Research On Reservoir Prediction Method Based On Probability Graph Model

Oil and gas reservoir prediction technology is a technology that comprehensively applies seismic,geological,drilling,logging and other data to predict the distribution,thickness,lithology and physical properties of underground oil and gas reservoirs.The main work contents of reservoir prediction are roughly divided into reservoir lithology prediction,reservoir morphology prediction,reservoir physical property prediction and comprehensive analysis of reservoir hydrocarbon content.Porosity is an important indicator reflecting reservoir oil and gas reserves,and is reservoir physical property.An important research content of the forecast.At present,the porosity prediction uses the impedance,density and other information obtained by seismic inversion and the porosity information obtained from the log to be regression,and then applies the relationship to the whole volume data.However,there are many problems in the practical application of this method,which are mainly reflected in: 1)the relationship between the porosity of different reservoirs and the elastic parameters is different.If the elastic parameters of all lithologies are related to the porosity,it may be There are large errors in the results;2)less well data may lead to weaker generalization of regression relations,resulting in unsatisfactory results;3)there may be errors in the elastic parameters obtained from seismic data,resulting in errors in prediction results.In order to solve the above problems,based on the conditional random field theory in the probability map model,this paper constructs a simultaneous prediction network of facies and porosity,and based on this,the following research work is carried out:1)Through the label pair of impedance and porosity,a porosity-rock facies synchronization prediction network with conditional random field and gradient lifting tree fusion is established.Through the conditional random field,the lithofacies iterative update under the joint control of porosity and impedance is realized.Under the control of petrographic facies,the gradient generation tree is used to solve the regression problem in the case of less well data sample labels.The fusion network is implemented in relatively few cases.Synchronous prediction of lithofacies and porosity under sample conditions.2)A network is constructed that directly predicts porosity and facies through waveform data.Considering that the impedance data may have errors,we replace the impedance with the seismic waveform and introduce the aforementioned conditional random field gradient generation tree fusion network model for porosity and facies prediction.Due to the multi-dimensional characteristics of seismic waveform data,we extract the texture features of seismic waveforms and use the gradient generation tree to reconstruct the features,and construct a new feature classification method.The classification results are used as the initial model input,which effectively improves the prediction of porosity and facies.The accuracy of the network.3)A conditional random field prediction method based on seismic waveform characteristics is proposed.The seismic waveform features are dimensionally reduced,and the features and porosity are reduced.The label is introduced into the gradient generation tree conditional random field.Network to achieve simultaneous prediction of seismic facies and porosity.The experimental results show that the use of seismic waveforms for porosity and facies prediction can effectively avoid the errors introduced by the calculation of impedance data and improve the prediction effect.Through the research of this thesis,it is proposed to explore a feasible technical solution for predicting reservoir lithofacies and reservoir physical parameters through artificial intelligence method,which provides a reference for promoting the combination of artificial intelligence technology and reservoir prediction.