With the mass construction of urban subways, the global greenhouse gas (GHG) emissions have been on the rise. This paper provides statistical evidence to support the infrastructure of subway emissions reduction through a study of GHG emissions during the construction stage of 6 stations and 7 sections of Chengdu Metro Line 18. Using the emission coefficient method, the GHG emissions from building material production, transportation and site construction in subway stations and shield sections were calculated, and a subway GHG emissions prediction model dependent on deep extreme learning machine (DELM) with whale optimization algorithm (WOA) was established(i.e., WOA-DELM). Compared with some optimized DELMs, namely wind driven optimizer (WDO) -DELM, grey wolf optimizer (GWO) -DELM, particle swarm optimizer (PSO) -DELM, artificial bee colony (ABC) -DELM, multi verse optimizer (MVO) -DELM, and atom search optimizer (ASO) -DELM, and some non-optimized algorithm models, namely back propagation neural network (BPNN), kernel extreme learning machine (KELM) and DELM, the correlation consistency of WOA-DELM algorithm prediction results (0.757) was found to be slightly higher. Through sensitivity analysis of the main input variables of subway GHG emissions with the WOA-DELM algorithm model, it was determined that the key influencing factors of station GHG emissions prediction were the station length and the depth of track surface, with relative change rates of corresponding variables of GHG emissions at 30.1% and 23.1% respectively. Finally, a rough prediction formula of GHG emissions from Chengdu Metro stations and shield sections were fitted based on the key influencing factors of GHG emissions. This study provides a practical and effective reference for reducing GHG emissions in subway construction and operation.

随着城市地铁的大规模建设，全球温室气体（GHG）排放量不断上升。本文通过对成都地铁18号线6站7路段施工期温室气体排放研究，为支持地铁基础设施减排提供统计依据。计算了地铁车站和盾构路段的场地建设情况，建立了基于深度极限学习机（DELM）和鲸鱼优化算法（WOA）的地铁温室气体排放预测模型（即WOA-DELM）。与一些优化的DELM相比，即风驱动优化器（WDO）-DELM，灰狼优化器（GWO）-DELM，粒子群优化器（PSO）-DELM，人工蜂群（ABC）-DELM、多节优化器（MVO）-DELM、原子搜索优化器（ASO）-DELM，以及一些非优化算法模型，即反向传播神经网络（BPNN）、核极限学习机（ KELM）和DELM，发现WOA-DELM算法预测结果（0.757）的相关一致性略高。通过WOA-DELM算法模型对地铁温室气体排放主要输入变量进行敏感性分析，确定车站温室气体排放预测的关键影响因素为车站长度和轨道面深度，相应的相对变化率温室气体排放量的变量分别为 30.1% 和 23.1%。最后，基于温室气体排放关键影响因素，拟合出成都地铁车站及盾构段温室气体排放粗略预测公式。该研究为减少地铁建设和运营过程中的温室气体排放提供了切实有效的参考。