Research On Intelligent Reservoir Prediction Method Under Small Sample Condition

In the field of oil and gas,reservoir information can reflect the favorable accumulation location of underground oil and gas.The traditional reservoir prediction method relies heavily on the experience of experts and is inefficient,so this thesis will use some artificial intelligence algorithms to solve these defects and realize the intelligent prediction of reservoirs.However,most of the existing artificial intelligence algorithms require abundant training samples.Due to the high cost of obtaining reservoir samples,the model is usually over fitted.Therefore,this thesis will study the problem of intelligent reservoir prediction based on the condition of small samples.Reservoir prediction mainly includes seismic facies prediction and porosity prediction.This thesis will study the above two aspects,including the following three points:(1)Build a lightweight network model to achieve seismic facies prediction under the condition of small samples.In this thesis,the sliding window is used to maximize the use of the limited seismic facies label information;then a better initial solution is provided for the network model through transfer learning;finally,the lightweight network model is used to solve the seismic facies prediction problem under the condition of small samples.(2)Based on the idea of self-supervised learning,the constraint of unlabeled data is added to the seismic facies prediction.In this thesis,the network model has the preliminary derivation ability of the work area data through the self-supervised learning method based on context generation.The specific implementation process is as follows:the model trains the auxiliary network to reconstruct the missing part of the seismic data through a large amount of unlabeled seismic data.Then,by transferring the training weight parameters of the auxiliary network,the downstream seismic facies identification network model can adapt to the difficulty of small sample conditions.(3)Based on the idea of semi-supervised learning,the problem of predicting porosity from seismic elastic data under the condition of small samples is realized.In this thesis,the network model is trained by combining a small amount of logging label information next to the well and a large amount of unlabeled seismic elastic data outside the well,thereby improving the accuracy of porosity prediction and reducing the dependence on label data.At the end,the thesis also conducts facies-controlled porosity prediction research on the basis of seismic facies prediction results.This thesis finally realizes the seismic facies prediction under the condition of small samples by improving the network structure and introducing the idea of self-supervised learning,and then completes the facies-controlled porosity prediction research on the basis of the seismic facies prediction results.Through the application of the actual work area,the method in this thesis can effectively improve the accuracy of the inversion.