Forecasting short-term traffic conditions in real-time

Taking advantage of Advanced data analytics: Real-time demand calibration/prediction is critical to ensuring that real time network traffic simulation models can forecast short-term traffic conditions and provide reliable transport operations decision support. That (linked) project has been completed and its final report s available for downlaod below.

Background

Traffic simulation models can simulate and predict traffic flows across entire metropolitan areas in real-time, empowering network managers and transport authorities to predict and alleviate road congestion under recurrent and non-recurrent events.

However the reliability and accuracy of these models depends on the quality of the inputs as well as the ongoing calibration and validation of the model parameters.

In collaboration with the Department of Transport and Main Roads (Queensland), Aimsun and Queensland University of Technology, this project aims to improve travel demand calibration and accuracy of 24 hour/ 7 days network simulation models (live and offline) for the AM and PM peak hours.

Using advanced data science techniques on real traffic monitoring data, the Queensland node of the iMOVE project focussed on modelling computationally-efficient and reliable travel demand inputs and driving behaviour parameter adjustment for a real-time traffic prediction and management in urban networks.

The project will integrate and evaluate the solutions in the Aimsun Live real-time transport network simulation and demonstrate the resulting performance in the two collaborating jurisdictions of Queensland and Western Australia.

Objectives

The objectives of this iMOVE project are to enhance short-term prediction performance in traffic simulation models, and to decrease the effort involved in model maintenance. It will utilise smart sensing for enhanced travel demand estimation and prediction, as well as artificial intelligence and machine learning for calibration against much larger real-time datasets.
The project will exploit the availability of extensive traffic monitoring datasets and utilise advanced data science techniques to address the calibration needs for simulation models. The development and testing of the project solutions will be first evaluated in the Aimsun Live testbed.

The objectives of this research are to develop Aimsun modules to:

• Establish a library of traffic states and OD matrices derived from a real-time data source
• Improve the short-term traffic demand matching model for real-time simulation
• Enrich the route choice calibration for simulation for the pilot system
• Deploy the developed solution into the pilot environment
• Demonstrate the transferability of the Queensland node solution into the Western Australian node

Aimsun Live methodology

Aimsun Live’s methodology takes advantage of both real-time and historical data.

An historical dataset is leveraged to establish a library of OD matrices and traffic states. The data is also used to train a machine learning based traffic forecasting model, for recurrent traffic conditions.

The traffic forecasting model is applied on real-time data for short-term analytical prediction and to identify OD matrices needed for the real-time simulation.

The simulation models (previously calibrated and validated) are then applied in real time for traffic prediction for the next hour.

Learnings fron the R&D

The project used flow and speed data from various sources, coupled with modelling section specified by Aimsun. Learnings included the need for data pre-processing steps to ensure mapping integration.

Aimsun Live requires a repository of traffic states and OD demand, for which traffic patterns are initially identified from the historical datasets. To address the need to consider multiple data sources with different spatial temporal information, an advanced data science based multi-view learning framework was proposed.

Aimsun adopted a bi-level OD estimation technique for dynamic OD adjustment, offering two distinct setups for adjusting OD intervals: simultaneous adjustment and rolling horizon adjustment. For the Aimsun Live process, Aimsun employs the simultaneous method, which was found to be time efficient as it adjusts all intervals at once.

The QUT team identified that the empirical traffic assignment has the potential of replacing the dynamic traffic assignment in the bi-level OD adjustment process. The developed empirical assignment technique underwent limited testing, hence it is highly recommended to extend the experimentation to assess the stability and performance of the technique over different days and scenarios.

When applying the OD adjustment technique, QUT noticed the demand factors for the first adjusted interval were disproportionately scaled compared to the subsequent intervals. It determined that the demand simulated during the warm-up period was not factored into the Aimsun Next OD adjustment process, and recommended incorporating an additional warm-up demand interval before the actual simulation period.

Benefits and future improvements

In this research, an advanced data science-based, multi-view learning framework is proposed for traffic pattern identification, which utilises both flow and speed information. This new approach is promising and, subject to further testing, can be considered in future Aimsun Live processes.

Among the two new OD adjustment techniques used, the bi-level OD adjustment process was explored in detail, while the Key-Frame Interpolation (KFI) method was tested at small scale. Performance wise, both provide similar results to the dynamic OD adjustment method Aimsun Next currently deploys. However, the KFI method has shown the potential for major run-time savings. Future research will focus on detailed testing of the KFI method but, overall, there is not enough evidence to support changing Aimsun Next’s current dynamic OD adjustment process.

QUT proposes a machine learning based traffic forecasting model, yet Aimsun Live currently feature a powerful analytical traffic forecasting module that performs a similar role. After comparing the performance of the two approaches, the current Aimsun Live analytical forecasting algorithm is considered satisfactory and no change is required.

Empirical traffic assignment is another exciting element of this research. The dynamic OD adjustment with empirical assignment would be much faster than the current method. However, empirical traffic assignment testing was limited, mostly due to time constrained. It also applied some hard assumptions, such as non-capacity constrained and requiring complete traffic state information. The feature should focus on ways to mitigate these limitations.

Expected project impacts

Real-time network prediction and response simulation is a desirable future capability for network operations, and understanding the capabilities of Aimsun Live and the opportunities in model maintenance explored by QUT, has been beneficial in raising our understanding of the implications of such technologies.

Ted Beak, Principal Advisor, Statewide Network Operations branch,

Department of Transport and Main Roads (Queensland)

Download the final report

Download your copy of the final report, iMOVE project 1-027: Transport Predictive Solution – Stage 2 – R&D – QLD Node: Advanced data analytics for real-time demand calibration and prediction in large scale networks , by clicking the button below.