节点文献
基于自适应秩动态张量分析的短时交通流预测
Short-term traffic flow prediction based on adaptive rank dynamic tensor analysis
【摘要】 在智能交通系统中,短时交通流预测可以为路线规划、交通管理和公共安全等领域提供数据支撑。为了提高数据缺失和异常情况下的预测准确性,提出了一种基于自适应秩动态张量分析的算法来进行短时交通流预测。首先构造了覆盖周、天、时间窗口和空间4个维度的张量,以挖掘交通流数据之间的多模相关性。其次,利用滑动窗口模型,形成动态结构的张量流数据。然后将主成分分析算法扩展成可以接收张量输入的离线张量分析算法,并引入自适应秩和遗忘因子形成自适应秩动态张量分析算法。最后将张量流数据输入自适应秩动态张量分析算法中,实现对短时交通流数据的预测。实验结果显示,即使在数据有缺失的情况下,自适应秩动态张量分析算法也能实现良好的预测。
【Abstract】 Short-term traffic flow prediction in intelligent transportation system can provide data support in areas such as route planning, traffic management, public safety and so on. In order to improve the prediction accuracy with missing and abnormal data, a short-term traffic flow prediction method based on the adaptive rank dynamic tensor analysis was proposed. Firstly, a four dimensional tensor consisted of week, day, time and space was constructed, which could excavate the multimodal correlation of traffic flow data. Secondly, tensor flow data with dynamic structure was formed by using sliding window model. The principal component analysis(PCA) algorithm was extended to an offline tensor analysis algorithm that could accept tensor input. Then the adaptive rank and the forgetting factor were introduced to generate an adaptive rank dynamic tensor analysis algorithm. Finally, the tensor stream data was inputted into the adaptive rank dynamic tensor analysis algorithm to realize the short-term traffic flow prediction. The experimental results show that a good prediction can be achieved even with data missing.
【Key words】 short-term traffic flow prediction; data missing; dynamic tensor analysis; multimodal information;
- 【文献出处】 物联网学报 ,Chinese Journal on Internet of Things , 编辑部邮箱 ,2019年03期
- 【分类号】O183.2;U491.14
- 【被引频次】1
- 【下载频次】152