| Travel time is an important indicator of road traffic state, an important aspect of Intelligent Transportation System. Accurate prediction of the arterial travel time is the foundation of the Traffic Guidance System and the Advanced Traveler Information System. Before travel time prediction, traffic data should be collected, and traditional methods for traffic data collection are mainly based on the inductive loops, the license plate recognition and floating car data. In the past, most of the research on travel time prediction is based on these three traffic data collection methods. With the development of technology, especially the popularity of smartphones and the improvement of vehicle information system as well as the advancements of Bluetooth technology, using Bluetooth technology to collect real-time traffic data becomes more and more popular abroad. The device collects Bluetooth signals from Bluetooth-enabled devices, such as smartphones, car hands-free systems and Bluetooth earpieces in the vehicles, and uses an electronic identifier called a machine access control(MAC) address to calculate travel time. The MAC address is a unique label to identify Bluetooth-enabled devices. In principle, the Bluetooth detectors are installed along the road to collect and store the MAC address and detect the time of passing vehicles. Travel time is calculated by matching Bluetooth MAC addresses at successive detection stations.This paper compares the advantages and disadvantages of traditional traffic data collection methods, and introduces the process of using Bluetooth detectors to collect traffic data in detail. Travel time data filtering and the average travel time estimation method are proposed based on the Bluetooth technology. In order to test the feasibility of the application of Bluetooth technology in China, we carried out a survey about the usage of Bluetooth devices and conducted a field experiment using Bluetooth devices in Shanghai, China. Based on the survey and the field experiment, we found that the rate of detected Bluetooth devices was about 2.7% to 4.3% among vehicles on the arterial. In order to forecast the travel time, two models are proposed, one is Kalman method for short-term travel time prediction based on Kalman filtering theory, the other is PSO-SVR model for short-term prediction based on support vector machine(SVM) theory and particle swarm optimization(PSO) algorithm. Filed traffic data were collected on the Humin road, Minhang district of Shanghai using Bluetooth detectors, the calculation and validation of the two proposed models was conducted with the above data. Also, the prediction results of the proposed models are compared with that of the BP neural network(BPNN) model, which indicate the feasibility and the effectiveness of the proposed prediction models. |
