Exploration of open rack vaporizers thermal performance based on dynamic evaluation of soft-sensing parameters

Open rack vaporizers (ORVs) use seawater to regasify liquefied natural gas (LNG), while their thermal performance varies substantially with operating conditions and may deteriorate under low-temperature seawater conditions. To characterize such dynamic behavior, this study proposes a mechanism–data-driven framework integrating physical heat-transfer relationships with dynamic soft-sensing parameter estimation. The logarithmic mean temperature difference and effective heat-transfer area are treated as operating-condition-dependent latent thermal parameters and estimated using a dual-critic actor–critic (DCAC) model. Generative adversarial network (GAN)-based augmentation is introduced to enrich the training space, while isolated training, validation, and test datasets are employed to prevent information leakage and assess model generalization. On the unseen industrial test dataset, the dynamic model achieves an RMSE of 0.106 °C and an MRE of 0.627% for LNG outlet temperature, while the MREs for LNG heat absorption, seawater outlet temperature, seawater heat release, and heat-transfer efficiency are 0.045%, 0.641%, 2.442%, and 2.420%, respectively, outperforming the static-parameter model. Engineering analysis reveals an efficiency–throughput trade-off, with a representative operating compromise at an LNG flow rate of approximately 406 m 3 ·h⁻ 1 and a heat-transfer efficiency of 72%. Low-temperature analysis indicates progressive icing-induced deterioration, intensified by insufficient seawater supply and high LNG throughput; sensitivity analysis supports a 2 °C icing-onset threshold as a conservative engineering criterion. Overall, the framework provides a physically interpretable and independently validated approach for dynamic ORV thermal-performance prediction, operating-state assessment, and reliability-aware optimization.