Practical travel time prediction algorithms based on neural network and data fusion for urban expressway

Current travel time prediction algorithms often need large amount of travel time data to identify the algorithms parameters. However, it is highly costly and time-consuming to obtain many travel time data. In this paper, we proposed an algorithm structure to calculate the travel time which utilize neural network to dynamically predict future speed and employ data fusion to integrate the speed data of different detectors in a urban expressway link. Based on the algorithm structure, two practical travel time prediction algorithms, called as Space Discretization Travel Time Calculation Algorithm (SDTCM) and Speed Integral Travel Time Calculation Method (SITCM), were developed by the discretization of the space and integral of the predicted speed. Vehicle plate recognition technology was used to collect the real travel time data in the test section on Beijing Third-Ring urban expressway to evaluate the algorithms. The obtained results show that the average prediction error of two algorithms are both less than 10% which can meet the requirement of the field applications and the algorithms are easy to be implemented as no travel time data collection are needed in advance. The two algorithms have some advantages and disadvantages over each other in accuracy and smoothness. Although SDTCM is more accurate than SITCM in general, the fluctuation of error of the SITCM is a little smoother than SDTCM.