An alternative reliability method to evaluate the regional traffic congestion from GPS data obtained from floating cars

Abstract

Abstract Fast and reliable evaluation of regional traffic congestion is beneficial to more effective traffic control. Based on data accumulation in modern society, more and more data‐driven methods are proposed. However, it is still not easy to process the raw data to an interpretable level in practical applications. In this article, the GPS data are obtained from floating cars covering a large scale region in Xi'an, China. To link the original data to the spatiotemporal relationship of driving behaviour, a pre‐processing method with specified time–frequency rules is proposed. Through map matching and landmark mapping, it can be seen that the data dispersion degree has decreased and the quality of the original data has been improved. At the same time, deep learning methods and non‐parametric survival analysis methods are used to compare and evaluate traffic congestion. In addition, four different distributions (Exponential, Weibull, Log‐normal, and Log‐logistic) are tested to fit the accelerated failure time model (AFT), which is then compared with the Cox proportional hazards model (Cox). It is concluded that the most suitable parameter model for the test section of Xi'an South Second Ring Road is AFT (Lognormal). All those methods are tested on a randomly selected segment on the ring road in Xi'an. The results suggest dramatic improvement of data quality and successful evaluation of traffic conditions with high reliability. Potential application could be effective methods for traffic control and management in the smart city.

Keywords

• failure analysis
• Weibull distribution
• Global Positioning System
• smart cities
• time‐frequency analysis
• deep learning (artificial intelligence)