论文部分内容阅读
本文将振幅随偏移距变化的AVO反演纳入贝叶斯统计框架中。在这一框架之下,利用模型参数和物理正演生成合成数据,然后将这些合成数据与观测资料进行匹配,获得模型空间内的后验概率密度(PPD)函数。遗传算法(GA)利用定向随机搜索技术来估算PPD的形状。与传统的反演方法不一样,GA不依赖于初始模型的选择,所以很适合于AVO反演。 文中先对单层模型资料进行了AVO反演,这时的振幅仅来自同一反射同相轴。用GA估算其法向入射反射系数(R_0)和泊松比(△σ),即使在信噪比较低的情况下其结果也达到合适的精度。对比表明,GA反演所获结果比生产中常用的反射系数最小平方拟合结果更精确。 对于多层模型数据的AVO波形反演,文中进行了全部或部分叠前资料的反演。这种类型的反演估算出的速度、泊松比和密度绝对值是非唯一的。然而,通过采用简化近似方法,文中验证了经反演可以估算出R_0、声阻抗比值(△A)和反射系数梯度(G)。根据R_o、△A和G的GA估算值和输入数据起始时间处的速度、泊松比等估算结果,能生成反演后的模型。对海上资料应用这一过程,证实了根据反演后模型计算出的合成数据与输入资料的匹配满足精度要求。与附近井中的测井资料进行比较,表明GA反演可以得到P波声阻抗的低频趋向和高频趋向(地震分辨率带宽范围内)。?
In this paper, AVO inversion with amplitude variation with offset is incorporated into the Bayesian statistical framework. Under this framework, synthetic data are generated using model parameters and physical forward modeling, and these synthetic data are then matched with the observed data to obtain the posterior probability density (PPD) function in the model space. Genetic Algorithms (GA) use directional random search techniques to estimate the shape of a PPD. Unlike conventional inversion methods, GA does not depend on the choice of initial model, so it is suitable for AVO inversion. In this paper, the AVO inversion of single-layer model data is carried out firstly, at this time, the amplitudes are only from the same reflection axis. Using GA to estimate its normal incidence reflection coefficient (R_0) and Poisson’s ratio (Δσ), the result is satisfactory even with low signal-noise ratio. The comparison shows that the results obtained by GA inversion are more accurate than the least squares fit of reflection coefficient commonly used in production. For the AVO waveform inversion of multi-layer model data, all or part of the pre-stack data is retrieved. This type of inversion estimates the velocity, Poisson’s ratio, and density absolute value non-unique. However, by using the simplified approximation method, it is verified that the R_0, the acoustic impedance ratio (ΔA) and the reflection coefficient gradient (G) can be estimated by inversion. Based on the GA estimates of R_o, ΔA and G and the estimated velocity, Poisson’s ratio at the start of the input data, the inverted model can be generated. The application of this process to maritime data confirms that the matching of the synthesized data calculated from the post-inversion model with the input data satisfies the accuracy requirements. Compared with the well logs in the nearby wells, it is shown that the low-frequency trend and the high-frequency trend of P-wave acoustic impedance (within the range of seismic resolution bandwidth) can be obtained by GA inversion. ?