Fusion Of Multi-Surce Heterogeneous Taffic Flow Data For The Assessment Of Traffic Operational Conditions

With the rapid increase of motorized vehicles, traffic congestion and traffic operational conditions of the road network in Beijing have been deteriorating continuously. An accurate assessment of traffic operational conditions is the basis to implementing effective traffic congestion mitigation strategies. Thus, the development of assessment models that can quantify traffic congestions accurately have become one of key research areas in traffic engineering. However, because traffic flow data used for evaluating traffic operational conditions have shown errors or losses in either the data collection stage or its transmission process, the assessment of traffic operational conditions based on single data source may bear substantial inaccurate results. The data fusion is an effective approach to counter this problem, whose purpose is to fuse the data from different detection techniques (e.g. different sources) for assessing the traffic operational conditions in order to make the results more accurate.In this background, the primary goal of this thesis is to develop a multi-source heterogeneous data fusion model for providing more accurate assessment of road traffic operational conditions based on the heterogeneous traffic flow data collected from three sources: floating car data, RTMS data and the license plate recognition data. Thereinto floating data and RTMS data would be used as the basic data for fusion and the license plate recognition data would be used as the true data for comparison purpose.First, the thesis analyzes the data collection methods and the data characteristics for the data from different sources including floating data, RTMS data and the license plate recognition data., and matches the data both temporarily and spacially, so that a group of data from different sources are derived as the basis for data fusion that can describe the traffic operational conditions for the same period on the same segments of the roads. Then, it compares the various data fusion methods in light of their strengths and shortcomings, and selects the neural network method for this study. In the process of data fusion, the thesis presents two data fusion models: Front-End Fusion (FEF) and Rear-End Fusion (REF). Finally, the thesis conducts a case study using the neural network method for both FEF and REF. In the case study, Travel Time Index (TTI) is used to evaluate the traffic operational conditions. By calculating the real data and analyzing the results, the thesis quantitatively compares the fusion results from different fusion techniques against the true data, and shows the effects of before- and after-fusion and different fusion methods on the assessment results.