Logging Evaluation Of Basic Parameters For Unconventional Reservoir Based On The Nonlinear Learning Theory

Nonlinear learning theories are applied to evaluate basic parameters ofunconventional reservoir in this dissertation, Logging prediction evaluation for coalstructure,coal-bed gas content, shale reservoir lithology and TOC is studyed. Based onthe logging response of coalbed methane and shale gas reservoir, the relations betweencoal structure,coal-bed gas content,shale reservoir lithology, TOC and logging areanalyzed. SVC and BP model are established to predict coal structure of studying area.BP and multiple regression model are built to predict coal-bed gas contents. SupportVector Machine lithology prediction model based on the optimal structure is establishedby cross validation and grid search algorithm. Using punishment parameter c andadding small sample weights is proposed for the imbalance samples. The predictionmodel of Total Organic Carbon content are established through the improved BP andSVM algorithm.and error analysis are carried out. The results show that nonlineartheories have fortissimo nonlinear approximation capability. It can reflect relationshipveritably between unconventional reservoir and log parameters. There is close tolerancebetween predicting outcome and measured value, it turn out to a good application effect.