Evaluation of Public Transit Systems under Short Random Service Suspensions: A Bulk-Service Queuing Approach

Baichuan Mo, Li Jin, Haris N. Koutsopoulos, Zuo-Jun Max Shen, Jinhua Zhao

Introduction

Public transit systems (PTSs) are the backbone of urban mobility, yet they are rarely in a state of perfect equilibrium. In major metropolitan areas like Chicago, the rail system can experience dozens of minor incidents daily—ranging from signal glitches to passenger-related delays—most of which last less than five minutes. While individually brief, these short-term suspensions create significant ripple effects, leading to increased headways, "bus bunching," and severe passenger overcrowding.

Traditional transit modeling often struggles to capture these dynamics accurately. Many models assume infinite vehicle capacity, ignoring the reality that passengers are often left behind when a vehicle is full. Others focus exclusively on single stations, failing to account for how a delay at an upstream stop propagates through the entire route. This research addresses these gaps by developing a novel stochastic framework that combines bulk-service queuing theory with a two-state Markov process to evaluate system performance under random disruptions.

Methodology

The researchers modeled the PTS using a bulk-service queue approach, where passengers are served in "batches" corresponding to the vehicle's remaining capacity. The framework is built upon several key methodological innovations:

1. Two-State Markov Process

To represent the stochastic nature of disruptions, the vehicle’s speed profile is modeled as an impulse function toggling between two states:

  • Normal State: The vehicle moves at a constant operational speed.
  • Disruption State: The vehicle is completely stopped due to a random suspension.

2. Headway Distribution Modeling

The authors demonstrated that under random suspensions, the headway (the time interval between consecutive vehicles) follows a distribution defined by the difference between two compound Poisson-exponential random variables. To ensure the model remains computationally tractable for calculating wait times, they approximated this complex distribution as a zero-inflated truncated normal distribution.

3. Route-Level Dynamics

Unlike isolated station models, this framework uses a Markov chain to track inter-station passenger flows. This ensures that the available capacity at any given station accurately reflects the number of passengers who boarded or alighted at all preceding stops, capturing the cumulative nature of transit delays.

4. Interpolation-Based Root Solving

Solving for queue lengths in bulk-service models requires finding all complex roots of a characteristic equation within a unit circle. The authors proposed a new algorithm utilizing interpolation to identify these roots more reliably than traditional optimization methods, which often struggle with convergence in highly non-linear scenarios.

Validation and Results

The analytical model was validated through a numerical case study of a 10-station bus route and a custom simulation involving 50,000 vehicle runs. The results demonstrated high consistency between the closed-form analytical equations and the simulation benchmarks.

Key Findings

  • Stability Thresholds: The research derived a closed-form stability condition ($\rho < 1$). It quantifies how incident rates and average durations directly reduce a system's effective capacity. If the incident-adjusted arrival rate exceeds the service rate, the queue grows indefinitely.
  • Variance Propagation: Headway variance is not static; it grows as a vehicle progresses down the line. Consequently, downstream stations are inherently more vulnerable to the ripple effects of upstream delays.
  • Wait Time Inflation: At congested stations, wait times do not simply increase by the duration of the delay. Instead, they increase non-linearly due to the "capacity constraint" effect, where passengers are forced to wait for multiple vehicle arrivals because the first arriving vehicles are already at capacity.
  • Impact of Incident Frequency: In test scenarios, increasing the incident rate from zero to 1/3 per minute at a specific station nearly doubled the expected queue length, from 4.5 to 8.3 passengers.

Conclusion and Future Work

This paper provides a robust mathematical toolkit for transit agencies to quantify system resilience. By treating headway as a difference of compound Poisson variables, the model captures the reality that vehicles can "catch up" to their predecessors (bunching) but cannot pass them.

However, the model currently assumes a "no overtaking" rule and a binary speed profile (full speed or stopped). Future research could expand this framework to include partial disruptions (e.g., slowing down for traffic or weather), passenger reneging (passengers leaving the queue during long delays), and complex network structures involving transfers between multiple routes.

Summary updated

BibTeX
@misc{mo2023evaluation,
  author = {Baichuan Mo and Li Jin and Haris N. Koutsopoulos and Zuo-Jun Max Shen and Jinhua Zhao},
  title = {{Evaluation of Public Transit Systems under Short Random Service Suspensions: A Bulk-Service Queuing Approach}},
  year = {2023},
  eprint = {2301.00918},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2301.00918},
}
Figure