Research And Implementation Of Large-Scale FCD Processing

With the deeply development of economy and urbanization process, the number of cars on the roads increases tremendously, which causes the traffic jam, incident and air pollution. ITS is publicly known to be the way to solve or relieve this situation, while every part of ITS is surrounded by the application of the traffic information. Therefore, it is important to provide the real-time and accurate traffic information. FCD technology is an approach of traffic information acquirement developed recently, which has a low cost, a wide range and can get the real-time information. The process of FCD processing is changing the large scale FCD records into the easily understanding road traffic information, such as travel time and road section speed. In this paper, we research into the larger scale FCD processing approach, including task decomposition and dispatch, parallel programming framework, and the important algorithms, including FCD map matching, travel time and road speed estimation. On the foundation of this research, FCD parallel processing system and data processing center are implemented. And also an FCD experiment platform is developed to support the research. The main work of this dissertation includes:1. A large scale FCD parallel processing framework is proposed on the platform of KD-50 HPC, and the definition of each step in the framework is given. After the study of multi-task decomposition and dispatch, an improved multi-task decomposition algorithm and a dynamic task dispatcher are developed. Furthermore,an FCD system design based on MapReduce programming model is proposed.2. FCD map matching algorithm is studied deeply in this paper. Comparing the incremental and global algorithms and analyzing the influence factor of FCD matching error, an FCD map matching algorithm based on multi-class SVM is proposed. The training process of parameters'selection and comparison of result are given.3. The edge speed estimation method based on the average speed of interval is introduced, which is analyzed by real testing car data in this paper. Aiming at compensating the missing speed data of some edges, a simple moving average method of prediction is given.4. An implementation of FCD parallel processing system is given based on the content researched in this paper. By means of improving the assignment approach of the tasks and resources, a traffic data processing center is developed in which multi-city FCD can be processed in same time. In order to research and develop the FCD processing algorithm in a better way, we developed the FCD experiment platform using visual GIS technology.