城市路网短期行程时间预测研究

Short-term Travel Time Prediction for Urban Networks

提出了一种基于状态空间神经网络(SSNN)和拓展卡尔曼滤波(EKF)的混合式行程时间预测模型.以往的研究表明,状态空间神经网络能够较好地处理复杂的非线性时空问题.然而,状态空间神经网络需要大量的历史数据作为离线训练之用.其不足之处在于,首先是需要花费大量的时间和精力去收集、准备数据,以及训练神经网络.其次,输入输出随着时间不断增加,训练过程需要不断的从新重复.为了提高状态空间神经网络的有效性,扩展卡尔曼滤波代替了传统的方法来对神经网络进行训练.荷兰的一条城市道路被选择为模型验证的试验路段.通过与另外两个预测模型之间的对比验证,该模型的预测能力能够达到满意的有效性、准确性和鲁棒性.

This paper presents a hybrid model for urban arterial travel time prediction based on the so-called state space neural networks （SSNN） and the extended Kalman Filter （EKF）. Previous research shows that the SSNN is able to deal with complex nonlinear spatio-temporal problems. However, the SSNN models required offline training with large data sets of input-output data. The main drawbacks of such a requirement are, first, the amount of time and effort involved in collecting, preparing and executing these training sessions. Second, as the input-output mapping changes over time, the model requires complete retraining. To improve the effectiveness of SSNN, the extended Kalman Filter is proposed to train the SSNN instead of conventional approaches. A densely used urban arterial in Netherlands was selected to test the performance of this model. This paper has compared the performance of this proposed model with two existing models. The results of the comparisons indicate that this proposed model is capable of dealing with complex nonlinear urban arterial travel time prediction with satisfying effectiveness, robustness, and reliability.