Research Of The Short-term Traffic Flow Forecasting Based On Vehicular Networks

With the increment of vehicle, urban traffic congestion becomes even worse and leads to many problems such as air pollution, a waste of energy and time, severe haze, and so on. The intelligent traffic control system has been proposed to solve the traffic congestion at home and abroad recently. The internet of vehicles(Io V) is a research hotspot in the intelligent traffic control system. The main contents of the Io V include the internet of things(Io T), the Intelligent Transportation Systems(ITS), the mobile internet and the cloud computing. The Io V can improve traffic condition and will change the way people travel in the future in an intelligent, safe, and environment-harmless way.This research focuses on the implementation of the real time short-term traffic flow forecasting in a more accurate way using the technologies of the Io V and provides the solution of the traffic control in the intelligent traffic control system. The contents of this research includes the following several parts:1. This research is based on the traffic data in California from Pe MS system. we make a Characteristic Analysis on the traffic flow data, and a time-segmented(weekday or weekend) multi-parameters(speed, flow, occupancy) forecasting method was proposed in this research.2. We built the traffic forecasting model using Kalman filtering, then we improved it by using phase-space reconstruction theory. We discussed the difference of basic Kalman forecasting model and the phase-space reconstruction forecasting model by simulation with real data. The prediction error of three parameters(speed, flow, occupancy) by the phase-space reconstruction forecasting model has been reduced by 23%, 1%, 28% on weekday traffic data, and 27%, 1%, 33% on weekend traffic data.3. We built the basic WNN traffic forecasting model. Then we improved it by using the space-time characteristics of traffic flow. We discussed the difference between the basic WNN forecasting model and the space-time WNN forecasting model by simulation with real data. The prediction error of three parameters(speed, flow, occupancy) by the space-time WNN forecasting model has been reduced by 18%, 9%, 8% on weekday traffic data, and 7%, 3%, 11% on weekend traffic data.4. We discussed the difference of four models mentioned above by simulation with real data. We found that the space-time WNN forecasting model made a better performance among four forecasting models. Among three parameters(speed, flow, occupancy), the forecasting result for speed is most accuracy. In addiction, the forecasting result for weekend has a better performance than weekday. In conclusion, this research gave a more effective short-term traffic flow forecasting model.