基于随机重复爬山法的交通状态预测

Traffic Status Prediction Based on Random Restart Hill-climbing

合理构造影响交通状态网络结构,是实现交通状态预测的前提条件.为克服爬山法易陷入局部最优的缺陷,提出一种基于随机重复爬山法的交通状态预测方法.对随机生成的有向无环图迭代运行爬山法得到多网络结构;通过有向边置信度的定义和置信度阈值的计算,确定了最优贝叶斯网络结构中节点和有向边选取准则;利用最优贝叶斯网络结构,实现了畅通、平稳、拥挤和阻塞等4种交通状态的预测并综合评价.分析结果表明,该方法仅选取时段、节假日等两变量时,对交通状态预测总体准确率超过85%,能够为高速公路运行状态监测预警和决策分析提供有效方法和数据支撑.

Construct of reasonable network structure which influencing traffic status is the prerequisite of realizing traffic status prediction. In order to improve Hill-climbing algorithm, which may trap into the local optimum instead of the global optimum, a new traffic status prediction method is proposed based on Random Restart Hill-climbing. Proposed multi-network structures are obtained by executing Hill-climbing algorithm iteratively, to create directed acyclic graphs randomly. Furthermore, selection criterion for nodes and directed edges in the optimal Bayesian network structure is determined by the definition of directed edges-confidence and the calculation of confidence-threshold. The intelligent predictions and comprehensive evaluations of four kinds of traffic status including free, smooth, congestion and jam are achieved by using optimal Bayesian network structure. Results indicate that the prediction results are satisfactory with a high accuracyrate of more than 85% only selecting two variables such as hour and holiday. Therefore, the proposed method provides an effective way and experimental proof for monitoring, warning and decision analysis of expressway operation status.