Geostatistics Stochastic Modeling With Machine Learning

Geostatistics stochastic modeling method is one of the important techniques for geological property simulation.The core idea is to calculate the experimental variance function from known well data,then construct and solve the kriging equation according to the random path for interpolation simulation,to produce the possible look of the geological model.In the traditional geostatistics stochastic modeling,the following problem generally exists: traditional geostatistics stochastic modeling method usually does not consider the influence of the division of seismic phase on the modeling process,leading to the modeling results can't fully reflect the geological spatial structure characteristics,phasecontrolled stochastic modeling method was born to solve this problem.However,traditional phase-controlled stochastic modeling still has shortcomings: 1)When traditional phase-controlled stochastic modeling method classifies the seismic phase,it is often to process a large number of geological property data points,and the calculation speed is very slow.2)In the traditional phase-controlled stochastic modeling,the traditional Kriging method is still used in the interpolation.The Kriging equation constructed by this method only considers the influencing factors of the known data points from the same seismic phase when interpolating the simulated points,does not consider the factors affecting the interpolation points by known data points from different seismic phases.We have reason to believe that even if the correlation between data points belonging to different seismic phases is less than the correlation between data points belonging to the same seismic phase,it is necessary to take this correlation into account.In order to solve the above problems,this paper introduces the sparse representation spectral clustering method in machine learning classification technology to cluster the seismic phase rapidly,and carry out stochastic modeling of geological attribute variables under the influence of different seismic phases.In the process of seismic phases classification,it is often necessary to deal with largescale geological attribute data,where the computational complexity is unacceptable.For this reason,based on the spectral clustering algorithm,a sparse representation spectrum clustering method based on real logging data is proposed in this paper,which can accelerate the calculation process.In general,the cost of accelerating clustering is the loss of clustering effect,but the sparse representation method based on real logging data proposed in this paper can reduce this loss to a certain extent.At the same time,in the process of stochastic modeling,this paper improves the traditional simple kriging equation to a certain extent,considering the correlation between the points belonging to different seismic phases in the equation,so that the modeling results can show the spatial geology structure more comprehensively.Seismic phase has important reference significance in geological analysis.In this paper,we try to quickly and accurately divide the seismic phases.When interpolating the unknown points,we consider the correlation between phase and phase as much as possible.The traditional phase-controlled stochastic modeling method has some improvements in phase division and interpolation in this paper.Our work has significance for geology exploration and geological reservoir predictions.