Study On Data Fusion Technology Of Multi-source Traffic Data Of Urban Expressway And Arterial Road

With the rapid growth of volume of motor vehicles in recent years, urban traffic problems, such as traffic congestion, accidents and environmental pollution have become more and more serious. It has been widely acknowledged that the development of intelligent transport systems (ITS) is one of the most important solutions to urban traffic problems, and accurate real-time traffic information is an important foundation and prerequisite for ITS’s effective operation. Besides the conventional roadside traffic detectors, floating car based traffic surveillance technology developed rapidly in the last decade. However, for the different characteristics between roadside detection technology and floating car data (FCD) acquisition technology, the collected information usually is heterogenous, inconsist, inaccurate and incomplete, so the output information of ITS is not so satisfactory in accuracy, completeness and reliability.Therefore, based on multi-source traffic data collected from urban expressway and arterial road, this paper proposes an advanced data fusion scheme for real-time speed estimation to provide more accurate, more complete and more reliable traffic information by complementing and validating each other of multi-source data. Thereinto RTMS (Remote Traffic Microwave Sensor) data and FCD are used as the elementary data source for urban expressway, while loop detector data and FCD are used as the elementary data source for the arterial road, and the license plate data are viewed as the true value for model examination.First of all, for the data missing and data error existed in multi-source data, practical methods are proposed separately to eliminate erroneous data and fill missing data. Secondly, based on the characteristicses of Genetic Algorithm and BP Neural Network, the GA-BP combination model is applied for multi-source data fusion to overcome the defect that BP Neural Network algorithm is not easy to find optimal solutions to problems. Thirdly, according to the differences in study area, spatial analysis unit, data loss situation and data-filling method, total33models belonging to3categories are established, including12plate number detection intervel based models of expressway,12link based models of expressway, and9plate number detection intervel based models of arterial road. Finally, the models are examined by realistic traffic data with least square error method (LSE) and mean relative error (MRE) as evaluation indicators. The results show that with the application of GA-BP neural network, each fusion model has a satisfactory accuracy up to82%-91%and the traffic information provided is more accurate than each single source traffic data. The accuracy and effectiveness of all the models are well verified.