Abstract: Given the complex deployment scenarios and dynamic operating conditions of rapidly deployable fuel pipelines, existing predictive maintenance and health management approaches suffer from limitations such as insufficient prediction accuracy under small-sample conditions, weak generalization capabilities, and high operational costs. To address these issues, this study investigates a mechanism-data hybrid-driven intelligent prediction method. We develop an edge-cloud collaborative system architecture tailored for rapid deployment scenarios and propose a hybrid prediction algorithm integrating corrosion mechanism constraints with dynamic bias long-term memory networks. By incorporating physical prior knowledge through a hybrid loss function, we alleviate the model generalization bottleneck in small-sample scenarios. Validation is conducted using 30 months of real-world measurement data collected from a rapidly deployable fuel pipeline. Results demonstrate that the proposed model achieves a root mean square error (RMSE) of 0.012 ± 0.001 and a mean absolute error (MAE) of 0.009 ± 0.001, improving prediction accuracy by 85.9% and 62.5% over pure mechanism-based models and long-term-memory network models, respectively. Furthermore, the proposed model exhibits significantly lower accuracy degradation under varying training dataset proportions than comparable methods, maintaining over 90% prediction accuracy in complex deployment environments while reducing overall operational costs by more than 40%. This study provides a technical framework for intelligent operation and maintenance of rapidly deployable fuel pipelines and offers valuable insights for health management of similar temporary industrial facilities.