Abstract: The rapid development of Internet of things (IoT) and in-stream big data processing technology brings new opportunity to intelligent transportation system (ITS). Traffic flow prediction is the key issue of ITS. In traffic flow prediction, one fixed model cannot get excellent performance under different environments. Predicting models should also be updated according to data stream. In order to resolve these problems, this paper proposes a traffic prediction method based on structure varying dynamic Bayesian network. Based on complex event processing, this method uses context clustering to partition historical data and uses online clustering to support the update of clusters. A search-score method is used to learn the structure of Bayesian network and Gaussian mixture model is used for approximate inference of Bayesian network. When predicting at run time, appropriate model or model composition are selected based on current context. The experiments on both real and simulation data show that the proposed  method has better performance than popular methods currently used.

Key words: intelligent transportation system, traffic flow prediction, complex event processing, structure varying dynamic Bayesian network

摘要： 物联网和大数据流式计算的快速发展为智能交通系统的研究带来新的机遇。交通流量预测一直是智能交通系统的关键问题。针对交通流量预测中一个固定模型无法适应多种环境的问题，以及面向数据流的模型更新问题，提出了一种基于变结构动态贝叶斯网络的交通流量预测方法。该方法以复杂事件处理和事件上下文为基础，通过上下文聚类进行历史数据的划分，并通过事件流在线聚类支持聚簇的更新。面向不同聚簇的数据，采取搜索-打分的方法学习对应的贝叶斯网络结构，基于高斯混合模型实现贝叶斯网络的近似推断。在线预测时根据当前上下文选择合适的模型或模型组合进行预测。真实和仿真数据上的实验结果表明，该方法能够获得比当前常用方法更好的预测效果。

关键词: 智能交通系统, 交通流量预测, 复杂事件处理, 变结构动态贝叶斯网络