摘要：Leakage from buried hydrogen-blended natural gas (HBNG) pipelines poses significant risks to hydrogen energy infrastructure. Existing pipeline leakage localization methods, which are considered classification tasks or reliant on fixed sensor layouts, struggle to achieve high accuracy in buried HBNG pipelines under sparse and constrained sensor deployment. Therefore, an end-to-end deep learning network, which achieves adaptive positioning of arbitrary sensor configurations through multiscale spatiotemporal feature fusion, is presented. The proposed network employs multiscale convolutions for temporal concentration evolution characteristics extraction, a bidirectional long-short term memory network for long-term time-series dependence modeling, an enhanced position coding mechanism for arbitrarily distributed monitoring point configuration, and a self-attention mechanism for key monitoring points with significant contribution to dynamic localization identification. The experimental results reveal that in the X and Y directions, the coefficients of determination are 0.9992 and 0.9756, respectively, indicating higher positioning accuracy than that of other advanced methods.

关键字：Hydrogen leakage; Leakage location; Deep learning; Spatiotemporal feature fusion

ISSN号：0360-3199

卷、期、页：卷222

发表日期：2026-03-31

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：INTERNATIONAL JOURNAL OF HYDROGEN ENERGY

参与作者：任杰

通讯作者：李佳美

第一作者：陈林,刘衡,荆琦

论文类型：期刊论文

论文概要：陈林,李佳美,任杰,刘衡,荆琦，End-to-end leakage localization in buried hydrogen-blended natural gas pipelines based on multiscale spatiotemporal feature fusion，INTERNATIONAL JOURNAL OF HYDROGEN ENERGY，2026，卷222

论文题目：End-to-end leakage localization in buried hydrogen-blended natural gas pipelines based on multiscale spatiotemporal feature fusion