TL;DR
This paper introduces a novel simulation-based optimization (SO) framework that efficiently integrates high-resolution traffic and emissions models for optimizing urban traffic signals. By combining detailed microscopic simulations with a differentiable macroscopic approximation, the method significantly improves computational efficiency, enabling the design of traffic signal plans that simultaneously reduce travel time and environmental pollution.
What Problem Does This Paper Address?
Urban transportation agencies face increasing pressure to mitigate vehicular emissions, but current practices often involve evaluating a limited set of pre-determined traffic management strategies rather than systematically designing them with environmental goals in mind. This is largely due to the computational expense of high-resolution traffic and emissions models, which are crucial for accurately capturing complex factors like individual vehicle behavior and instantaneous speed/acceleration. Existing optimization techniques struggle to handle these computationally intensive models within practical time budgets, leading to a gap in tools that can directly optimize for environmental metrics alongside traditional traffic performance.
Methods
The core of this paper's contribution is a metamodel simulation-based optimization (SO) approach. This framework leverages the strengths of both microscopic and macroscopic models:
- Microscopic Models: High-resolution traffic simulators (like Aimsun) and emissions models (based on Panis et al.) provide detailed, albeit computationally expensive, insights into vehicle behavior and emissions.
- Macroscopic Models: A differentiable, analytical approximation of the microscopic emissions model is derived. This approximation, while lower-resolution, provides crucial structural information and analytical tractability.
- Metamodel: The SO algorithm constructs a metamodel that combines information from both the microscopic simulator and the analytical macroscopic model. This metamodel acts as an efficient surrogate for the complex objective function.
- Simulation-Based Optimization: The algorithm iteratively refines the metamodel using simulation observations and then uses the metamodel to derive trial points for optimization. This process allows for efficient exploration of the solution space, even with computationally demanding simulators.
The key innovation lies in using the analytical structure provided by the macroscopic model to guide the optimization process, making the computationally inefficient microscopic models usable within a tight budget.
Datasets
The study utilizes a calibrated microscopic model of the Swiss city of Lausanne for the evening peak period (5-6 pm). This model represents individual vehicles and includes approximately 12,400 trips generated from an origin-destination matrix. The network consists of 47 roads and 15 intersections, with 9 signalized intersections.
Key Insights and Ideas
- Bridging the Gap: The paper demonstrates that combining detailed, high-resolution simulation data with the analytical structure of lower-resolution macroscopic models is a powerful strategy for efficient optimization.
- Analytical Structure is Key: Embedding analytical information from macroscopic models into SO algorithms significantly enhances computational efficiency, even when the macroscopic model's direct approximation of the objective function is not perfectly accurate.
- Beyond Evaluation: The proposed framework moves beyond simply evaluating pre-determined schemes to actively designing traffic signal plans that optimize for both traffic flow and environmental impact.
- Monetary Value of Emissions Reduction: The study quantifies the significant monetary savings achievable by implementing optimized signal plans, highlighting the economic benefits of emissions mitigation.
- Link-Level Improvements: The analysis shows that optimized signal plans lead to reductions in average emissions per vehicle at the link level, contributing to overall network-wide improvements.
Key Results
The proposed method, denoted "Am," consistently outperformed other methods ("Aφ" which used only microscopic simulation, and "AfA" which used only the macroscopic analytical model) in a case study of the Lausanne city network.
- Performance Improvement: Signal plans generated by Am yielded systematic improvements across all five performance metrics (average travel time, CO2, NOx, VOC, and PM emissions) compared to the initial signal plan.
- Superiority over Aφ: Am significantly outperformed Aφ, demonstrating the added value of incorporating analytical information from macroscopic models.
- Superiority over AfA: Am also outperformed AfA, highlighting the importance of using detailed simulated observations for optimization.
- Computational Efficiency: The Am method achieved these improvements within a tight computational budget of 150 simulation runs, a stark contrast to typical genetic algorithm approaches that require tens of thousands of runs.
- Monetary Savings: The best Am signal plan, when compared to the existing Lausanne signal plan, showed significant monetary savings, ranging from 200 to 700 EUROs for CO2 emissions and 0 to 300 EUROs for travel time during a single evening peak hour.
Limitations and Future Work
- Non-differentiable Emissions Models: The paper acknowledges that microscopic emissions models are often non-differentiable. While a differentiable macroscopic approximation is formulated, this simplification might introduce some inaccuracies.
- Single Vehicle Type Assumption: For simplicity, the study assumes a single vehicle type. The authors note that this assumption can be relaxed to include different vehicle types.
- Future Work: The authors suggest extending the framework to address dynamic SO problems, such as traffic-responsive signal control and dynamic traffic management strategies that account for spatio-temporal variations in emissions. They also plan to investigate combining analytical models, simulation-based models, and data-driven approaches for sustainable transportation problems.
Summary updated