Building and Calibration of Microscopic Simulations Models for the Assessment of Traffic Control Strategies

🚀Project overview

This project develops and calibrates a microscopic traffic simulation in SUMO (Simulation of Urban Mobility) to assess the performance of different traffic signal control strategies on Plymouth Road, Ann Arbor (MI, USA). The study covers 11 signalized intersections along the corridor, integrating field traffic counts with a detailed digital roadway network built from OpenStreetMap and GIS data.

The model was calibrated against observed traffic volumes using the GEH statistic, achieving >85% of movements with GEH < 5, ensuring high fidelity between simulated and real-world traffic behavior.

📃Methodology

Corridor Geometry Development

The Plymouth Road corridor network was developed using OpenStreetMap (OSM) data and converted to a SUMO-compatible network with the osmWebWizzard.py script. The initial network was refined by:

• Removing unnecessary elements like polygons, bicycle lanes, pedestrian paths, and streets outside the corridor.
• Adjusting intersections manually using Google Maps satellite imagery, ensuring accurate lane counts and geometry.
• Correcting unrealistic U-turns and removing redundant nodes.

This process produced a high-fidelity network representing all 11 intersections along the corridor.

Traffic Demand Generation

Traffic flows were generated using JTRrouter, based on field traffic counts for each movement at the intersections. Key steps included:

1. Defining turning ratios: Field counts were converted into probabilities for each movement using .turn.xml files.
2. Defining network exits: Specified exit edges to ensure vehicles leave the network correctly.
3. Defining traffic flows: Used .flow.xml files to define entry points and vehicle generation probabilities per second.

JTRrouter was executed via command line, producing routes compatible with SUMO.

Data Collection for Model Calibration

• Induction loop detectors were placed on all intersection lanes to monitor simulated traffic.
• Data cleaning involved combining through and right-turn movements where detection was imprecise.
• Model calibration was assessed using the GEH statistic, targeting ≥85% of movements with GEH < 5.

Traffic Signal Control Plans

Four signal control strategies were implemented:

1. Fixed-time coordinated plan (entire corridor)
2. Fixed-time coordinated plan (corridor divided into two sections)
3. Actuated control (uncoordinated)
4. Actuated coordinated control

PTV VISTRO was used to optimize the fixed-time plans:

• Objective function: Minimize delay and number of stops.
• Optimization method: Genetic algorithm adjusting cycle times and phase splits.
• Actuated plans: Derived from best-performing fixed-time plan, extending phases 150–200% to allow dynamic responsiveness.

Real-Time Simulation and Metrics

Simulations were run in SUMO with TraCI (Traffic Control Interface) and Python for real-time data collection. Performance metrics included:

• Intersection delay: Average vehicle delay per intersection.
• Travel time: Time to complete a route.
• Lost time: Time spent driving below ideal speed.
• Number of stops and stop time for each vehicle.
• Space-time diagrams: Visualizing vehicle trajectories and traffic patterns.

Data were analyzed separately for the main corridor and secondary streets to assess localized performance.

📊Results

For the complete set of performance metrics and intersection-level results, refer to the complete document. An English version will be available soon.

Fixed-Time Plans

• Single coordinated plan (entire corridor):
  While synchronization across all intersections helped reduce stops along the mainline, congestion persisted in sections with uneven demand, leading to localized delays.

• Two-section fixed-time plan:
  Splitting the corridor improved local synchronization and mitigated bottlenecks in the busiest zones. However, performance remained limited during variable demand conditions.

Actuated Plans

• Uncoordinated actuated plan:
  Reduced delays at lightly trafficked intersections but increased variability in travel times due to lack of coordination. Particularly problematic for through traffic on Plymouth Road.

• Coordinated actuated plan:
  This was the most effective strategy overall. It dynamically adapted to fluctuations in demand while maintaining progression along the corridor.

  • Reduced average vehicle delay by ~45% relative to fixed-time.
  • Lowered lost time and smoothed travel time distributions, reducing variability.
  • Improved performance on both primary and secondary roads, offering more balanced operations.

Additional Observations

• While the coordinated actuated plan combines responsiveness with corridor-level synchronization, results show that it does not always minimize delay at every single intersection.

• The coordinated plan prioritizes flow along the main corridor, which explains why it performs better on metrics such as lost time, travel time, number of stops, and stop time along Plymouth Road. This is clearly seen in the space-time diagrams, where vehicles under coordinated control maintain more consistent speeds and experience fewer interruptions.

• The trade-off is on side streets: coordinated control increases delay and waiting time for minor approaches. As a result, in some intersections, the uncoordinated actuated plan may show lower average delay.

In summary, actuated coordinated control is the most effective strategy for the main corridor, maximizing traffic flow and reducing key performance indicators. However, this comes at the cost of increased delay on side streets, which explains why intersection-level delay alone can sometimes suggest that uncoordinated actuation performs better.

🗂️Repository Layout

this-repo/
├── models/
│   ├── fixed_time_full/          # SUMO XML + control logic + extras
│   ├── fixed_time_split/         
│   ├── actuated/                 
│   └── actuated_coordinated/     
├── img/ 
├── results/
│   ├── simulation-stats/         # distribution histograms, boxplots, etc.
│   └── space_time_diagrams/
├── full-report-ES                # thesis document
└── README.md

For the complete set of performance metrics and intersection-level results, refer to the complete document. An English version will be available soon.

👤Autor

Daniel Avila, MSc in Mobility Engineering.

📄License

This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.