Research On Virtual Detection Technique In Mixed Adaptive Traffic Control System

The traffic congestion has become the bottleneck, which restricts the development of the urban city all over the world. It not only causes the enormous losing, but also builds the traffic accident, environment pollution, and energy waste. Up till now, the Intelligent Transportation Systems (ITS), which mainly use the information technology, are considered as one of the best ways to solve the traffic problems of the urban city. In order to make full use of ITS, we need timely, accurate and dependable information. And it's also very important to provide such information for the traffic users. So it requires higher for the traffic systems which obtain and cope with information.In our country, the density of the city route is becoming higher and higher, and the number of the nodes is becoming more, under this situation, to realize the macroscopical management, we must get most (even all) the traffic information of the nodes. Most of the traffic information we get from the road is through detections which can be divided into fixed detections and floating detections, especially the inductive loop detector has become the most popular form of detection system.There are inevitably some faults in some dynamic data which we get from traffic system, and because the data's numerous characteristic, if we use these data without selection and disposal, it will surely be harmful to the models' validity.In SCOOT, detectors are setting at outbound of signalized intersection, system can optimize the offset base on the real-time traffic condition which reflected by detector data, but it is not good at optimizing split, because absents of traffic data near stop-bar. In contradiction to SCOOT, SCATS' detectors are setting near stop-bar. System is good at optimizing split but no offset. In engineering, it is impossible to setting detector both at upstream and near stop-bar. So this paper puts forward virtual detector in traffic signal control system. It analyzes vehicles fleet discrete and departure, data mining, multi-sensor data fusion, data relevancy and correlation and provides traffic parameters. Under cooperation with upstream outbound detector and virtual detector, traffic signal control system's strategical and tactical control achieves better coordinate and performance.This thesis is based on National High-tech Research and Development Program (863 Program): key technologies research and development on new generation of intellectualized traffic control system under grant 2006AA11Z228. The thesis studies the installation of the detection now used in the traffic control system and fixes on the problems in the process of installation. Taking account of their theoretical significance and value in practice, the thesis also studies the collection and disposal of the data and multi-sensor data fusion (MSDF) technology, then improves the MSSDF technology. Based on it, makes use of the virtual detection technique to predict the traffic parameters and improves the veracity of the parameters. And some conclusions are obtained based on the data from the survey, which testify the control systems which use virtual detectors technology get better effects.The thesis comprises of 5 chapters and their contents are as follows:Chapter One: Introduction. First, explain what and why to be investigated in the thesis. Point out the thesis's value both from the perspective of theory and practice, and then states the deficiency of the data detection in SCOOT and SCATS , brings forward the virtual detection technique to predict the traffic parameter. After that, the main content and structure of the thesis is given.Chapter Two: The installation of the fixed detections in MATCS. Analyses of the different methods in traffic information collection and summarizes the means which are mainly used in the traffic information collection at present. According to different place, the detections are divided into fixed detections and floating detections. The thesis describes the two technologies simply, and introduces the fixed detector especially. Then introduce several sorts of fixed detector, express their characteristics and adapted situations, which gives a criterion of choosing detectors in data collection in MATCS. There different methods of different detectors in installation, the thesis gives different suggestions according different situations, and expresses the factors when they are installed, which purposes are making best use of the detectors in data collection. Chapter 3: The research of virtual detector technology in MATCS system. The chapter 3 of this paper expatiated the basic principle of virtual detector,analyzed the instance which lead to abnormal data when detector is working. Through the analysis and anomaly recognition of collected data, it judged the reason leading to abnormal data and discriminated the state of detector. It gived the definition and measurement of abnormal data and introduced several method of judying abnormal data, then classified abnormal data and introduced different method in different circumstances. This paper introduced the integration method of multi-source data and compared them, BP neural network method was discussed in detail. The discrete situation of the traffic flow ,which started from the stop line of the upper intersection, in the process of running and the correlation of the adjecent detector in the same intersection were analyzed in this papar and this provided a basis for initially predicting the traffic volume of the vitual detector. BP neural network method was used for minnig multi-sensor data at the intersection. Because of the correlation in time of virtual detector's data, Exponential smoothing method was used for achieving the excavation and prediction of traffic data in time. Finally, the use of traffic data predicted by virtual detector achieved a further extrapolation.Chapter 4:The data testing of virtual detector in MATCS. According to the actually measured data and choosing representative of traffic data: traffic flow, the average speed of time, time occupancy and so on, this chapter validated the virtual detector model based on BP neural network and verified that the model has achieved positive results, then it validated the traffic parameters forecast techniques for the second time based on exponential smoothing and the results show the forecasted data of the model can accord with the actually measured data. Finally, the control system used vitural detector was compared with that not use virtual detector and the result show the system used virtual detector had better effect.Chapter Five: Conclusions and the future work. The thesis is summed up and the research process and the main contents of the thesis are looked back. At the same time, point out the innovation and the existing problems, as well as the problems that need to further study in the future.According to the study on MSDF and the traffic parameters collection based on virtual detection technique in the thesis, timely, accurate and dependable information can be acquired, which is significant to improve the ITS, especially thevalidity in the mixed Adaptive Traffic Signal Control system.