Stress prediction model of oil and gas pipeline based on magnetic-force coupling and machine learning

Abstract

This study proposes a novel machine learning-based model for accurate stress monitoring in gas pipelines, essential for ensuring safe and efficient operations. The model leverages data on stress and magnetic flux density derived from parameters such as pipe external diameter, wall thickness, elastic modulus, relative permeability, and structural characteristics, combined with internal pressure loads and environmental magnetic fields, obtained via a magnetic simulation platform. So, the data mainly comes from the magnetic simulation platform, but also includes some model tests and the accumulation of measured data of the study team. Data preprocessing included missing value imputation, outlier processing, and normalization. A 3D parallel computing machine learning model was developed to predict stress, incorporating pipeline structural parameters, internal pressure loads, and magnetic parameters. Model parameters were optimized using a grid search method with 50 % cross-validation, and performance was evaluated using R², RMSE, and MAE metrics. Among RF, SVR, CART, and ANN algorithms, Random Forest (RF) performed best, achieving R² = 0.87, RMSE = 0.045, MAE = 0.01 for stress prediction, and R² = 0.97, RMSE = 0.05, MAE = 0.02 for magnetic flux density prediction. Comparisons with finite element method calculations across 12 pipeline parameter sets showed a maximum accuracy value error is within 6 %. The model’s robustness allows accurate predictions even with incomplete data, enabling non-excavation stress assessment using design data, field surveys, and tests. This provides valuable insights for pipeline lifecycle management and preventive maintenance, offering effective technical support for stress monitoring in oil and gas pipelines.