Rank Predictions of Internal Corrosion of Gathering Pipelines in a Natural Gas Field With a Multi-Kernel SVM Method

Internal CO₂/H₂S corrosion of gathering pipelines is a serious problem in natural gas plant. It is important for field engineers to assess the corrosion degree and control corrosion risk. A multi-kernel support-vector-machine (SVM) method is presented to rank internal corrosion of gathering pipelines according to the NACE RP-0775-91 standard. By considering the nonlinear indivisibility between data, we combined three kinds of kernels (linear kernel, polynomial kernel, and Gaussian kernel) into a multi-kernel SVM to rank the internal CO₂/H₂S corrosion of gathering pipelines. The method was applied to a natural gas field in northwest China. Corrosion data were collected and analyzed. The prediction accuracy of the multi-kernel SVM method for ranking CO₂/H₂S corrosion was 66%, which is higher than the results of the single-kernel SVM methods (linear kernel, polynomial kernel and Gaussian kernel), whose prediction accuracies are 50%, 48% and 54% respectively. These findings could help field engineers rank corrosion and reduce the corrosion risk.