时间价值分布预测的混合密度模型.PDF

第22 卷 第4 期 中 国 公 路 学 报 Vol. 22 No. 4 2009 年7 月 China Journal of Highw ay and T ransport July 2009 : 1001-7 72( 2009) 04-0096- 06 俞 礼 军 ( , 510641) : 针对使用常规简单分布密度函数难以拟 复杂时间价值数据的问题, 提出了基于混 分布的 建模方法, 采用期望最大化( EM) 算法推导出了Weibull 混 分布参数估计的迭代算式, 并将其用 于时间价值数据的拟 为考察混 分布模型对复杂数据分布的拟 性能, 进行了仿真对比研究, 并将基于EM 算法的Weibull 混 分布时间价值模型应用于城市客运交通方式分担率的分析和计 算结果表明: 基于EM 算法的混 分布模型是最有效的; Weibull 混 分布模型能够有效用于复 杂时间价值数据统计分布建模和交通方式分担率预测研究 : 交通工程; 时间价值; 混 模型; EM 算法; Weibull 分布; 参数估计 : U 491. 1 : A Finite Weibull Mixture Di stributi on Model of VOT Distribution Forecasting YU L-i jun ( School of Civil Engineering and Transportation, South China University of Technology , Guangzhou 510641, Guangdong, China) Abstract: Aimed at problem that data of complicated value of time ( VOT ) can not be fitted by simple distribution mixture functions frequently , the finite Weibull mixture distribution model for modelling VOT distribution w as proposed. Expectation maximum ( EM) algorithm w as used to deduce the interative formulae of the model and the formulae w ere applied to fitting of VOT . In order to explore fitting characteristics of the model on complicated data distribution, simulated comparisons w ere carried out and the model based on EM algorithm w as applied to analysis and calculation of urban passenger traffic mode sharing ratio. Results show that the model based on the EM algorithm is the optimal choice. The model can be effectively applied to VOT distribution modelling and forecasting of traffic mode sharing ratio. Key words: traffic engineering; value of time; mixture model; EM algorithm; Weibu