利用浮动车大数据进行稀疏路段行程时间推断

针对利用实时浮动车数据估计路段行程时间时存在的数据稀疏性问题，提出了构建三层神经网络模型，以目标路段与邻接路段间的特征关系为输入、目标路段与邻接路段行程时间比值为输出，利用浮动车历史大数据获取路段之间的交通时空关联关系，继而用于路段行程时间的推断。采用武汉市2014年3~7月的浮动车GPS历史数据进行验证，得到的路段行程时间估计值的平均绝对百分比误差小于25%，证明了所提方法的有效性。

Sparse Link Travel Time Estimation Using Big Data of Floating Car

Although there exists quantities of GPS data of floating car, partial links lack real data during some certain period of time. Therefore, we can't estimate target link travel time. Considering the problem of sparse data when using floating car data estimating link travel time, we put forward a kind of inferred method based on big data of floating car. We designed a three-layer artificial neural network model, whose input information and output information are the feature relationship and the travel time ratio between target link and adjacent link respectively. We obtained traffic spatiotemporal association relationship using historical big data of floating car and then inferred link travel time. The model was verified by historical big data of Wuhan's floating car from March to July, 2014 and the MAPE of estimated value of link travel time is less than 25% which proved the effectiveness of the proposed method.