Research On Freeway Traffic State Estimation Via Multi-Source Heterogeneous Data

Under the current freeway network, the increasing traffic will lead to more traffic problems,i.e., jams and accidents. The Intelligent Transportation Systems (ITS) are regarded as the most efficient technique to solve these problems. The real-time traffic state is the fundamental information for ITS. There are various methods for the collection of traffic state information,including the inductive loops, microwave sensors, the satellite positioning technology, the video and image processing technology and the cellphone-based positioning technology, etc.Although more traffic information is available, a series of issues are arising at the same time,such as how to deal with the heterogeneity of these data when they differ in semantic, resolution and precision? In essence, these issues are the multi-sensor data fusion in the field of traffic engineering, which have been extensively studied. However, it is still hard to determine the optimal fusion method, and there are challenges for the existing fusion method when the data collection methods are updated or new technologies are emerging. The cellphone-based approaches using the big data from the wireless communication records show great potential to realize the traffic monitoring for a large-scale freeway network. On the other hand, the traditional fixed sensors have been applied in the field for decades. Therefore, the subject of this study is to investigate the integration of cellphone-based technologies and the traditional fixed sensors which have not been explored.This dissertation was supported by the National Key Basic Research Development Program of China (973 Program) and the technology demonstration project of the Ministry of Transport of the People's Republic of China "The intelligent platform for the Jiangsu Freeway operation and information service". This study aims to furtherly investigate the possibility of newly developed cellphone-based technology, and extend the existing data fusion theory and the traffic state estimation technology. Moreover, the study using field data will be a good reference for the field application, and thus, it will benefit the freeway operation and information service. The study focus on the integration of the cellphone handoff based traffic data collection method (hereafter referred to as the "HO method")and the fixed sensors, as well as the combination of the cellphone activity data-based traffic data collection method (hereafter referred to as the "CA method"). Using data from these sources, the study explored some key issues of the appropriate fusion and estimation methodology by first analyzing the characteristics of the multi-source heterogeneous data, then constructing an optimal fusion and estimation method, and finally proposing a method to deploy different combinations of data collection technologies. In summary, the main content and key findings of the study are as follows:First, the study classified the different kinds of traffic data collection technologies, and then analyzed their applications on the freeway and the future development prospects. Based on the analysis, the study furtherly analyzed the rationale of the technologies and their data characteristics using the field collected data. Specifically, the impact of the freeway length between two adjacent handoffs, the traffic volume, and the distribution of the valid handoff sample size were explored for the HO method. Besides, the study proposed a Dynamic Time Warping (DTW) based hierarchical clustering and classification method to investigate the measurements collected by the CA method. Via this DTW-based method, the study found out the spatial and temporal differences in the measurements. Moreover, the study evaluated the accuracy of the fixed sensors based on the fundamental diagram. Generally, the above analysis is basic and essential for the following research.Second, the study proposed a neural network-based fusion and estimation method to integrate the data from the HO method and the fixed sensors. There are three main modules in the proposed method, i.e., the data transformation module, the neural network-based estimation module and the neural network-based fusion module. To test the proposed method, the study established a VISSIM simulation model based on the real freeway segment of Xi-Cheng freeway in Jiangsu, China. The sensitivity analysis indicated that the length of handoff link and handoff sample size had an impact on the fusion accuracy. Afterward, the proposed method was tested using the simulated data and was also evaluated by the comparison with other fusion methods, i.e., the simple convex combination and Dempster-Shafer evidence theory. The results showed that the proposed method successfully integrated the multi-source data, improved the estimation accuracy compared with other methods, and enlarged the temporal and spatial coverage of measurements.Third, since the traffic features extracted from the CA data have not been applied to the traffic state estimation, the study established the relation model to transform traffic features to the state variables. And the traffic features were selected via the previously proposed DTW-based method. The relation model combined with a second-order macroscopic traffic flow model were rewritten as a state-space model. Using the Extended Kalman Filter (EKF)technique, the traffic features were successfully used to estimate the freeway traffic state.Accordingly, the study proposed a Progressive EKF (PEKF) based fusion and estimation method to adapt the multi-source data from the CA data and fixed sensors. The test using the field data shows that the PEKF-based method could reconstruct the freeway traffic state with multi-source heterogeneous data. Besides, the results indicated that the estimates could exhibit the impact of the accident on the freeway traffic state.Finally, for the combination of the HO method and the fixed sensors, the study set the length of handoff link and the distance between fixed sensors as the variables, and set the accuracy of fused traffic state as the dependent variable. Accordingly, several schemes of the sensor placement were presented and the simulation models were constructed accordingly. The study used several accuracy indexes to evaluate the schemes and find out the optimal sensor placement. For the combination of the CA method and the fixed sensors, the study proposed the rules to determine the sensor placement by a step-to-step consideration as follows: the application needs for the spatial resolution, the surveillance needs for the critical freeway segments, the areas that do not have qualified traffic featured from CA data and the data requirements of the proposed fusion and estimation method.The works in the dissertation contribute to the theory of traffic data collection technologies,the multi-source traffic data fusion techniques, the estimation approaches for freeway traffic estimation and the placement of multiple collection technologies. Moreover, the simulation and field test and validation provide a good reference for the field application.