基于MLE-LM算法估计的交通流断面速度Weibull分布模型

摘要

For the problems that one-sided hypothetical relations,incomplete regression data and fitting models lacking of universality and accuracy in the traditional velocity distribution of section traffic flow,the paper uses the bisecting K-FCM clustering algorithm to cluster partition the large sample traffic flow data of velocity,flow and occupancy.Then,MLE-LM algorithm model is established to fit three parameters Weibull distribution of section velocity,and compares its fitting results with the normal distribution and multi-peaks gaussian fitting.Such an algorithm model is featured with fast convergence,high precision of fitting results and simple steps of solutions,and it is not easy to converge in local extremum and sample capacity has the strong adaptive capacity.The results show that non-parametric estimation of multi-peaks Gaussian has the best fitting effect,but its degree of freedom is smaller than the parameter estimation.That means the variables is limited,and it is difficult to reflect the distribution characteristics of the expressed data.The fitting results of Weibull-3 distribution are all greater than 0.7,which is better than the normal distribution.The overall results show that the velocity distribution obeys the Weibull-3 distribution in different traffic flow.%针对传统断面交通流速度分布假设关系较片面、回归数据不全面和拟合模型不具有普遍性和精确性等问题,本文采用二分K-FCM结合算法对连续收集的大样本交通流数据中速度、交通量和占有率三者进行聚类划分,再利用自主建立的MLE-LM算法拟合运行车速三参数Weibull分布,并将其拟合结果与正态分布和高斯多峰拟合结果对比.该算法模型收敛快,拟合结果精度高,不易收敛于局部极值,且对样本容量的适应能力强.研究结果表明:各交通流状态下高斯多峰非参数估计拟合效果最好,但其自由度最小,即被限制的变量最多,难以反映数据的分布特征;而参数估计中Weibull-3分布拟合结果均比正态分布拟合结果好,且Adj R-square值都大于0.7,故总体认为不同交通流状态下速度分布基本服从Weibull-3分布.