Research On Lithofacies Identification Method Based On Residual Recurrent Neural Network

Lithofacies identification is an important task in geological exploration and energy development.The location,scale,and characteristics of oil and gas reservoirs can be determined by analyzing different lithofacies categories,providing useful guidance for their development.Traditional lithofacies identification methods obtain lithofacies information by quantitatively or qualitatively interpreting initial parameters,which suffer from problems such as strong subjectivity,high cost,and low efficiency.Data mining-based machine learning,especially deep learning algorithms,has become a hot technology for lithofacies identification.It plays an important role in accurately establishing the nonlinear correspondence between characterization targets and reservoir parameters.In this paper,based on deep learning technology,with shale reservoirs as the research object,a lithofacies identification method based on residual recurrent neural network is proposed for shale reservoirs,achieving intelligent identification of unconventional shale reservoir lithofacies.The main research contents are as follows.(1)To address problems such as data heterogeneity,noise,and outliers in the initial dataset,preprocessing work such as well logging attribute analysis,well logging parameter sensitivity analysis,and standardization processing are carried out on the well logging data in the work area,which can provide a high-quality dataset for the research on lithofacies identification based on machine learning.(2)Decision tree,random forest,K-nearest neighbor,and support vector machine algorithms are used to construct lithofacies identification models for heterogeneous reservoirs.Evaluation metrics such as accuracy,recall,F1 score are applied to verify and evaluate the model.The limitations of traditional machine learning methods in lithofacies identification for heterogeneous reservoirs are analyzed.(3)Develop an intelligent identification model for heterogeneous reservoir lithofacies with residual connection technology.First,based on long short-term memory networks,a sequence feature module is constructed to extract the temporal information of parameter features in logging data.The multi-layer stacking of this module further enhances the model’s ability to extract key feature information.Secondly,residual connection technology is introduced on the basis of the sequence feature module to solve the degradation problem of conventional networks,and to capture important parameter features at different levels in the stacked network,and obtain comprehensive logging information.Finally,the model is applied to shale reservoirs,and the method is evaluated and analyzed in terms of lithofacies identification accuracy,recall rate,and other aspects.(4)Based on the research content,a reservoir lithofacies intelligent identification visualization system is developed by QT Designer.The system incorporates human-computer interaction functions such as parameter optimization,parameter settings,and reservoir lithofacies identification.It improves the operability,maintainability,and intelligence level of the lithofacies identification system.