基于BP神经网络的交通数据序列动态可预测性分析方法

Dynamic Predictability Analysis for Traffic Data Serials Based on BP Neural Network

为了进一步改善交通数据序列短时多步预测的效果,提出了交通数据序列动态可预测性分析的思想,在设计了交通数据序列动态可预测性关联数据特征指标的基础上,基于BP神经网络建立了交通数据序列动态可预测性分析方法,运用某城市快速路主线与匝道车辆检测器的实际数据对该方法进行了验证,并与不同固定预测步数条件下的预测效果进行了对比分析.结果表明,所提出的方法能对交通数据序列的可预测性进行在线分析,在保持预测精度的情况下,可最大限度地增加交通数据短时预测的步数.

In order to improve the short-term multi-step prediction effect of traffic data serials, a dynamic predictability analysis of traffic data serials is first proposed by this paper. Then the characteristics associated indexes of traffic data serials predictability is designed, and the predictability analysis method for traffic data serials is established based on BP neural networks. The traffic data for test are obtained from a real section of urban expressway, the performance of proposed method is compared to different fixed prediction steps. The experimental results indicate that the proposed method can predictability analyze the predictability of traffic data serials on-line, and short-term prediction step was maximized under the premise of maintaining the prediction accuracy.