A physics-informed neural network-based computational method for forward and inverse analysis of buried pipelines subjected to lateral landslides

Buried oil and gas pipelines in mountainous regions are frequently subjected to transverse landslide loading, necessitating accurate forward mechanical prediction and reliable inverse load identification for structural safety evaluation. Conventional finite element methods serve as the standard tool for forward analysis of such structures, yet they inherently lack the capability to assimilate sparse field monitoring data—a critical prerequisite for quantifying unknown landslide-induced loads. Although physics-informed neural networks (PINNs) offer unified forward–inverse modeling capacity, their practical application to large-scale pipeline problems is limited by inherent spectral bias, stringent strong-form continuity requirements, and automatic differentiation overhead. This study proposes a domain-decomposed energy-based PINN framework for Winkler-supported pipeline beams, adopting single-hidden-layer networks with cubic Hermite interpolation, analytical differentiation, and a weak-form energy loss functional. The established framework supports both single-stage static load identification via bilevel optimization and Tikhonov regularization, as well as two-stage time-varying load evolution tracking through dual-network temporal coupling. Benefiting from the unified PINN architecture, heterogeneous monitoring data including deflection, rotation, and strain measurements can be flexibly fused across staged monitoring procedures. Forward validation against ABAQUS benchmarks under both cosine and asymmetric ridge load distributions yields relative errors below 0.2% for deflection, rotation, bending moment, and strain. Inverse tests for both load patterns demonstrate that deflection-based monitoring achieves high-accuracy load reconstruction (RMSE <80 kN / m ) with five or more sensors, and two-stage incremental monitoring effectively refines time-varying load amplitude estimation, albeit constrained by the baseline accuracy of initial sparse inversion. Further validation with 5% peak-amplitude Gaussian noise shows that the load identification accuracy and the inherent sensitivity hierarchy of structural responses are well preserved, demonstrating the favorable noise robustness and reliability of the proposed framework. This work presents an accurate, robust, and efficient physics-informed paradigm for geohazard monitoring and safety assessment of buried pipeline infrastructures.