the far-array. This study proposed a new model-based, data-driven surface wave attenuation method, which uses dispersion curve analysis of surface wave and joint inversion based on genetic algorithm (GA) and conjugate gradient algorithm, to build an accurate surface wave model, and then to subtract the model data from seismic data to improve the signal to noise ratio (S/N ratio). Located in the southern part of the Ordos Basin, the Loess Plateau has the characteristics of large thickness of loess sedimentary layer, diverse geomorphology, complex geological conditions and developed surface waves. After applying the method to the field seismic data, numerical tests show that joint inversion of dispersion waves was stable, accurate, and efficient in multi-type models, including models with a low-velocity layer and with a high-velocity layer. The result showed that the 3D surface wave attenuation method is effective in surface wave suppression. The low-frequency information and the reflected signal are efficiently preserved. Seismic faces are more apparent for reservoir characterization. This method provided credible data for the exploration of the loess tableland area.

厚厚的黄土沉积物在地震采集过程中会产生大量噪声，通常是表面波。传统方法很难抑制 3D 表面波，因为它们在远阵中是双曲线的。本研究提出了一种基于模型、数据驱动的面波衰减新方法，利用面波频散曲线分析以及基于遗传算法（GA）和共轭梯度算法的联合反演，建立精确的面波模型，然后从地震数据中减去模型数据以提高信噪比（S/N 比）。黄土高原位于鄂尔多斯盆地南部，具有黄土沉积层厚度大、地貌多样、地质条件复杂、面波发育等特点。将该方法应用到现场地震数据中，数值试验表明，在低速层和高速层等多类型模型中，频散波联合反演稳定、准确、高效。结果表明，3D表面波衰减方法对于抑制表面波是有效的。低频信息和反射信号被有效保留。地震面对于储层表征更为明显。该方法为黄土塬区勘探提供了可靠的数据。