地震资料逆时偏移中的图形处理器加速算法①.PDFVIP

2013 年 第 22 卷 第 11 期 计 算 机 系 统 应 用 地震资料逆时偏移中的图形处理器加速算法① 柯 璇, 石 颖, 刘诗竹 (东北石油大学 地球科学学院, 大庆 163318) 摘 要: 叠前逆时偏移(RTM)方法是目前地震勘探领域最为精确的一种地震数据成像方法, 其运用双程声波方程 进行波场延拓, 可实现对复杂构造介质的准确成像. 文中采用互相关成像条件对震源波场与检波点波场在同时 刻相关成像. 针对 RTM 方法计算量大的问题, 将图形处理器(GPU)引入到RTM 计算中, 充分挖掘 GPU 的众核结 构优势, 利用基于 CUDA 架构的并行加速算法取代传统 CPU 的串行运算, 对逆时偏移算法中较为耗时的波场延 拓和相关成像过程进行加速. 复杂模型测试结果表明, 在确保 RTM 成像精度的前提下, 相比于传统 CPU 计算, GPU 并行加速算法可大幅度地提高计算效率, 进而实现基于 GPU 加速的叠前逆时偏移算法对复杂介质的高效 率、高精度成像. 关键词: GPU; CUDA; 逆时偏移; 加速; 地震资料 GPU Acceleration Algorithm of the Reverse Time Migration with Seismic Data KE Xuan, SHI Ying, LIU Shi-Zhu (School of Earth Sciences, Northeast Petroleum University, Daqing 163318, China) Abstract: Currently,prestack reverse-time migration is the most accurate imaging method for seismic data in seismic prospecting domain. It extrapolating the wave field with the two-way acoustic wave equation, and it can image complex geological structure accurately. The cross-correlation imaging condition is used for the imaging of source wavefield and receiver wavefield at the same time in the paper. For computationally intensive problems of RTM,we introduce the graphics processing unit (GPU) into RTM algorithm, and exploit the multicore advantages of GPU. In this paper,we use the parallel acceleration algorithm base on the CUDA architecture to replace the serial computation on the traditional CPU and accelerate the process of the wavefield extrapolation and cross-correlation imaging in reverse time migration. The test on complex modeling show that we can achieve imaging result for complex medium with high efficiency and precision by pre-stack reverse time migration algorithm base on GPU acceleration. Under the premise of ensuring the calculation accuracy of the RT