| Pre-stack inversion is a key technique used in oil and gas field exploration and development to obtain reservoir properties,such as elastic parameters(longitudinal and transverse wave velocity,density,Poisson’s ratio,etc.)and physical parameters(porosity,fluid saturation,etc.)from seismic data.However,the quality of seismic inversion results is usually governed by a number of factors.Traditional inversion algorithms,such as linear inversion methods,have low accuracy results,and numerical inversion methods tend to trap the solution in local minima.In addition,noise,finite bandwidth and underdetermination problems in seismic data can lead to multiple solutions and uncertainties in seismic inversion,thus posing challenges for accurate reservoir interpretation and evaluation.For complex geophysical problems with non-linear and non-unique properties,this thesis proposes a Markov Chain Monte Carlo(MCMC)non-linear stacked forward inversion algorithm based on structural constraints to characterize the solution space of reservoir elastic parameters(longitudinal and transverse wave velocity and density).The method is based on Bayesian theory and uses Markov chains to randomly sample the parameter space,which in turn gives the opportunity to jump out of local extremes in the sampling process.It also provides a framework for quantifying the uncertainty of the inversion results based on the estimated posterior probability distribution of the model parameters,resulting in a more comprehensive and accurate inversion parameter space.Compared to the conventional method,this thesis also uses the Plane Wave Deconstruction Filter(PWD)to extract the tectonic dip and introduces geological constraints to the inversion,which improves the spatial continuity of the results and obtains more geologically appropriate inversion results.At the same time,this thesis uses a modified method of hybrid Bayesian Linear Inversion(BLI)and MCMC non-linear inversion method,as well as Block coordinate descent(BCD)algorithm,to solve the problems of long sampling burn time,high computational cost and memory consumption due to the drastic increase of data dimensionality.The problems of long sampling and burning periods,high computational costs and memory consumption due to the dramatic increase in data dimensionality are solved,making the MCMC inversion algorithm based on constructive constraints more feasible in practical applications.The results of model and actual data inversion show that the method proposed in this paper can effectively improve the accuracy and efficiency of reservoir parameter inversion,while ensuring the continuity of inversion results along the stratigraphic direction,which has certain practical application value. |
