Minmin Huang Leyi Xu Yanhui Zhu Ye He Zhiye Li Ying Lin ^*

• CNOOC China Limited, Shenzhen Branch, Shenzhen, China

The stochastic inversion method using logging data as conditional data and seismic data as constraint data has a higher vertical resolution than the conventional deterministic inversion method. However, it remains challenging to reduce the randomness of the prior obtained through conventional random simulation techniques and to enhance its accuracy. To address this, we propose a stochastic inversion method based on compressed sensing frequency-division waveform indication prior. This method fully considers the geophysical mapping relationship between the observed seismic data and the parameters to be inverted across different frequency bands. And the correlation coefficients between the seismic data at the known points and the predicted points are obtained by solving the low-rank system of equations through the compressed sensing method. Consequently, pseudo-kriging simulation of well data is performed based on the similarity between known and predicted seismic waveforms, thus establishing prior information indicated by the seismic waveforms. On this basis, the stochastic inversion results are solved using a very fast simulated annealing method. Both model calculations and field data inversion effects demonstrate that the compressed sensing frequency-division waveform indication method effectively improves the accuracy of solving prior information under a low-rank matrix. Ultimately, the proposed stochastic inversion method based on compressed sensing frequency division waveform indication prior enhances the inversion accuracy and provides advantages in identifying underground oil and gas reservoirs.