Artificial intelligence-based vehicular traffic flow prediction methods for supporting intelligent transportation systems

In recent years, the Intelligent transportations system (ITS) has received considerable attention, due to higher demands for road safety and efficiency in highly interconnected road networks. As an essential part of ITS, traffic prediction can provide support in many aspects, such as road routing, traffic congestion control, etc. To provide a more comprehensive overview of the role of traffic forecasting in ITS systems, we will first introduce the corresponding ITS applications and discuss how traffic forecasting can improve the performance of these applications. Next, we will introduce the general prediction procedure as well as some basic concepts of traffic flow prediction, followed by a description of a general framework for implementing the traffic flow prediction. In this survey, mainly two sorts of prediction methods are focused, statistics-based and machine learning (ML)-based. These two types of approaches are more used in ITS traffic flow predictions these years, and service for different contexts. Generally speaking, the statistics-based models have better model interpretability, but the rigorous model structure limits the adaptability, while ML-based models are more flexible. Accordingly, we will introduce the characteristics of these two types of methods through specific examples of state-of-the-art approaches. Last but not least, some potential and meaningful development directions corresponding to this domain are introduced to do a great favor for future research.

Duke Scholars

Author Peng Sun DKU Faculty