Mitigating Bus Bunching with Reinforcement Learning Enhanced by Semantic Stop Embedding

2026-08-10

AI Summary

This study addresses the irregular service and increased passenger waiting times caused by vehicle bunching on high-frequency bus routes. The authors propose an event-driven holding control method that integrates semantic embeddings from large language models (LLMs) with deep Q-learning. By leveraging LLMs to generate offline semantic representations of stops—incorporating physical attributes, surrounding land-use activities, and historical operational patterns—the approach overcomes limitations of conventional methods that rely solely on instantaneous variables or route-specific identifiers. This significantly enhances cross-route policy transferability and control performance. Experimental results demonstrate that, compared to the Daganzo benchmark, the proposed method reduces headway variation by 32.0%, decreases bunching events by 69.2%, and shortens average passenger waiting time by 24.0%. Furthermore, fine-tuning enables faster learning convergence and improved initial performance during policy transfer.

Abstract

Bus bunching degrades service regularity and increases passenger waiting in high-frequency transit. Existing reinforcement-learning-based holding controllers primarily rely on instantaneous operational variables or route-specific stop identifiers, which provide limited information about the functional and operational context of individual stops and constrain policy reuse across routes. This study introduces an LLM-assisted semantic stop representation for event-driven bus holding control. An LLM is used offline to transform heterogeneous stop information, including physical attributes, surrounding activity context, and historical operational characteristics, into fixed semantic embeddings that are incorporated into a deep Q-learning controller without requiring real-time LLM inference. Experiments are conducted in stochastic simulations calibrated with observed data from two bus routes. Compared with the best calibrated Daganzo baseline, the semantic controller reduces headway variability, bunching events, and passenger waiting time by 32.0%, 69.2%, and 24.0%, respectively. A route-specific stop identifier does not improve the spacing-only controller, whereas semantic stop information improves headway regularity, waiting time, and holding effort, providing a more favorable overall trade-off across control objectives. Cross-route experiments further show that zero-shot transfer provides limited immediate generalization, while warm-start fine-tuning accelerates early-stage learning and improves transferred policies; cold-start training nevertheless achieves the best final performance. These findings suggest that semantic state representations can complement conventional operational states and support adaptation-based policy reuse across related transit routes.

Problem

• bus bunching
• reinforcement learning
• semantic representation
• transit control
• policy transfer

Innovation

• semantic stop embedding
• reinforcement learning
• bus bunching mitigation
• LLM-assisted representation
• policy transfer