An efficiency-oriented physics-informed transfer learning framework for transient simulation across diverse properties and conditions in liquid energy pipelines

As pipeline-transported liquid energy has expanded to include hydrogen-based and bio-based fuels, achieving efficient and rapid simulations across diverse physical properties and conditions is essential for timely early warning of anomalies. However, conventional numerical computation is hindered by iterative parameter identification, precluding computational efficiency. Emerging physics-informed machine learning techniques necessitate excessive retraining whenever fluid properties or conditions change, which might take hours. Furthermore, they struggle to accurately capture the transient oscillatory characteristics induced by coupled trends of flowrate and pressure. To address these challenges, this study proposes an efficiency-oriented physics-informed transfer learning framework for accelerated transient simulation under diverse properties and conditions. First, a Trend-Aware Gating Physics-Informed Neural Network (TAG-PINN) is designed to characterize the trend synchrony and divergence of transient hydraulics by incorporating a trend-gating layer to quantify the dependency strength between flowrate and pressure. Subsequently, a Similarity and Physics-driven Transfer (SPT) learning framework is developed, which quantifies similarities between domain samples to achieve high-fidelity simulation while minimizing the data overhead. Moreover, the contribution of network layers is quantified to enable the selection of high-contribution layers for a cost-effective, high-precision adaptation. Pragmatic experiments demonstrate that TAG-PINN captures transient oscillations more accurately, significantly reducing flowrate and pressure simulation errors by 58% and 51%, respectively, over conventional PINNs. The proposed SPT framework reduces training time by one to two orders of magnitude, bringing fine-tuning time from the hour scale to the minute scale, thereby providing effective technical support for timely early warning and decision-making in industrial pipelines.