Abstract
Public transit passengers need guidance during service disruptions. This study proposes an individual-based path (IPR) recommendation model. The model decides which paths to recommend for each passenger with the objective of minimizing system travel time and respecting passengers' path choice preferences. We assume the recommendations could affect passengers' path choice probabilities, but their actual choices are uncertain. This behavior uncertainty makes the problem a stochastic optimization with decision-dependent distributions. We propose a single-point approximation method to eliminate the expectation operator by introducing two new concepts: epsilon-feasibility and Gamma-concentration, which control the mean and variance of path flows in the optimization problem. The approximation yields a tractable single-stage mixed integer linear formulation, which can be solved efficiently with Benders decomposition. The model is implemented in a real-world case study using data from an urban rail disruption in the Chicago Transit Authority system, and a synthetic case study with varied network sizes and demand situations. Results show significant reductions in system travel time compared to status quo and capacity-based benchmarks.
Introduction
Public transit disruptions often lead to chaotic conditions where passengers flock to the most obvious alternative routes, causing secondary bottlenecks and severe congestion. Traditional navigation apps often exacerbate this by providing "selfish" shortest-path recommendations to all users simultaneously. This paper addresses the challenge of system-wide coordination through an Individual Path Recommendation (IPR) model.
The core problem lies in the "compliance myth"—the assumption that passengers will perfectly follow system-optimal instructions. In reality, passengers have heterogeneous preferences and varying levels of trust. Furthermore, the act of giving a recommendation changes the probability of a passenger's choice, a phenomenon known as performative prediction. This creates a feedback loop that makes traditional optimization extremely difficult.
Methodology
The proposed framework balances system efficiency (total travel time) with individual preferences (utility). The model treats passenger choices through a Random Utility Maximization (RUM) framework, where a choice depends on inherent preferences and the "information shock" of a recommendation.
To handle the computational complexity of stochastic optimization over millions of passenger combinations, the researchers introduced a "Single-Point Approximation" method using two key constraints:
- $ε$-feasibility: Ensures actual passenger flow remains close to the expected flow.
- $Γ$-concentration: Constrains the variance of the flow, serving as a risk-management tool to maintain system stability even when passengers deviate from recommendations.
The problem is formulated as a Mixed Integer Linear Program (MILP) and solved using Benders Decomposition. This approach separates the master problem (recommendation assignment) from the subproblem (network flow effects), allowing for real-time solution speeds.
Case Studies and Datasets
The model was validated through two primary lenses:
- Chicago Transit Authority (CTA) Case Study: Utilizing smart card "tap-in" data from a Blue Line suspension on February 1, 2019, the authors compared the IPR model against the observed "status quo" behavior.
- Synthetic 3-Line Network: A controlled environment with varying station counts ($N=2$ to $N=20$) and incident locations to evaluate scalability and performance under different demand intensities.
Results and Discussion
The IPR model demonstrated significant improvements in transit efficiency:
- Travel Time Savings: In the CTA case, the model reduced average travel times by 6.6% compared to the status quo and 4.2% compared to a capacity-based benchmark.
- Demand Sensitivity: Synthetic tests revealed that the model is most effective (up to 15% improvement) in networks with moderate demand. In fully saturated networks, the improvement drops to 1.8% as alternative capacity becomes exhausted.
- Computational Efficiency: The Benders Decomposition allowed the CTA scenario (over 5,800 passengers) to be solved in 17.8 seconds, proving its viability for real-time deployment.
A key insight is that explicitly modeling behavior uncertainty—the fact that people might ignore advice—actually leads to better system outcomes. By not "over-betting" on a single alternative path, the model creates a more robust distribution of passengers across the network.
Conclusion
This research provides a tractable solution for real-time transit management during disruptions. By accounting for human behavior and uncertainty, the IPR model offers a more realistic and effective alternative to traditional routing algorithms. Future work will explore the co-design of supply-side responses (like shuttle bus scheduling) alongside recommendation strategies and address equity concerns to ensure no passenger group is unfairly burdened by system-optimal routing.
Summary updated