为解决成都地铁设计和修建过程中碳排放计量问题，以成都地铁18号线6车站7区间为研究对象，采用机器学习算法对成都地铁建设阶段碳排放进行预测研究。基于生命周期评价（life cycle assessment, LCA）框架对地铁车站和盾构区间建筑材料生产阶段、建筑材料运输阶段和现场施工阶段温室气体排放量进行计算，建立基于鲸鱼优化算法（whale optimization algorithm, WOA)的深度极限学习机（deep extreme learning machine, DELM）地铁碳排放预测模型，并与基于风驱动优化（wind driven optimizer, WDO）、灰狼优化(grey wolf optimizer, GWO）、粒子群优化(particle swarm optimizer, PSO)、人工蜂群优化（artificial bee colony, ABC）、多元宇宙优化（multi-verse optimizer, MVO）、原子搜索优化（atom search optimizer，ASO）的深度极限学习机（DELM）和未优化的BP（back propagation neural network）、KELM(kernel extreme learning machine)、DELM算法预测结果进行对比分析。研究得到： 1)WOA-DELM算法预测结果相关一致性为0.757，略高于其他算法； 2）根据WOA-DELM算法对地铁碳排放主要输入指标进行敏感性分析，得到地铁车站碳排放预测的关键影响因素为车站长度和轨面埋深，对应指标碳排放相对变化率分别为30.1%和23.1%。

To achieve low carbon emission and reduction in the design and construction of Chengdu metro, the machine learning algorithm is employed to predict the carbon emissions in the construction phase of the Chengdu metro based on a case study of 6 stations and 7 sections of the Chengdu metro line 18. The greenhouse gas emissions from building material production, transportation, and onsite construction in metro stations and shield sections are then calculated using a life cycle assessment method, and a metro carbon emission prediction model based on deep extreme learning machine (DELM) optimized by whale optimization algorithm (WOA) is established. The prediction results of DELM algorithms optimized by the winddriven optimizer, grey wolf optimizer, particle swarm optimizer, artificial bee colony, multiverse optimizer, and atom search optimizer are compared with those of nonoptimized back propagation neural network, kernel extreme learning machine, and DELM algorithms. The research results reveal the following: (1) The correlation consistency of WOADELM algorithm prediction results is 0.757, slightly higher than other algorithm models. (2) The sensitivity analysis of the leading input indicators of metro carbon emission is conducted using the WOADELM 〖HJ〗algorithm model, obtaining the key influencing factors of station carbon emission prediction, i.e., station length and rail surface buried depth, and the relative change rates of corresponding indicators of carbon emission are 30.1% and 23.1%, respectively.