中文摘要：

作为高精度波形反演或逆时偏移的重要组成部分，地震波数值模拟对计算速度和效率提出了更高要求．GPU通用计算技术的产生及其内在数据并行性，为高效地震波数值模拟应用和研究得以有效开展奠定了基础．本文借助交错网格的Fourier伪谱微分矩阵算子和GPU上高效矩阵乘法，实现了复杂介质中地震波模拟的高效算法．数值试验表明，优化后的GPU计算相比CPU单核计算在大规模二维地震波场计算中获得至少100x以上的加速比．这对我们快速分析目标反射层在地震剖面中同相轴位置，制定优化采集方案具有重要意义．

英文摘要：

As a crucial part of high-resolution waveform inversion and reverse time migration, seismic modeling demands a high efficiency of computation. The general-purpose GPU computing technology provides a fundamental basis as for GPU was naturally designed with data parallelism. By virtue of staggered-grid Fourier pseudo-spectral differentiation matrix operator, solving wave equation can be easily implemented on GPU by matrix multiplication. In large-scale twodimensional wavefield computation, optimized single GPU forward seismic modeling can achieve at least 100x faster than that of single CPU using only one core. The high speedup of GPU/CPU implementation enables us to quickly recognize seismic events of the reflectors of different layers in common shot profile, which is helpful for optimizing the acquisition configuration in field work.