A physics-informed neural network method for predicting the remaining life of energy pipelines

The safe and stable operation of energy pipelines is directly related to energy supply security and public security. Accurate remaining service life (RSL) prediction for energy pipelines is critical to improving transmission system safety, reducing operational risks, and maintaining stable, efficient operation of the energy supply chain. However, conventional methods struggle with complex corrosion environments, limited data, poor physical consistency, and low interpretability, limiting their practical use in pipeline operation and maintenance. This paper proposes a physics-informed neural networks prediction method synergistically guided by physical consistency and feature confidence. The method uses a convolutional residual neural network as the base architecture and introduces a semi-empirical physical consistency loss function based on normative corrosion mechanisms. By constraining the model through learnable physical parameters, it ensures that both the prediction trends and numerical magnitudes conform to the physical laws of corrosion. Meanwhile, a feature confidence guidance mechanism based on gradient sensitivity is integrated to implement directional constraints on the model's dependence on key corrosion features, enhancing the interpretability of the model's feature-based decision-making. Research results indicate that compared to conventional models, the proposed model achieves a mean absolute percentage error of 9.3% and a coefficient of determination of 0.94, representing the best prediction performance. Furthermore, the model output and physical feature constraints demonstrate excellent physical consistency and interpretability. This model can effectively provide reliable technical support for the life-cycle planning, condition-based maintenance scheduling, risk prevention, and efficient stable operation of integrated energy transmission systems, thereby improving the overall safety, economy, and sustainability of integrated energy systems planning and management.