基于FCM聚类模型约束的二维初至旅行时反演

2D inversion of seismic first⁃arrival traveltime based on FCM clustering model constraint

最小结构模型约束正则化二维地震初至旅行时反演中存在模型边界刻画不清的问题,尤其是地质体内射线分布稀疏的情况下,反演效果不理想。为此,引入模糊C均值(FCM)聚类模型约束函数,旨在提高反演结果对模型边界的成像精度。该约束项将先验信息作为参考聚类中心,在迭代过程通过反复修改聚类中心及每个网格单元对聚类中心的隶属度,实现对速度的自动分类。在此基础上,采用以模型灵敏度信息为依据的多重网格反演策略,以提高反演的稳定性及效果;应用简单模型讨论了FCM聚类模型约束权重、先验信息引导项权重等参数选取方案;对比无监督学习与先验信息监督学习的反演效果,后者改善了反演速度模型边界刻画模糊现象,有效提高了反演结果的分辨率;最后,通过实测数据反演,验证该方法在实际应用中的实用性和有效性。

The 2D inversion of seismic first⁃arrival traveltime with minimum structure model constraint regularization suffers from the issue of insufficient delineation in model boundaries,particularly when dealing with sparse ray distribution within geological structures,leading to unsatisfactory inversion results.To address this challenge,this paper introduces a model constraint function based on fuzzy C⁃means(FCM)clustering,so as to improve the accuracy of the inversion results in delineating the model boundaries.This constraint incorporates prior information as reference cluster centers and employs an iterative process to repeatedly modify the cluster centers and the membership degrees of each grid cell to the cluster centers,enabling automatic classification of velocities.On this basis,a multi⁃grid inversion strategy guided by model sensitivity information is adopted to enhance the stability and effectiveness of the inversion.Parameter selection schemes for FCM clustering model constraint weights and weights of the prior information guidance term are discussed using simplified models.A comparison is made between the inversion results of unsupervised learning and prior information⁃supervised learning approaches.The latter approach successfully addresses the issue of blurring in depicting the boundaries of the velocity model during inversion,effectively improving the resolution of the inversion results.Finally,the inversion of real measured data verifies the applicability and effectiveness of the approach in practical applications.