An ensemble model based on relevance vector machine and multi-objective salp swarm algorithm for predicting burst pressure of corroded pipelines

Abstract

Burst pressure is the key to the design of pressure pipelines, and its accurate prediction is of considerable significance to pipeline safety and integrity management. Due to the different corrosion forms of different pipelines, the traditional empirical formula is difficult to predict the burst pressure with high accuracy. This paper proposes an ensemble data-driven model that can train the model based on existing pipeline parameters, corrosion characteristics, and burst pressure data. The trained model can be used to predict the burst pressure of pipelines with other pipeline parameters and corrosion characteristics. In the ensemble model design, a more advanced machine learning model, relevance vector machine, is used as the basic predictor. Multi-objective salp swarm algorithm is utilized to optimize the relevance vector machine, to improve the prediction accuracy and stability. The proposed model is applied to three datasets with different features and inputs. The results reveal that the average absolute percent errors of the proposed model in the three datasets are 3.724%, 1.788%, and 4.133%, respectively, and the standard deviations of the percentage errors are 9.07%, 2.36%, and 5.42%, respectively, which are all in a low level. The results indicate that the proposed model has high prediction accuracy and stability in different datasets, even when the data amount is small. Moreover, the Diebold-Mariano test implies that the proposed model's prediction performance is better than benchmark models, and much better than the model without the optimizer or with the single-objective optimizer.