An Artificial Intelligence Platform for Network-wide Congestion Detection and Prediction Using Multi-source Data


Provide general overview of project in this box and link to any previous or related work.With the rapid growth of urban populations, traffic problems are getting worse in terms of increased traffic congestion and lengthened delays. Urban traffic congestion has become a critical problem that not only affects the people’s daily lives, but also restricts the development of the economy. Smart Mobility Report (2015) indicated that the average American spends about 34 hours every year in traffic. In Europe, the cost of road congestion is estimated to exceed € 110 billion a year, or even be equivalent to around 1% of the GDP, although this varies among countries. In Asia, the average speed of vehicles in Bangkok was 15 km/h, while those in Manila, Jakarta, and Singapore were 18 km/h, 19 km/h, and 20 km/h, respectively. During peak hours, the speed on roads in Delhi and Mumbai drops to 10–20 km/h. Hence, it is urgent to find solutions that can mitigate traffic congestion and reduce traffic delays.

Since urban traffic flow is complex and constantly changing, it is difficult for travelers to acquire information describing current and estimated future traffic conditions for various roadway facilities. As a result, congestion detection and prediction have been proposed to support transportation agencies and help them establish effective traffic management measures, as well as aid road users in their adoption of smarter trip strategies, including route and departure time selection. Understanding how congestion at one location can cause ripples throughout a large-scale transportation network is vital for transportation researchers and practitioners to be able to pinpoint the locations of traffic bottlenecks as points to focus on for congestion mitigation. Some traditional studies rely on either mathematical equations or simulation techniques to model traffic congestion dynamics. However, most of the approaches have limitations, largely due to unrealistic assumptions and cumbersome parameter calibration processes. Ultimately, there are two major challenges in urban traffic congestion detection and prediction: (1) How to estimate and predict traffic congestion in large-scale urban areas? (2) How to improve the accuracy, instantaneous nature, and stability of traffic congestion detection and prediction?


New Developments in Data-Driven Congestion Detection

With the development of data collection technologies, transportation data have become more and more ubiquitous. This has triggered a series of data-driven research projects to investigate transportation phenomena. Some recent studies have proposed data-driven methods for congestion detection and prediction. Typical approaches for congestion detection include Global Positioning System (GPS) trace analysis, use of back propagation (BP) neural networks and Markov models, real-time adaptive background extraction, undedicated mobile phone data analysis, space-time scan statistics (STSS) based non-recurrent congestion (NRC) detection, etc. Several congestion prediction methods have also been developed such as adaptive data-driven real-time congestion prediction, traffic flow prediction using floating car trajectory data, Bayesian network analysis, deep learning theory, data mining based approaches (integration of K-means clustering, decision trees, and neural networks), Hierarchical fuzzy rule-based systems optimized with genetic algorithms, etc. These existing studies have made significant contributions to development of the methodologies and technologies for traffic congestion detection and prediction, but with the development of Intelligent Transportation Systems (ITS) and Internet of Things (IoT) technologies, new challenges and opportunities are continuously emerging with higher requirements for metrics such as detection and prediction accuracy, real-time results, and stability.

Recently, artificial intelligence (AI) has become one of the most promising techniques to tackle tremendously high-dimensional data analysis tasks. AI technologies have been applied for transportation analysis applications such as traffic signal control, network design, pedestrian crossing detection, travel time prediction, short term traffic volume prediction, and car ownership determinants. Specific applications include locating inspection facilities in traffic networks, real-time highway traffic condition assessment, and decision support in real-time for traffic flow management. However, the applications of AI technologies, especially deep learning, are still in their early stages in the transportation area. This proposed research attempts to extend AI technologies into large-scale transportation network analysis.

The research team has already established an on-line transportation platform, named the Digital Roadway Interactive Visualization and Evaluation Network (DRIVE Net). DRIVE NET can be used for sharing, integration, visualization, and analysis of transportation-related data. The proposed research aims to extend the functions of DRIVE Net by developing an AI platform for network-wide congestion detection and prediction using multi-source data.


Research Objectives

The primary objective of this project is to extend the functions of DRIVE Net by developing an AI platform for network-wide congestion detection and prediction using multi-source data. In particular, the research team aims to achieve the following research objectives:

  • Develop new databases by employing the Microsoft Trusted Data Platforms for improving the efficiency of data management.
  • Design an AI platform architecture to fully apply/use big data resources.
  • Develop AI-based analytical models for network-wide congestion detection and prediction.
  • Validate the research findings from the model results of the AI platform with real-world, multi-source data.

To achieve the above objectives, the team plans to work closely with WSDOT and Seattle DOT on a variety of multi-source data access and technology transfer activities. Furthermore, the team will work with industry partners for outreach activities and additional data support.



Yinhai Wang

Professor, UW

Yinhai Wang is the Principal Investigator for this project

Xuegang (Jeff) Ban

Professor, UW

Jeff Ban is a Co-Principal Investigator for this project