基于状态空间模型的道路交通状态多点时间序列预测

Road Traffic State Multi-spot Time Series Forecasting Based on State Space Model

以多点的道路交通状态为研究对象,把道路交通状态单点预测向多点同时预测扩展,提出了基于状态空间模型的道路交通状态多点时间序列预测方法。首先,利用道路交通状态的多点时间序列数据建立多维自回归模型,转化状态空间模型形式,接着利用EM算法估计状态空间模型参数,从而得到多点道路交通状态的状态空间模型;其次,根据时间序列数据估计系统状态,利用卡尔曼滤波算法进行一步预测,补充新的数据并更新系统状态递推预测;最后,利用某城市快速路上相邻6个交通检测器采集的多点时间序列数据验证模型的有效性,并与卡尔曼滤波单点预测方法相对比。结果表明:该模型是可行和有效的。

Regarded multi-spot road traffic state as research object, a method of road traffic state multi-spot time series forecasting based on state space model was proposed and road traffic state forecasting was extended from single-spot forecasting to multi-spot forecasting. Firstly, using road traffic state multi-spot time series data, multidimensional auto-regressive model was established and converted to format of state space model, then, EM algorithm was used to estimate the state space model parameters in order to establish state space model of multi-spot traffic state. Secondly, system state was estimated based on time series data, then, Kalman filtering method was used to perform one-step prediction and simultaneously new data was supplied to update system state so as to perform next prediction. In the way, prediction was performed on and on. Finally, the effect of proposed method was tested by multi-spot time series data collected by six traffic detectors on certain urban expressway and was compared with forecasting method of single-spot Kalman filtering. The result shows that the proposed model is feasible and effective.