ACCELERATION OF FULL WAVEFORM INVERSION BASED ON RANDOM SUPER-SHOT AND GPU

作者：Wang Baoli^*; Gao Jinghuai; Chen Wenchao

来源：Journal of Seismic Exploration, 2011, 20(4): 331-346.

• full waveform inversion
• GPU
• CUDA
• velocity analysis
• super-shot

Wang, B, Gao, J. and Chen, W., 2011. Acceleration of Full Waveform Inversion based on Random Super-shot and GPU. Journal of Seismic Exploration, 20: 331-346.
The full waveform inversion (FWI) method is more and more important in the field of seismic exploration, which can be used to estimate the subsurface model by matching the simulated data with the acquired records. However, due to extreme computer-intensive time, FWI cannot be applied to real data widely. This paper describes a new strategy for the acceleration of FWI algorithm. Firstly of all, by generating the super-shots, the numbers of shots are reduced effectively, and thus, the computational cost is also reduced. Besides, the super-shots are regenerated randomly for suppressing the crosstalk noises and the artifacts caused by the summation during iterations. Furthermore, FWI is accelerated by the introduction of the hardware of graphics processing units (GPU) and the parallel programming based on GPU: Then, the synthetic records forwarded by finite-difference method in the time domain based on the Marmousi velocity model are employed to examine the proposed algorithm. Test results show that, as far as a same inversion level is concerned, our algorithm reduces the computational cost by 80 times than the conventional FWI method. Result also indicates that the super-shots regenerating randomly can suppress the crosstalk noise and the artifacts caused by summation than conventional methods. Finally, a marine real dataset is used to prove the superiority of RSSFWI based on GPU.

• 出版日期2011-12
• 单位西安交通大学