Novel method for residual strength prediction of defective pipelines based on HTLBO-DELM model

Residual strength prediction of defective pipelines is critical to pipeline reliability assessment, which can affect the remaining useful life of pipelines. In this paper, we propose a novel method for predicting residual strength of defective pipelines based on deep extreme learning machine (DELM). To obtain the high accuracy, the hybrid teaching-learning-based optimization (HTLBO) algorithm with multiple adjustment strategies is designed to improve the DELM model. The experimental data of pipeline burst pressure is selected for the training and validation of proposed method. The interactions of input parameters on residual strength are investigated using response surface method. After comparisons of key model parameters, the optimal model is established to ensure the prediction accuracy. Through the validation of benchmark functions, the HTLBO performs well in convergence and optimization performance. The prediction results show that the proposed method has higher precision than other models, and it can predict the residual strength within the relative error of 6%. This study can provide a basis for reliability engineering and transportation safety of defective pipelines.