Abstract
Ramp metering is a critical tool for managing urban freeway congestion, yet traditional methods like the ALINEA algorithm often struggle with parameter estimation and ramp queue spillover. This paper presents CS-ALINEA, an enhanced control strategy that utilizes real-time traffic flow data and Gaussian Mixture Models (GMM) to adaptively manage congestion. By implementing a segmented queue control mechanism, the system effectively balances freeway efficiency with the need to prevent on-ramp queues from paralyzing local city streets.
Introduction
Urban freeways are the arteries of modern cities, but their efficiency is frequently compromised by bottlenecks at entry points. Ramp metering—the use of traffic signals to regulate the flow of vehicles entering the freeway—has long been used to mitigate this. However, the industry-standard ALINEA algorithm relies heavily on "critical occupancy," a parameter that is notoriously difficult to measure accurately under varying environmental conditions. Furthermore, many existing models are "freeway-centric," prioritizing mainstream flow at the expense of local streets. This often leads to ramp queue overflows that cause secondary gridlock in urban centers.
The CS-ALINEA Approach
To address these challenges, the authors developed CS-ALINEA (Congestion Status based ALINEA). This approach introduces three core innovations designed to make ramp metering more practical and responsive:
- Parameter Substitution: Instead of relying on occupancy, the model uses traffic flow as the primary control variable. This makes the system more intuitive and responsive to the actual number of vehicles merging into the mainstream.
- Adaptive Congestion Classification: The system employs Gaussian Mixture Models (GMM) and the Expectation-Maximization (EM) algorithm to classify traffic into four distinct states: smooth, mild, moderate, and heavy congestion. This allows the control logic to adapt dynamically to real-time traffic conditions.
- Segmented Queue Control: To protect local surface streets, the algorithm monitors ramp queue lengths ($L'$). If the queue exceeds predefined thresholds ($L_1$), the signal timing is adjusted to prioritize clearing the ramp, acting as a "safety valve" for the surrounding city grid.
Methodology and Data
The model was validated using the SUMO (Simulation of Urban Mobility) platform, incorporating the Intelligent Driver Model (IDM) for realistic vehicle behavior. The simulation was grounded in real-world data collected from microwave detectors on the Shangtang freeway in Hangzhou, China. The dataset spanned 10 working days (June 25 – July 6, 2018), focusing specifically on the morning peak hours (5:00 AM to 9:00 AM) at the Daguan intersection.
Key Results
Simulation results demonstrated that CS-ALINEA significantly outperforms traditional and even some modern AI-based methods:
- Reduced Congestion: Compared to standard ALINEA, CS-ALINEA reduced average ramp waiting times by 17.31% (saving approximately 5.82 seconds per vehicle) and average queue lengths by 20.35%.
- Freeway Efficiency: Mainstream traffic throughput saw a 36.19% increase compared to scenarios with no signal control, proving that prioritizing the ramp does not necessarily sacrifice the freeway.
- Comparison with Reinforcement Learning: CS-ALINEA outperformed a Q-learning (Reinforcement Learning) model, reducing waiting times by an additional 4.88 seconds. This suggests that well-tuned classical control theory, when augmented with modern classification techniques, remains highly competitive for infrastructure management.
Conclusion
CS-ALINEA provides a robust solution for urban traffic management by balancing the competing needs of freeway throughput and local street stability. By moving away from static occupancy thresholds toward dynamic flow-based classification, the system offers a more resilient approach to ramp metering. Future research will focus on extending this local control logic to multi-ramp coordinated systems to manage entire freeway corridors effectively.
Summary updated