文献类型：期刊文献

英文题名：Physical field distribution-driven scenario reduction for gas sensor layout optimization within hydrogen-fueled gas turbine enclosures

作者：Feng, Yu[1];Lang, Ziqiang[1,2,3];Jin, Xisheng[4];Wang, Bing[1];Cao, Chenxi[1,2]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Engn Res Ctr Proc Syst Engn, Minist Educ, Shanghai, Peoples R China;[3]Univ Sheffield, Dept Automat Control & Syst Engn, Sheffield S1 3JD, England;[4]East China Univ Sci & Technol, Sch Chem Engn, Key Lab Green Chem Engn & Ind Catalysis, Shanghai 200237, Peoples R China

年份：2026

卷号：214

外文期刊名：PROCESS SAFETY AND ENVIRONMENTAL PROTECTION

收录：;EI(收录号：20262420884304);WOS:【SCI-EXPANDED(收录号:WOS:001797741800001)】；

基金：The work was supported by the National Science and Technology Major Special Project (2025ZD1607100), National Natural Science Foundation of China (62394343, 62394345), the Programme of Introducing Talents of Discipline to Universities (the 111 Project) under Grant B17017, Natural Science Foundation of Shanghai (24ZR1414900) and Fundamental Research Funds for the Central Universities.

语种：英文

外文关键词：Gas turbine; Hydrogen leakage; Scenario reduction; 3D convolutional neural network; Stochastic programming; CFD

摘要：Hydrogen leakage is a critical safety concern for hydrogen-fueled gas turbines. Optimizing a gas detector network essential for early warning and risk mitigation inside turbine enclosures necessitates the consideration of numerous uncertain leakage scenarios, which results in substantial computational cost. However, existing scenario reduction strategies for detector placement are mainly based on source parameters or scenario probabilities and do not explicitly preserve the concentration-field morphology most relevant to sensor deployment in congested enclosures. An integrated framework is proposed that reformulates scenario reduction for gas sensor layout optimization as a concentration-field-aware representation problem and couples physical field distribution-driven scenario reduction with multi-objective stochastic optimization for effective sensor placement. A three-dimensional convolutional neural network equipped with a spatial attention mechanism is employed to encode leakage concentration fields into compact features that preserve plume morphology. These features are then fused with leakage source parameters to generate a small set of representative scenarios with associated probability weights. Stochastic optimization based on only 8 physical field-induced typical scenarios, reduced from a full ensemble of 183 scenarios, produced sensor layouts consistent with those obtained using the full scenario set. The reduced scenario set preserved the weighted concentration field relative to the full ensemble, while achieving a 14.6-fold speed-up for the joint objective.