中文摘要：

短时交通流量预测，是交通系统信息化和智能化交通运输管理技术领域研究的关键问题．目前的方法对历史数据具有较高的依赖程度，或者具有较高的计算成本，或者不能有效反映实际中较复杂的交通网络及各结点之间的相互关系、以及依赖的不确定性，或者多种模型的组合使得预测方法较复杂．贝叶斯网是一种重要的概率图模型，本文以交通网络结构为基础，利用概率图模型在不确定性知识表示和推理方面的良好性质，考虑路口交通流量及其预测的时序依赖特征，构建了带有时序条件依赖关系的交通贝叶斯网．进而针对短时交通流量预测的实时性和高效性要求，提出了基于Gibbs采样的交通贝叶斯网近似概率推理算法，并进行交通流量的短时预测．实验结果表明，本文提出的交通贝叶斯网构建、近似推理以及相应的短时交通流量的预测方法，具有高效性、准确性和可用性．

英文摘要：

Short-term traffic flow prediction is the critical problem of informative and intelligent management technology in traffic transportation and systems. Current research depends much on historical data, has high computational cost, cannot reflect the dependencies among roads in complex traffic networks nor the uncertainty of these dependencies, or derives complex methods by integrating various underlying models. Bayesian network （BN） is an important probabilistic graphical model （PGM）. In this paper, we construct the traffic BN （TBN） with time-series dependencies based on the traffic network and well-behaved properties of PGM on representing and inferring uncertain knowledge. Further, we propose an approximate algorithm for TBN inferences and the corresponding short-term traffic flow prediction based on the basic idea of Gibbs sampling and the inherence of instantaneous and efficient traffic prediction. Experimental results show that the methods for TBN construction, inference and the corresponding short-term traffic flow prediction are efficient, accurate and applicable.