基于CSSOA-DSRF模型的致密砂岩储层流体测井智能识别OA

Reservoir fluid identification is of great significance for the evaluation and development of tight sandstone oil and gas reservoirs.Tight sandstone reservoirs have characteristics such as low porosity,low permeability,and strong heterogeneity,which result in complex gas-water relationships.Traditional reservoir fluid identification methods mainly rely on data from resistivity logging and other techniques,making it difficult to identify reservoir fluids with weak conductivity contrast.With the development of machine learning and artificial intelligence technologies,the integration of well logging techniques and intelligent algorithms has played a key role in fluid identification.However,traditional machine learning models lack the ability to distinguish highly redundant and imbalanced samples,limiting their prediction capabilities.This paper proposes an intelligent well logging fluid identification model for tight sandstone reservoirs based on the chaos sparrow search optimization algorithm-double cost sensitive random forest(CSSOA-DSRF)model.The double cost sensitive random forest(DSRF)introduces cost-sensitive learning during both the feature selection and ensemble voting stages of the random forest algorithm.By assigning weight coefficients to different fluid types,it enhances the model's focus on minority class samples,making feature selection more targeted and selecting a set of decision trees that are more sensitive to minority class data,thus solving the issue of sample class imbalance.To overcome the limitations of traditional optimization methods that easily fall into local optima,the chaos sparrow search optimization algorithm(CSSOA)incorporates an improved Tent chaotic mapping and Gaussian mutation mechanism within the framework of the sparrow search algorithm(SSA),enhancing population diversity and global search capabilities,and reducing the risk of premature convergence.The model combines five well logging response feature curves from the study area:acoustic time-difference logging,compensated neutron logging,density logging,natural gamma logging,and deep lateral resistivity logging,to output the corresponding fluid type prediction results.The prediction accuracy based on the borehole results is 90.46%.Compared with DSRF,random forest(RF),K-nearest neighbors(KNN),and support vector machine(SVM),the method demonstrates high accuracy and maintains good robustness and stability,providing a feasible solution for fluid identification in tight sandstone reservoirs.