中文摘要：

为精确估计路段平均速度,提出了基于BP神经网络与D-S证据理论的路段平均速度融合方法。通过训练完成的BP神经网络估计概率密度函数值,进而通过D-S证据理论进行数据融合,整合了BP神经网络自学习的特点与D-S证据理论推理的能力。提出了融合方法的框架,给出了具体的计算模型。利用京藏高速公路上的实测浮动车数据、微波检测器数据、车牌识别数据对融合方法进行了验证,并分析了当微波检测器失效时融合方法的鲁棒性。分析结果表明：融合数据的平均绝对误差百分率比仅使用浮动车数据或微波检测器数据分别提高了7.90%、20.72%,融合方法能够得到较好的效果。微波检测器失效的情况下,融合精度有所下降,但融合数据的误差仍然小于仅使用浮动车数据的误差,说明融合方法具有一定的鲁棒性。

英文摘要：

In order to estimate road section average speed accurately, a fusion method of road section average speed based on BP neural network and D-S evidence theory was proposed. The values of probability density function were estimated by trained BP neural network, and the data were fused by D-S evidence theory. The self-learning ability of BP neural network and the reasoning ability of D-S evidence theory were combined in the fusion method. The framework and model of the fusion method were presented, and each process of the method was analyzed. The fusion method was verified by using floating car data （FCD）, microwave detector data, and license plate recognition data from Beijing-Xizang Expressway. The robustness of the fusion method was verified in the case that microwave detector failed to work. Analysis result indicates that the mean absolute percentage errors of fusion data are 7.90%, 20.72% better than that of FCD and microwave detector data respectively. When microwave detector fail to work, the fusion accuracy reduces, but the errors of fusion data is still smaller than that of FCD, and the fusion method is proved to be robustness. 2 tabs, 6 figs, 26 refs.