Research On Thin Interbed Frequency Band Extension Method Based On Compressed Sensing

Thin bed is a layer that can't be identified on the seismic profiles and its round travel time thickness is less than the incident wave tuning thickness(1/4 wavelength),while thin interbed is a set of strata consisting of multiple thin layers that cannot be distinguished by seismic reflection for each single layer.Owning to main frequency is low,thin interbeds often appear as mixed composite waves on seismic profile to make seismic description and prediction difficult.While current research difficulties of thin interbeds include three problems.1.Thin interbed reservoir is too thin to identify,so seismic resolution is not enough.2.Due to the strong shielding cap,lower part reflection is weak.3.The lateral heterogeneity of the reservoir is very strong,which continuous is not easy to track and control.Therefore,this paper has carried out a series of studies on thin interbeds.Firstly,according to seismic convolutional forward modeling,wedge-shaped model of the two-layer thin interbeds and typical multi-layer model of thin interbeds are both discussed to study characteristics and difficulties of thin interbeds on synthetic seismic record,which provides a theoretical basis for sparse inversion and frequency band extension of thin interbed seismic data.Secondly,compressed sensing is introduced and the feasibility of sparse signal reconstruction is verified.For seismic data,we deduce the conventional parse inversion formula of reflection coefficient based on compressed sensing.In order to to improve the seismic data resolution,the low frequency information of seismic data is compensated by the reflection and high frequency information is extended using wide band wavelet convolution.Thirdly,we discuss effects of parameters,such as algorithms,regular terms and wavelets in the process of sparse inversion and expanding frequency.Parallel computing,de-biase idea,retrospective thinking,multi-channel weighted complex spectrum and wavelet spectrum simulation are introduced to improve iterative efficiency,algorithm effect,regular term operator selection,inversion wavelet extraction and wide band wavelet selection.Finally,since noise and strong shielding problems,combining the Gaussian function and the information along the layer,we present both a method for Gaussian frequency domain to extend frequency based on compressed sensing and a de-strong shielding method on reflection coefficient domain by the L2-norm constrained based on compressed sensing.Model test and actual data application demonstrate the effectiveness of two new methods.