Optimising rule-based variable speed limit control

The iMOVE project Optimising motorway control algorithms: New technologies/data has developed a data-driven framework that makes Variable Speed Limit (VSL) systems more adaptive and easier to manage. Using real traffic data, the framework automatically calibrates VSL parameters to reduce manual effort and improve road safety.

The project was led by the Queensland Department of Transport and Main Roads (TMR), in collaboration with the Queensland University of Technology (QUT), and Transmax.

Background and motivation

Effective motorway control plays an important role in maintaining safe and efficient traffic flow, and VSL systems are one of the most widely used tools to achieve this. Traditional rule-based VSL systems rely on predefined activation and deactivation thresholds for speed and occupancy to detect congestion and apply upstream speed reductions. While advanced Reinforcement Learning (RL) methods have been explored in research, rule based VSL remains the best approach because of its simplicity, transparency, and lower computational demands.

However, manual calibration of these rule-based parameters is labour intensive and less adaptive to changing traffic conditions. Moreover, loop detectors on motorways typically spaced around 500 metres apart, provide limited spatial resolution, making it difficult to accurately identify the back of queues and predict congestion propagation.

This project sought to address these limitations by developing a systematic, data-driven framework for calibrating rule-based VSL parameters using real world traffic data.

Project phases

Phase 1: Motorway Control Algorithm Documentation

The first phase involved documenting existing motorway control strategies and their operational characteristics. The team analysed current rule based VSL algorithms used in Queensland’s managed motorways, focusing on how thresholds and control parameters are currently determined. This helped establish the baseline for comparison and identified the limitations of manual calibration methods, such as subjectivity and inconsistency between sites.

Phase 2: Sensitivity analysis
Next, a sensitivity analysis was conducted to understand how variations in the key VSL parameters; activation thresholds, deactivation thresholds, and queue growth rate parameters affect overall system performance. This analysis highlighted how even small deviations in threshold values can impact congestion detection accuracy, queue protection efficiency, and traffic stability, emphasizing the need for automated, data-driven calibration.

Phase 3: Enhanced traffic state estimation

This phase improved how congestion is detected and visualised. Since the detector data is often sparse, the project used an Adaptive Smoothing Method (ASM) to estimate traffic conditions between sites. Combined with Edge Detection (ED), it captured sudden speed drops marking the start or end of congestion, giving a clearer picture of how traffic conditions evolve.

Phase 4: Data-driven calibration framework

In the final phase, the project developed the data-driven calibration framework that automates the tuning of VSL parameters. This framework integrates two key components:

• A Speed-Based Congestion Detection module that uses the Smoothed Difference of Squared Speeds (SDSS) to dynamically detect congestion onset and dissipation.
• An Adaptive Smoothing Method with Edge Detection (ASM-ED) module to estimate realistic back-of-queue locations and congestion growth patterns.

These modules allow the VSL system to automatically identify the appropriate activation and deactivation thresholds, and calibrate queue propagation parameters with minimal human input.

Methodology: The VSL framework

The framework analyses speed data from motorway loop detectors to spot when traffic slows or clears. It calculates the Smoothed Difference of Squared Speeds to detect sudden changes, identifying when congestion starts or ends. Using long-term speed and occupancy data, it applies Kernel Density Estimation (KDE) to find the most typical conditions that trigger or clear congestion, setting accurate activation and deactivation thresholds for VSL.

To track how queues form and spread, the Adaptive Smoothing Method with Edge Detection (ASM-ED) creates detailed congestion profiles. These are compared with results from the Congestion Detection-Protection (CD-CP) algorithm, which estimates queue growth using two parameters: average vehicle length and time headway. An optimisation process automatically fine-tunes these values for better accuracy and less manual work.

Implementation and results

The framework was tested on the Centenary Highway inbound corridor in Brisbane; a 7 km stretch between Bullockhead Street and Centenary Bridge with 17 detector sites and 9 VSL gantries. Using 20-second aggregated detector data from 2023, the system applied smoothing and lane selection to prepare high-quality inputs.

The SDSS-based calibration found activation thresholds near 50 km/h and 25% occupancy, and deactivation thresholds around 57 km/h and 15% occupancy. Optimisation of 26 congestion events showed an average headway of 1.5 seconds and vehicle length of 3.8-3.9 metres, consistent across methods. This automated approach takes the place of manual calibration and gives road operators a more consistent and scalable way to manage congestion and set VSL parameters.

Discussion and conclusions

This study shows that a data driven calibration framework can automatically fine-tune rule based VSL parameters, providing a reliable alternative to manual methods. Using historical traffic data, it identifies activation, deactivation, and queue growth thresholds through the SDSS and ASM-ED methods.

The calibrated parameters improve congestion detection and queue protection, reducing rear-end collision risks and enhancing traffic flow stability. The approach retains the simplicity and transparency of rule-based systems while making them more adaptive to real-world traffic.

The framework can also be expanded to other motorway corridors and tested under varying traffic and weather conditions, supporting safer and more adaptive road management.

Download the final report

Download your copy of the final report, Optimising Rule-Based Variable Speed Limit Control – A Data-Driven Approach for Parameter Calibration, by clicking the button below.

This work was completed in mid-2025.