地震波能量补偿的并行反Q滤波方法研究

Research on Parallel Inverse Q Filtering Methods for Seismic Wave Energy Compensation

张全 王一品 张伟 彭博 胥林

ZHANGQuan WANGYipin ZHANGWei PENGBo XULin

西南石油大学计算机科学学院, 四川 成都 610500 电子科技大学信息与通信工程学院, 四川 成都 611731 西南石油大学信息学院, 四川 南充 637001

School of Computer Science, Southwest Petroleum University, Chengdu, Sichuan 610500, China School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China School of Information, Southwest Petroleum University, Nanchong, Sichuan 637001, China

在石油勘探地震资料处理中，反Q滤波方法能有效地对地震波进行振幅补偿和相位校正，为地震反演和储层预测提供更准确的信息。对于大规模的地震道集数据处理，反Q滤波方法在CPU计算平台上执行时间较长，影响了地震解释的效率。分析发现，反Q滤波方法大量时间消耗在振幅相位补偿与短时傅里叶变换。在GPU平台上，首先，对振幅相位补偿部分进行并行化；其次，对批量短时傅里叶变换用CUFFT库进行加速；最后，对批量短时傅里叶变换进一步优化并将其应用于反Q滤波方法。实验结果表明，相比CPU计算环境，基于CUFFT库的反Q滤波并行算法效率提升了3.9倍，优化后的批量短时傅里叶变换进一步将效率提升了12%。

In seismic data processing of petroleum exploration, the inverse Q filtering method can effectively perform amplitude compensation and phase correction on seismic waves to provide more accurate information for seismic inversion and reservoir prediction. In large-scale seismic data processing, the inverse Q filtering method takes longer operation time under the CPU computing platform, which affects the efficiency of seismic interpretation. After analysis, it is found that the inverse Q filtering method consumes a lot of time in the short-time Fourier transform and calculates the amplitude and dispersion compensation terms. On the GPU platform, we first parallelizes the amplitude and dispersion compensation calculations, and accelerates the batch short-time Fourier transform with the CUFFT library, and then further optimizes the batch short-time Fourier transform and applies it to the inverse Q filtering method. The results show that compared with the CPU computing environment, the efficiency of the inverse Q filtering parallel algorithm based on the CUFFT library is improved by 3.9 times, and the optimized batch short-time Fourier transform further improves the efficiency of the parallel inverse Q filtering method by 12%.

10.11885/j.issn.1674-5086.2021.02.03.03