| With the increasing difficulty of oil and gas exploration, higher accuracy of seismic imaging methods are required, and the traditional methods can not satisfy the need of seismic data processing and interpretation. Full waveform inversion (FWI) is a novel velocity building method with high resolution. By matching the simulated seismic records corresponding to the starting model with the observed seismic records, the velocity model is continuously updated using one optimization algorithm, and finally obtain a velocity result which can be accurately described subsurface medium. FWI makes full use of kinematic and dynamic information of pre-stack seismic data, so the resolution is much higher than other traditional methods.Since FWI uses local optimization algorithm based on Born approximation, and has some limitations in solving strongly nonlinear problems, it highly depends on the quality of initial velocity model and low frequency information in the seismic data. FWI is easily trapped in local minima if seismic data lacks of low frequency information or using bad initial velocity model, and failure to obtain an accurate inversion result. Therefore, how to get a global optimal solution is one of the hotspots of FWI.This paper performed a deep research on the theory and application of frequency domain FWI. And we proposed some solutions for several problems and difficulties of FWI in this paper.Multi-scale strategy is commonly used to improve the stability of FWI. In order to reduce the dependence on the initial model of FWI, this paper proposed a new multi-scale FWI method combining with multi-scale characteristic of Curvelet transform. FWI goes from low frequency to high frequency to update the velocity.And at the different stages of FWI, select different scales of observed seismic data involved in the inversion according to the use of frequency. That is, first extract large scale data of observed seismic data involved in inversion at the beginning stage of FWI, and gradually add smaller scale data with the increase of frequency and accuracy. Thus simulated data can be better matched with observed data, and reduce the dependence on initial model of FWI. This paper compares the inversion results of multi-scale FWI method based on Curvelet transform and conventional FWI method when using bad initial model, and verifies the effectiveness of our new method.Cycle skipping is one of the main problems hindering FWI from practical application. And it occurs when the phase difference between simulated seismic data and observed seismic data is greater than half a cycle, which leads to wavefields incorrect match. Low frequency information is not sensitive to cycle skipping and plays a crucial role in FWI. In this paper we proposed Wavefield Phase Correlation Shifting (WPCS) FWI method drawing lessons from digital image correlation method to improve the effect of cycle skipping. Before inversion, we first calculate the phase cross correlation between simulated seismic data and observed seismic data. Then shift simulated seismic data according to the correlation coefficients to reduce the phase difference, making simulated data meet the requirement of FWI. We test the new method with a synthetic data and shows that the new method can obtain a global optimal result without low frequency in seismic data, while conventional FWI method falls into local minima and failure to get an accurate solution.As FWI is strict with low frequency and signal to noise ratio in seismic data, this paper applied over/under towed streamer data to FWI. The depth of streamer affects the frequency band and ghost notch frequency of data. That is, the deeper streamer depth is, the more low frequency information can be received, and the shallower streamer depth is, we will receive more high frequency information. Thus, frequency band can be extended and ghost suppressed by combining over and under streamer data. Hence it improves the stability and inversion depth of FWI when applying over/under streamer combined data. |
PDF Full Text Request