Residual Strength Hybrid Prediction of Hydrogen-Blended Natural Gas Pipelines Based on Fem-Fc-Bp Model

Abstract

Utilizing existing natural gas pipelines for hydrogen transportation is deemed crucial in developing the hydrogen energy industry, thus necessitating urgent safety evaluations. However, current experimental research on hydrogen-blended natural gas pipelines confronts challenges like high cost and risk, making obtaining relevant experimental data difficult. In addition, the oil and gas pipeline field research also faces the problem of limited data features. To address these problems, this paper organically integrates the finite element method (FEM), feature crossing (FC), and backpropagation (BP) neural network, and proposes a residual strength hybrid prediction model of hydrogen-blended natural gas pipelines based on FEM-FC-BP. The model first uses the FEM to simulate a data set. Then, the FC is introduced to enhance the data feature reasonably. Finally, the optimal BP neural network is trained using the five-fold cross-validation method. Experiments on API SPEC 5L X52 pipeline steel show that the proposed FEM-FC-BP hybrid prediction model can effectively predict the residual strength of hydrogen-blended natural gas pipelines, in which FEM can effectively overcome the difficulty of data acquisition, and FC can reasonably enhance the data feature to improve the accuracy of the model. This research provides valuable references for the integrity management of hydrogen-blended natural gas pipelines.

Keywords: Hydrogen-blended natural gas, Finite element method, Feature crossing, BP neural network, Residual strength