Post-dryout heat transfer is critical for the thermal management of boiling water reactors (BWRs) and fourth-generation supercritical water reactors (SCWRs) operating under subcritical, transcritical, and accident conditions. The acquisition of experimental data for post-dryout heat transfer in high-pressure supercritical environments poses significant challenges, resulting in a scarcity of available data. This study presents a novel approach to investigate post-dryout heat transfer through a sample expansion strategy that integrates neural networks and noise error mechanisms. The existing datasets of supercritical water and R134a under post-dryout conditions are expanded by introducing controlled noise. The augmented dataset is subsequently employed to train and optimize a neural network model for predicting post-dryout heat transfer. And this research analyzes the impact of noise arising from experimental errors on data fitting. This research offers a methodology to advance the understanding and application of post-dryout heat transfer, aiming for more precise predictions.