关键词: 全波形反演, 频率域, L-BFGS算法, 同时激发震源, 多尺度

Abstract:

Full waveform inversion (FWI) can reconstruct underground velocity while utilizing the kinematic and dynamic information of pre-stack seismic data, which could reveal detail information of the structure and lithology under complex geological background. However, huge calculating amount and storage space requirements confine the development of full waveform inversion. A combined method of L-BFGS algorithm and simultaneous sources technology is applied to ameliorate this problem during frequency multi-scale full waveform inversion. First we carry out velocity inversion based on the Marmousi model. It can be obviously found that the memory spending has been improved considerably in the process of calculation, and the fitting error between the final inversion result and true Marmousi model is as small as 0.095 9. Meanwhile, single shot forward modeling 384 times takes about 32 640 seconds when using 10 frequency bands, while simultaneous sources which contain 384 shots forward modeling one time needs only about 700 seconds. Then we conduct research on the anti-noise ability based on the high-speed wedge model. The signal to noise ratio of original seismic data is 11.147 3 dB, while the signal to noise ratio of the seismic data obtained from forward modeling based on inversion velocity model is 22.251 8 dB. Finally we conduct research on inversing the medium with velocity perturbation based on the thrust fault model. The final inversion model is very clear and very satisfied with the real model with velocity perturbation characteristics, and the fitting error between them is only 0.036 0. Numerical simulation experiment result indicates that the inversion precision of this method is higher enough, and the memory spending is less, which could significantly increase calculation efficiency. What is more, this method has good anti-noise ability, and could inverse the medium with velocity perturbation characteristics.

Key words: full waveform inversion, frequency-domain, L-BFGS algorithm, simultaneous sources, multi-scale