Predicting Solid-Particle Erosion Rate of Pipelines Using Support Vector Machine with Improved Sparrow Search Algorithm

The erosion of elbows by particles in shale gas pipelines poses a great threat to the production safety of pipelines. Accurate prediction of the erosion rate is critical to ensuring the safe operation of pipelines. Previous predictions of the erosion rate is based on experiments and numerical simulation. However, because many factors of affecting pipeline erosion lead to strong nonlinearity of the physical scene, these two methods are time-consuming and labor-intensive and have many assumptions. This paper presents a data-driven approach for efficiently predicting erosion rates in shale gas pipelines, which has a high prediction accuracy based on an amount of data without experiments and numerical simulation. The model mixes the adaptive t-distribution-based sparrow search algorithm and support vector machine. Through case studies, the following results were obtained: (1) the minimum mean square error (MSE) of the proposed model is less than 10%; (2) compared with the hybrid model without considering the adaptive t-distribution, the MSE of the proposed model is reduced by 32%; and (3) the MSE, Theil U statistic 1 (U1), and Theil U statistic 2 (U2) of the proposed model in the test set are 47%, 47%, and 34% lower than the average MSE, Theil U statistic 1 (U1), and Theil U statistic 2 (U2) of benchmark models, respectively. Two cases of multiphase flow and gas–solid conditions showed that the improved hybrid model has higher prediction accuracy than support vector machines with other optimizers and has strong generalization performance. This study may be helpful for pipeline maintenance and repair in practical engineering.