基于RFID电子车牌数据的公交行程时间预测方法

Bus Travel Time Prediction Method Based on RFID Electronic License Plate Data

为了给公交优先信号配时系统提供足够的'思考'时间和准确的控制依据,基于重庆市RFID电子车牌数据提出了一种采用自适应渐消卡尔曼滤波和小波神经网络组合模型动态预测公交行程时间的方法。综合分析公交行程时间的动态和静态影响因素,选取的模型输入参量为标准车流量、路段车辆平均行程时间、平均车速离散性和前班次公交行程时间。利用RFID电子车牌系统采集重庆市鹅公岩大桥路段车辆行驶数据,选取3 000组实际运行数据完成公交行程时间预测模型的训练,另筛选50组数据验证模型的有效性和准确性。研究结果表明:组合模型可动态自适应预测公交行程时间,预测值平均相对误差为3.23%,绝对误差集中在8 s左右,明显优于2种单一模型和基于传统GPS数据的公交行程时间预测模型,可认为选择RFID电子车牌数据作为组合模型的输入,能够明显改善模型预测精度;组合模型预测值的残差分布更为集中、鲁棒性较好,泛化能力强。选择平均绝对误差值、均方根误差值和平均绝对百分比误差作为模型评价指标,结果进一步表明,组合模型的综合预测效果明显优于单一的自适应渐消卡尔曼滤波和小波神经网络。研究方案可为先进公交信息化系统提供良好的技术支撑。

To provide sufficient'thinking'time and accurate control for a transit signal priority(TSP)timing system,this study proposed a method that combines an adaptive fading Kalman filter and wavelet neural network to predict bus travel times dynamically.This is based on the Chongqing radio frequency identification(RFID)electronic license plate data.This study comprehensively analyzed the dynamic and static influencing factors of bus travel times and used the following model input parameters:standard traffic volume,average travel time over the road segment,average vehicle speed dispersion,and travel time of a pre-class bus.The RFID electronic license plate system was used to collect vehicle driving data of Chongqing Egongyan Bridge and the bus travel time prediction model was trained using 3 000 sets of actual operational data.Fifty groups of actual running data were used to verify the validity and accuracy of the model.Experimental results show that the combination model can dynamically and adaptively predict bus travel times.The average relative error of the predicted value is 3.23%,whereas the absolute error is approximately 8 s.These errors are obviously better than those of the two previous single models and the travel time prediction model based on traditional GPS data.This study shows that selecting RFID vehicle license plate data as input for the combination model can significantly improve the model’s prediction accuracy.The residuals distribution is more concentrated in the combination model,which has good robustness and a strong generalization ability.The mean absolute error,root mean square percentage error,and mean absolute percentage error were selected as model evaluation indicators.Comprehensive prediction capabilities of the combination model are obviously better than that of both the single fading Kalman filter and wavelet neural network.This research can provide effective technical support for advanced public transportation information systems.