One-Shot Traffic Assignment with Forward-Looking Penalization

TL;DR

The paper tackles efficient traffic assignment under real-time conditions by moving from purely individualistic routing to a cooperative, one-shot framework. It introduces METIS, which combines Forward-Looking Edge Penalization (FLEP), k-most diverse near-shortest paths (KMD), and a pattern-aware route scoring to balance route diversity with high-capacity edges. Empirical results across Milan, Florence, and Rome show substantial CO$_2$ reductions (18–46%) while maintaining modest computation times, outperforming several state-of-the-art one-shot baselines. The work demonstrates that penalizing edges anticipated to be used by nearby vehicles and favoring less-popular, high-capacity routes can distribute traffic more effectively and reduce environmental impact in urban transportation systems.

Abstract

Traffic assignment (TA) is crucial in optimizing transportation systems and consists in efficiently assigning routes to a collection of trips. Existing TA algorithms often do not adequately consider real-time traffic conditions, resulting in inefficient route assignments. This paper introduces METIS, a cooperative, one-shot TA algorithm that combines alternative routing with edge penalization and informed route scoring. We conduct experiments in several cities to evaluate the performance of METIS against state-of-the-art one-shot methods. Compared to the best baseline, METIS significantly reduces CO2 emissions by 18% in Milan, 28\% in Florence, and 46% in Rome, improving trip distribution considerably while still having low computational time. Our study proposes METIS as a promising solution for optimizing TA and urban transportation systems.

Figures (8)

• Figure 1: Graphical representation of the bipartite network of road edges and areas. $K_{\text{road}}^{\text{\tiny (source)}}$ is the in-degree of edge nodes, $K_{\text{road}}^{\text{\tiny (end)}}$ is the out-degree of edge nodes.
• Figure 2: FLEP with $p=0.1$ applied to road network $G$, resulting in penalized network $H$. Grey circles represent estimated vehicle positions. FLEP applies cumulative penalization to edges based on the vehicles' expected traversal, with a multiplicative factor of $(1+p)$. Darker red color indicates higher penalties imposed on road edges. For example, edge $e_4$ is traversed by vehicles $c_3$ and $c_4$, leading to a penalty of $(1+p)^2$, resulting in $w(e_4) = 72 \cdot (1.1)^2 = 87.12$.
• Figure 3: Routes generated by KMD (a) and METIS (b) in Milan for 150 trips. METIS exhibits a more spatially uniform distribution of traffic than KMD, which tends to concentrate routes on highly popular routes. RC indicates the road coverage and RED the time redundancy (5-minute window).
• Figure 4: Comparison of METIS (black bar) with the baselines in Florence, Milan, and Rome on CO2 emissions (in tons), road coverage (in %), and time redundancy. To ensure statistical reliability, we run non-deterministic algorithms (GR, PR, KMD, PP, PLA, KD) ten times and present the average values and the corresponding standard deviation.
• Figure 5: Pearson correlation ($r$) between time redundancy (5-minute window) and CO2 emissions. Black dots represent TA algorithms. The grey dashed line represents the curve fit.