摘要：The prevention of natural gas hydrates is critical to oil and gas flow assurance. The nucleation process of hydrates has always been a research hotspot, yet its randomness makes the induction time of hydrates difficult to predict. To address this issue, this paper uses a Noise Injection Target Autoencoder (NITAE) to augment data, followed by a GBRT model for predicting hydrate induction time. Finally, the gplearn method is employed to generate an empirical equation for the hydrate induction time. The GBRT model achieves an R2 of 0.9858, with an absolute error within +0.02, addressing poor prediction performance due to data scarcity. The gplearn-based empirical equation achieves an R2 of 0.8353, with an error within +20 %. These results provide a new direction for predicting the hydrate formation induction time in actual field conditions and the prevention of hydrate formation in oil and gas pipelines.

关键字：Driving force; Hydrate induction time; Oil-water emulsion; Small sample; Data augmentation; Machine learning

ISSN号：0009-2509

卷、期、页：卷307

发表日期：2025-03-15

期刊分区（SCI为中科院分区）：二区

收录情况：SCI（科学引文索引印刷版）,EI（工程索引）,SCIE（科学引文索引网络版）

发表期刊名称：CHEMICAL ENGINEERING SCIENCE

参与作者：吕晓方,陈坤凯,柳扬,彭明国,段继淼,王传硕,马千里,周诗岽,李晓燕

第一作者：史博会,宋尚飞

论文类型：期刊论文

论文概要：吕晓方,陈坤凯,柳扬,彭明国,段继淼,王传硕,马千里,周诗岽,李晓燕,史博会,宋尚飞，Intelligent prediction of hydrate induction time in oil-water emulsion system based on data-driven and driving force，CHEMICAL ENGINEERING SCIENCE，2025，卷307

论文题目：Intelligent prediction of hydrate induction time in oil-water emulsion system based on data-driven and driving force