大部分基于浮动车GPS数据的速度估计模型仅适用于GPS数据采样时间间隔小、样本量空间分布密集的理想情况,无法准确计算样本量不足情况下的实时速度.根据浮动车GPS数据点在空间上的分布情况,提出组合三种速度估计模型,以最大限度地提高GPS数据利用率；考虑到GPS数据点在时间上分布不均,在GPS数据不足的情况下,结合神经网络预测和数据融合的技术,根据误差方差融合速度估计模型的测量值和神经网络拟合的预测值,以减少实时估计误差.选择广州市东风路作为测试实例,在高峰和平峰两种交通场景下比较了融合值、测量值和预测值的误差,结果表明结合神经网络和数据融合技术的城市路段速度估计精度和稳定性均优于速度估计模型.%Most link speed estimation models based GPS data are only suitable for ideal condition in which the sampling inter-val is close and the sample amount is sufficient, but they are unable to estimate the real time link speed if GPS data are not enough. According to the space distribution of GPS data, combined three models, which could maximize the utilization of GPS data and the road network coverage of speed estimating results. It taken uneven distribution of GPS data into account and adopt-ed back propagation neural network ( BPNN ) to repair the link speed when insufficient amount of GPS data occurred. Accord-ing to the error variance of measured value and predicted value, designed a self-adaptive filter to merge the historical data and the neighboring links' data. It selected an arterial road in Guangzhou city as a case study to test the proposed method. The performance test illustrates that the proposed method provides better results than normal speed estimation method.
展开▼
