Link Travel Time Estimation Based On Data Fusion

Link travel-time is reflected in the traffic flow at an important indicator. Real-time or quasi real-time travel time data is an important basis to build ITS subsystems, such as dynamic traffic information service system, signal coordination control system and urban traffic guidance systems. ITS is the research hotspot at home and abroad. Properly giving real-time link travel-time has two meanings. For one thing, it can provide accurate traffic data for traffic management department for management, scheduling. For another, it can provide meaningful information for travelers to enable them to choose the appropriate travel routes, and to avoid wasting time and money while improving road network utilization. Therefore, real-time link travel-time estimation on the road is important to run the system.In this paper, we first make a brief introduction of information collection and data fusion theory, and then get the link travel time by loop detectors and GPS floating car. Lastly, make data fusion by neural network to get more accurate estimation. Contents completed in the paper are as follows:Firstly, make a review of traffic information collection and data fusion theory. Introduce the advantages and disadvantages of commonly used detectors, and then analyze different detectors can get data content. Then focus on data fusion theory, summarize commonly used data fusion algorithm.Secondly, get section travel time estimates through single detector. Analyze the necessity of pretreating the information obtained by detectors, and summarize methods using to identify and repair the fault data. Then introduce a method to get section travel time using loop detectors. The emphasis of this chapter is to research methods of obtaining section travel time using GPS floating car. this thesis makes use of data with similar time features to calibrate state transition parameter throughout citing Kalman filter.Thirdly, based on comparing different fusion methods, the thesis choose diagonal recurrent neural network to fuse section travel time from multi-source. Then based on introductions of diagonal recurrent neural network, the thesis make data fusion of data obtained by single detector before to get more accurate travel time estimates.Finally, the thesis makes a brief introduction of VISSIM simulation platform. Then, set up the simulation platform for the simulation data to verify the previous methods. There is clear evidence that link travel time estimation based on data-fusion is more precise than that based on single detector. Furthermore, data fusion model can better trace the change in traffic flow state and possess the advantages of the estimation model based on probe vehicle data.