Abstract: Aiming to solve the deficiency of structural equation method in predicting the improvement effect of rail transit passenger satisfaction, Bayesian network method that was good at dealing with uncertainty problem was directly put forward. Machine learning and expertise were combined to build the Bayesian prediction network between passengers overall satisfaction and quality indicators. The absolute effect, relative effect as well as the elastic modulus and synergic relationship to judge improvement effect of the established predictive model were defined. The data of Wuhan rail passenger satisfaction survey was used to establish uncertainty Bayesian network and the improvement effect of “Transfer” and “Ticketing System” which mostly needed to be improved in priority was predicted and analyzed. The research results indicate that “Transfer” indicator has a better improvement effect; however, there is no synergistic effect between two indicators.

Key words: traffic and transportation engineering, rail transit, passenger satisfaction, improvement effect, uncertainty predict, Bayesian network

摘要： 为克服结构方程式法对轨道交通乘客满意度改善效果预测的不足，提出了直接采用擅长处理不确定性问题的贝叶斯网络方法。采用机器学习和专家经验相结合的方法构建了乘客总体满意度与质量指标间的贝叶斯网络预测模型。定义了改善方案预测的绝对效果、相对效果以及指标改善效果有效性判别的弹性系数和协同关系。利用武汉市轨道交通乘客满意度调查数据，建立了不确定性贝叶斯网络，预测、分析了最需要优先改善的“换乘接驳”和“售票系统”两个指标的改善效果。研究结果表明：“换乘接驳”指标的改善效果较好，但两个指标之间无协同效果。

关键词: 交通运输工程, 轨道交通, 乘客满意度, 改善效果, 不确定性预测, 贝叶斯网络