Research On Key Technologies In Optimal Sensing For Urban Traffic Flow Data

Traffic data collection is one of the core sub-system of intelligent transportation systems, and which is the basis of all kinds of traffic applications. Collecting traffic data with higher spatiotemporal precision by advanced information technologies, and controlling or guiding traffic flow with microscopic traffic control systems and traffic information systems to balance the spatioal-and-temporal distribution of traffic flow on road networks, which is key to resolve the global and growing crisis of urban traffic congestion problem. Traditional technologies can only detect data in fixed locations, and in actual applications fixed detectors are installed in only arterial roads and main intersections, that causes many blank space with information vacuum existing in the urban road networks, actually can not sense the dynamic characteristic of urban traffic flow comprehensively and thoroughly, and the optimization performance of traffic control systems is restricted. Recently, new information technology continuously appears, such as mobile Internet, wireless sensor networks, Internet of vehicles, etc. If the data generated by these systems can be connected with intelligent transportation systems, new way will be provided to urban traffic sensing. It is of momentous current significance and far reaching critical significance to explore ways in traffic sensing which has high accuracy, good real-time performance, lower cost and easily to install and maintain, and more importantly, the new method should accommodate the big data age.Firstly, in point data collection, optimal model and algorithms for traffic surveillance using wireless sensor networks is studied. Wireless sensor networks can deploy in large scale, which has a good application prospect in intelligent transportation systems. In this paper, aiming at the problems of threshold update and vehicle length existing in the adaptive threshold detection algorithm proposed by P. Varaiya, signal correlation model and binary proximity sensor network are introduced to improve the performance of vehicle speed estimation and vehicle classification, and the robustness under situations of threshold shifts or overlapping signals.Secondly, the application of crowd participatory sensing for traffic data collection is studied, and Lagrangian sensing for link traffic sensing is proposed. In this method, wireless sensor data is utilized to resolve traffic equations to predict the internal operation behaviours of traffic flow, and meanwhile, with participatory data as measurement data, Kalman filtering algorithm is employed to estimate traffic parameter optimally integrating traffic equations and actual observation equations, to achieve continuous, high resolution traffic flow data. On this basis, congestion factor is proposed to evaluate real-time traffic congestion situation and used in traffic light timing optimization. Signal phase sequence is optimized with particle swarm optimization to avoid the formation of traffic jam.Thirdly, data set selection problem in crowd participatory sensing is studied. Existing research shows that sensor locations has larger impact on traffic flow estimation than sensor quantity. In large scale urban road network, the volume of participatory sensing data is huge, and important problem appears that how to differentiate data value and select the optimal set in massive data. In this paper the sensor data selection problem under the condition that optional locations are given is studied. A multi-objective optimization model is built with mutual information entropy as objective function, and mean square error as constraints, and finally a sequential selection algorithm for sensor data set selection based on Bayesian optimization is proposed.Fourthly, process of dynamic uncertainty caused by time-dependent network topology variation and traffic flow is studied based on the characteristics that sensor is moving with traffic flow in vehicular sensor network based participatory sensing. Time-varying data value network is defined based on utility function of sensor data, and concurrently select data-set using ant colony optimization. Furthermore, arming at the characteristics that sensor moving and selective data collection, a Internet based transmission protocol is proposed to enable control node to make sense of traffic pattern change and select the optimal data, and control the data sensing and transmission of sensor nodes with feedback.