Study On Interval Speed Fusion Algorithm Of Urban Expressway Based On Multi-source Detectors

In order to solve the traffic problems caused by the rapid growth of urbanization, many cities develop the intelligent transportation system as an important measure to solve the urban traffic problems. To obtain the dynamic traffic information timely and accurately, it is necessary to make sure the intelligent transportation system can play the effective role. In recent years, in addition to the traditional fixed detectors, the floating car system based on mobile vehicles is more and more applied to data acquisition. But because of the differences between the fixed detection technology and the floating car data acquisition technology, there are all kinds of problems of the collected information, like heterogeneity, inconsistent, imprecise and data missing, causing the limitations in the accuracy, integrity and reliability of the information from the intelligent transportation system.This paper is expected to realize the mutual verification of multi-source traffic detection data through data fusion, and to get more accurate, comprehensive and reliable urban expressway traffic information.Firstly, based on multi-source traffic flow data collected on the expressway, this paper analysis the temporal characteristics and the relational models of the three traffic flow parameters, and introduces several common multi-source data fusion methods, considering their advantages and disadvantages and then proposing fusion model of this paper.Secondly, for the noise existing in the data, the abnormal data and other issues, this paper using the Kalman filter algorithm to reduce noise, combining the threshold and traffic flow mechanism to identify erroneous data. It provides a simple and practical way to repair abnormal floating car data, and uses radial basis (RBF) neural network to reconstruct surface for repairing microwave erroneous data.Finally, establishing a more complete and efficient data fusion programs, this paper puts forward two models for multi-source traffic data fusion. One is the adaptive neural network fuzzy inference system (ANFIS), and the other is BP neural network optimized by the genetic algorithm, and to compare the fusion results with the simple BP neural network. The microwave detection data and floating car data are the basic data of the data fusion, and the license detection data is as the real traffic situation of the real data to test the model. Using the mean absolute error (MRE) and least square error method (LSE) to test the models. Test results show that:the two fusion models’accuracy were both above 90% and after fusion of the models, the speed value were more loser to the true value than any single source traffic flow data before fusion, and the accuracy and effectiveness of the fusion models have been well verified. Furthermore, a combination of two models is superior to single BP neural network in fusion results.