Link Travel Time Estimation Model Fusing Data from Mobile and Stationary Detector Based on BP Neural Network

Real-time transportation information is the foundation and ensure of dynamic route guidance system. Stationary sensors´ detecting precision is high. But because stationary sensors only can detect point information of links, stationary sensors´ maturity degree is bad. On account of mobile sensors´ detecting links´ livelong transportation information, mobile sensors´ maturity degree is high. However because of GPS data´s errors and probe vehicles´ randomness mobile sensors´ detecting precision is bad. Considering colligating the mobile and stationary sensors´ advantage, this paper proposes a new mobile and stationary sensor fusion model based on BP neural network to improve the accuracy and maturity degree of estimating travel time. The model consists of three modules: (1) mobile detecting module which measure first part, second part and third part travel time over a link using taxis equipped with differential global positioning system receivers; (2) loop detecting module which measure travel time using fixed detectors fixed in roads and traffic signal timing parameters; and (3) data fusion module which uses a neural network to combine outputs from the above two modules to improve the travel time estimation accuracy. This model´s inputs respectively are: travel time detected by mobile sensors, travel time detected by stationary sensors, mobile sensors´ density in the link and stationary sensors´ density in the link. This model´s output is the link´s travel time. To validate the validity of this model, this paper presents the test of this model using a great deal of real data in Guangzhou city. The result indicates that this model is valid