Finite Capacity Queueing for Traffic Spillback Modeling

21 concepts17 papers1 proposal

Traditional traffic macro-models and signal control algorithms often rely on a convenient abstraction: the point queue (or vertical queue) model. In this paradigm, vehicles waiting at an intersection take up zero physical space. They stack vertically, as if the road possesses an infinite reservoir to store idling cars.

While this assumption makes network optimization mathematically elegant, it ignores a disruptive physical reality. In actual urban networks, roads have finite physical dimensions. When a queue of vehicles grows long enough to fill a road segment entirely, it overflows into the upstream intersection. This phenomenon—known as spillback or queue spillover—fundamentally alters traffic dynamics, turning localized delays into network-wide gridlock.

To design resilient urban infrastructure, researchers have shifted toward modeling frameworks that replace infinite buffers with finite-capacity queues. Understanding why this shift is necessary requires looking at how physical queues break classical traffic assumptions and how new mathematical tools capture these constraints.

The Mechanics of Spillback and Intersection Coupling

In an infinite-buffer model, the traffic flowing out of an upstream intersection depends entirely on two things: the upstream demand and the local signal timing. The downstream road is assumed to be a passive receiver, always capable of accepting more vehicles.

When finite capacity is introduced, this independence vanishes. As explored in Scalable Adaptive Traffic Light Control Over a Traffic Network Including Turns, Transit Delays, and Blocking, a full downstream lane creates a hard "blocking effect." Once a queue reaches the upstream boundary of a link, vehicles can no longer enter it, regardless of whether their traffic light is green.

Infinite Buffer Model (Vertical Queue):
[Intersection A] ------> [Intersection B] (Vehicles stack vertically; never blocks A)

Finite Capacity Model (Horizontal Queue & Spillback):
[Intersection A] ===[CRITICAL QUEUE]===> [Intersection B] (Queue blocks A from discharging)

This blocking mechanism couples adjacent intersections in a backward-propagating wave. In dense urban grids where intersections are separated by only a few meters, Enming Liang et al. note that this queue spillback can occur near-instantaneously. If a control system is too slow to recognize that a link is full, it will continue to feed vehicles into a dead end, causing the blockage to cascade upstream.

This spatial constraint is a universal bottleneck in network flows. For example, in transit networks, Yueying Huo et al. demonstrate that a vacant bus berth can be rendered completely useless if the physical entry path leading to it is blocked by another vehicle. Similarly, in industrial logistics, Yanqiong Zhang et al. show that throughput maximization models for narrow environments must explicitly bound the number of active transit units based on the physical dimensions of the channels to prevent total system stagnation.

The Mathematical Cost of Physical Realism

If point queues are so unrealistic, why did early traffic models rely on them so heavily? The answer lies in computational tractability. As Victoria Guseva et al. demonstrate, modeling traffic with point queues allows researchers to formulate the user equilibrium problem as a standard, convex linear program that can be solved efficiently even for large networks.

Introducing physical queue lengths destroys this convexity. Walid Krichene et al. explain that when horizontal queuing and spillback constraints are added to System Optimal Dynamic Traffic Assignment (SO-DTA) formulations, the state space becomes highly non-convex. The derivative of the flow with respect to density changes discontinuously when a link transitions from free-flowing to completely blocked, making global optimization significantly more difficult.

To restore mathematical rigor without sacrificing physical accuracy, researchers use finite capacity buffers as dynamic couplers within macroscopic conservation laws. In A macroscopic traffic flow model with finite buffers on networks: well-posedness by means of Hamilton–Jacobi equations, Alexander Keimer proves that by lifting standard traffic flow equations into Hamilton-Jacobi equations, one can rigorously model these finite buffers. When a buffer reaches its capacity limit, it naturally restricts the incoming boundary flow, forcing the mathematical model to propagate the congestion wave backward across network nodes.

Traditional Model:  Density increases ──> Flow increases (up to capacity) ──> Flow plateaus
Spillback Model:    Density increases ──> Queue overflows ──> Effective capacity drops ──> Flow drops to zero

Accounting for these physical queues also changes how we calculate a road's capacity over time. In Modeling the residual queue and queue-dependent capacity in a static traffic assignment problem, Hao Fu et al. show that "residual queues"—the vehicles left over from a previous congested period—physically occupy lane space and actively shrink the effective capacity available for incoming traffic in the next period. This aligns macroscopic models with real-world sensor data, capturing the severe drop in flow that occurs during gridlock.

Designing Spillback-Aware Control Systems

When traffic signal controllers assume infinite link capacities, they optimize for local throughput, often inadvertently triggering network-wide failures. For instance, a controller might extend a green light to clear an intersection, unaware that the departing vehicles will immediately get stuck in a saturated downstream link, blocking a cross-street.

To prevent this, modern control frameworks embed finite capacity constraints directly into their optimization loops using three primary strategies:

1. Event-Driven Perturbation Analysis

Instead of trying to predict exact future traffic arrivals, controllers can optimize signal timings by observing discrete network events. Yanfeng Geng et al. and Ying Qing Chen et al. utilize Infinitesimal Perturbation Analysis (IPA) to calculate the sensitivities of system delays relative to green-light durations. By explicitly tracking when queues empty and when blocking events occur, the IPA controller adaptively adjusts "green waves" across multiple intersections, scaling linearly with network size while actively mitigating spillbacks.

2. Constrained and Penalized Optimization

Another approach is to treat physical lane capacity as a hard boundary condition. In the online signal control system TUSERACT, developed by Srishti Dhamija et al., sample-based constrained optimization is used to explicitly prevent vehicle lines from exceeding block lengths.

In decentralized environments where intersections cannot communicate globally, Vinzenz Tütsch et al. introduced the Coordinated Maximum Pressure-plus-Penalty (CMPP) method. This policy modifies traditional maximum pressure control by adding a severe mathematical penalty term that activates as a downstream lane approaches its physical capacity limit, shifting green time away from paths that would cause an imminent spillover.

3. Integrating Spillback into Behavioral Routing

Finally, managing finite capacity requires changing how we model driver behavior. When choosing lanes or routes, drivers are highly sensitive to downstream conditions. As Ruolin Li et al. demonstrate when modeling lane choices at complex highway diverges, assuming infinite capacity on exit links leads to highly inaccurate behavioral predictions. Incorporating downstream spillback effects reveals that a backup on a single exit drastically alters the utility of upstream lanes, forcing a redistribution of traffic long before vehicles reach the bottleneck.

By moving away from the mathematical convenience of infinite point queues and embracing finite capacity queueing, these models provide the precision needed to understand, predict, and ultimately prevent urban gridlock.

Key Foundational Work

Accounting for congestion and spillbacks in fixed-time traffic signal optimization: an analytical queueing model approach (2008) — contributed 2 concept elements to this direction.
Dynamic Congestion and Tolls with Mobile Source Emission (2012) — contributed 1 concept elements to this direction.
Multi-intersection Traffic Light Control Using Infinitesimal Perturbation Analysis (2012) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_spillback_modeling_gap_0ec63d7e,
  title = {spillback modeling gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spillback-modeling-gap/rdr_0ec63d7e24864d3da59ebf0234f9afc1},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_spillback_modeling_gap_0ec63d7e_v1,
  title = {spillback modeling gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spillback-modeling-gap/rdr_0ec63d7e24864d3da59ebf0234f9afc1/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake