Research On The Optimized Algorithm And Application Of Least-squares Reverse Time Migration

As the exploration and development of oil and gas in China go deep into a new era,the key exploration targets have been gradually changed from conventional clastic reservoirs and carbonate reservoirs into unconventional tight oil reservoirs and shale oil reservoirs.The exploration and development of unconventional reservoirs have the problems including complex geological structures,large buried depth and low quality seismic data,thus it is important to develop the precise and amplitude-preserved imaging methods for deep reservoirs,providing reliable data bases for well location arrangement and reservoir description.The paper first introduces the theory of least-squares reverse time migration(LSRTM),with a detailed introduction of the forward operator,the conjugate operator and the data processing flow of LSRTM.LSRTM has the advantages of high signal-to-noise ratio,high resolution and high fidelity;therefore it is a high precision imaging method suitable for the exploration of deep reservoirs.However,LSRTM has some drawbacks including poor sensitivity to the quality of seismic data,slow convergence rate and huge computational cost which limits its further application to the industrial field data.To solve the imaging problem of the seismic data with low signal-to-noise ratio,the paper analyzes the influence of noise on LSRTM and verifies that reducing the noise residual improves the image quality of the seismic data with low signal-to-noise ratio.Therefore,a weighted misfit function is proposed based on the predictability differences between the desired signals and noise to suppress the noise residual by imposing constraint on the data residual.And in particular,a primary-guided weight matrix is derived to suppress the data residual associated with unpredictable noise such as random noise and spiky noise,while a multiple-guided matrix is derived to suppress the data residual associated with free surface related multiples.A weighted conjugate gradient method is used to minimize the weighted data residual and find the best migration image.The numerical tests on the synthetic data and field data demonstrate that the reweighted LSRTM is more robust that standard LSRTM for imaging the seismic data with low signal-to-noise ratio.The weighted matrix only reduces the influence of noise on LSRTM,but cann't fully removes the migration aritfacts.Therefore,the regularization methods of LSRTM are presented to suppress the migration artifacts and accelerate the convergence rate.The singular spectrum analysis(SSA)algorithm is introduced into LSRTM to derive an adaptive singular spectrum constraint.Additionlly,a regularization method using the prior model is presented to provide the image of the structures that the seismic data cannot illuminate.The prior model constructed by well data imposes constraint on the migration image to fit the well data.The numerical tests on the synthetic data and field data demonstrate that the regularization methods can improve the image quality and accelerate the convergence rate of standard LSRTM.To improve the computational efficiency of LSRTM,the paper combines the plane-wave encoding technique and the stochastic optimization algorithm to present the plane-wave LSRTM method with preconditioned stochastic conjugate gradient.In plane-wave LSRTM,the plane-wave encoding technique is applied to the massive seismic data to produce several plane-wave gathers,thereby reducing the computational cost.Based on plane-wave LSRTM,the paper derives a plane-wave angle domain preconditioning operator which suppresses the migration artifacts in the take-off angle domain common image gathers(TADCIGs)by the singular spectrum constraint.Then,the plane-wave encoding formula is presented for plane-wave LSRTM from rugged topography that improves the image quality of near-surface structures.In addition,the paper derives the hybrid stochastic conjugate gradient(HSCG)method for LSRTM.HSCG method reduces the computational cost by applying a shot sampling method to the seismic data,and it uses a hybrid searching direction of the conjugate gradient iteration and stochastic gradient iteration such that it has faster convergence rate than conventional stochastic optimization method.At last,the HSCG method is combined with the plane-wave angle domain preconditioning operator to propose a preconditioned stochastic conjugate gradient(PSCG)method which is then applied to plane-wave LSRTM.Plane-wave LSRTM with PSCG method can produce not only high quality migration images but also high quality TADCIGs.LSRTM produces the image of the full observed data which cann't fully uses the information of diffractions.The last part of the paper reports a target-oriented LSRTM method to improve the image quality of deep-part and subsalt inhomogeneous structures.The paper introduces the workflow of the diffraction separation method on plane-wave gathers and the LSRTM method of diffractions.LSRTM of diffractions can produce high quality images of inhomogeneous structures.Then,the paper derives the misfit function of the target-oriented LSRTM method which uses a reflection damping operator to control the inverted wavefield.In detail,target-oriented LSRTM inverts the large-scale background structures by using the reflection data while it improves the image quality of the small-scale inhomogeneous structures.Proper reflection damping operator should be selected according to the imaging demand in practical application.