Abstract:The study on real-time travel speed estimation for urban arterial road based on multi-source traffic data is presented in this paper. Loop detector data (LDD) and floating car data (FCD) are applied in back-propagation (BP) neural network to estimate the travel speed. Firstly, data characteristics of loop detector data and floating car data are analyzed. Secondly, according to the data missing status, data is complemented and then four data fusion models are established based on BP neural network. The input vector is composed of travel speed, data number from FCD and traffic volume, spot speed, time occupancy rate from LDD, and the output of fusion model is the travel speed estimation. Finally, the four models are examined by the realistic traffic data, and least square error method (LSE) and mean absolute relative error (MARE) are evaluated. The results show that the mean estimation accuracy of travel speed for four fusion models achieves 84% and meets the accuracy requirements of practical application.

Estimation of urban arterial real-time travel speed based on multi-source data fusion