Abstract: In order to find a suitable prediction method for carbon emission of urban traffic, 7 indexes were selected as the influencing factors of carbon emission of urban traffic based on the STIRPAT model. The 7 indexes included total population, per capita GDP, vehicle ownership, carbon emission intensity, urbanization rate, passenger turnover, cargo turnover and etc.. The prediction models of GA-SVM, PSO-SVM and GS-SVM were established respectively.And the relevant indicators of traffic carbon emissions from 1995 to 2016 were taken as the basic data for case study. The results show that: compared with PSO-SVM and GS-SVM, the correlation coefficients of training set of GA-SVM increase by 2.74% and 1.07% respectively, and the correlation coefficients of test set of GA-SVM increase by 1.04% and 0.29% respectively. Compared with the other two prediction models, GA-SVM model has good learning and generalization ability, which indicates that GA-SVM model is more suitable for the prediction and analysis of urban traffic carbon emissions.

Key words: traffic and transportation engineering, carbon emissions, GA-SVM model, influence index

摘要： 为寻找合适的城市交通运输碳排放预测方法，基于STIRPAT模型，选取人口总量、人均GDP、机动车保有量、碳排放强度、城镇化率、旅客周转量和货物周转量等7项指标作为城市交通运输碳排放影响因素，分别建立基于遗传算法优化支持向量机、粒子群优化支持向量机、网格搜索优化支持向量机预测模型，并以1995—2016年交通运输碳排放相关指标作为基础数据做实例分析。结果表明：GA-SVM对比PSO-SVM与GS-SVM所得出训练集的相关系数分别增长了2.74%和1.07%，测试集的相关系数分别增长了1.04%和0.29%，较其它两种预测模型具有良好的学习和推广能力，说明GA-SVM模型更适合对城市交通碳排放进行预测分析。

关键词: 交通运输工程, 碳排放预测, GA-SVM模型, 影响指标