基于FFT-MA谱模拟的快速随机反演方法研究

虽然基于地质统计学的随机反演方法能够有效融合测井资料中的高频信息,但计算效率低,占用内存大,限制了它在实际资料中的应用.本文在保留传统随机反演方法优点的基础上,创造性地引入傅里叶滑动平均(Fast Fourier Transform-Moving Average, FFT-MA)谱模拟进行频率域的地质统计模拟,并利用逐步变形算法(Gradual Deformation Method,GDM)确保模拟结果与实际地震数据的匹配,构建了基于FFT-MA谱模拟的新的快速随机反演方法.与常规随机反演相比,新方法不仅分辨率高,而且能够使反演解得到快速收敛,有效提高计算效率,减少内存占用.模型试算获得了与理论模型吻合度较好的高分辨率反演结果.实际资料分析也表明新方法所得到的高分辨率反演结果能够对薄互储层进行良好的展示,为薄储层的识别提供高效可靠的技术支持.

Study of fast stochastic inversion based on FFT-MA spectrum simulation

Although the stochastic inversion based on geostatistics can effectively integrate high frequency information from well logs with higher resolution, however, it has low computational efficiency and too much memory consumption, which limits its application in real seismic data. The stochastic inversion is used to improve the operation efficiency of inversion and obtain high-resolution inversion results.#br#On the basis of the traditional stochastic inversion, a new method of stochastic inversion is proposed, i.e. the fast stochastic inversion based on fast Fourier transform-moving average (FFT-MA) spectrum simulation. This method is used to conduct geostatistical simulation in the frequency domain, and the gradual deformation updating method (GDM) can be used to accelerate the simulation convergence through continuous modification of the reservoir model until it matches the real seismic data.#br#FFT-MA simulation is a spectrum simulation method which can perform fast simulation in the frequency domain by Fourier transform. Compared with the sequential Gaussian simulation algorithm, the FFT-MA spectrum simulation can greatly improve the computational efficiency while has better simulation result. The conditioned FFT-MA spectrum simulation method can reconstruct the data which satisfy the specified covariance structure and grid size as well as the known well data. The GDM can get a series Gauss white noises realization which having the same mean and variance distribution. By changing the Gauss white noise, the FFT-MA disturbance simulation results will have all or part of the continuous modification. The synthetic seismic data can be obtained from the simulation results by forward modeling. Fast optimized stochastic inversion results will be obtained while the synthetic seismic data match well with the real seismic data. Compared with the conventional stochastic inversion, the fast stochastic inversion method can get high-resolution inversion result. Moreover, it can enhance the computational performance and reduce the memory consumption,and also accelerate the convergence process of the inversion. The model tests show that the proposed stochastic inversion method can obtain higher resolution of inversion results than the deterministic inversion method and match the model well. The analysis of the real data also shows that the high-resolution inversion result of the stochastic inversion method can have better exhibit thin interbedded reservoirs.#br#The fast stochastic inversion method can get higher resolution of inversion result than deterministic inversion method and has faster computational efficiency than conventional stochastic inversion. The proposed method can be used in the identification of thin reservoirs. The proposed method is not affected by the sequential factors; therefore it can be parallel computed in the future research which will further improve the computational efficiency.