Feng Xiao

This paper revisits the linear adaptive bus holding control framework with three modifications. First, we replace its deterministic dwell-time model with a more realistic M/G/1 boarding process in which passengers who arrive during dwelling can also board. Second, we shift control decisions from bus arrival times to ready-to-depart (RTD) times, the moment boarding is complete, so the control law uses the realized dwell time instead of predicting it. Third, we apply a nonlinear ramp function that keeps the computed holding time nonnegative, so the slack no longer has to be large to guarantee nonnegativity and instead becomes a free tuning parameter. The resulting control subsumes the classical adaptive controls in the literature as special cases. We prove finite-horizon probabilistic stability of the nonlinear RTD-based Simple Control under the coupled M/G/1 boarding model. We derive an explicit slack condition for this stability guarantee. We test these modifications in simulation on both an idealized homogeneous route and a route calibrated to bus line No. 8 in Chengdu, China. Our proposed nonlinear RTD control outperforms both the conventional linear arrival-based control and carefully tuned reinforcement learning agents, the latter even on the headway objective they are trained for. Within our nonlinear RTD framework, traditional schedule-based holding is generally Pareto efficient for schedule adherence despite using no real-time headway information. This contradicts the long-standing view that it is inferior to adaptive control. Therefore, we argue that the proposed nonlinear RTD-based holding controls, despite their simplicity, suffice for effective single-line bunching mitigation.

This study introduces a novel decompose and distilled informer (D2-Informer) framework designed to address the challenge of anomaly detection in highfrequency, long-sequence vibration data from aircraft engines. By employing the variational mode decomposition (VMD), the model effectively decomposes complex vibration signals into intrinsic frequency components, isolating noise and enhancing the detection of key features. Building upon this, the Wav2Vec-inspired convolutional module compresses the decomposed signals, reducing sequence length while retaining critical temporal patterns. This operation ensures that essential features are preserved, aiding in precise anomaly classification. The integration of the informer network enables the capture of long-range temporal dependencies using an advanced self-attention mechanism, optimizing the model’s ability to detect anomalies even in highly imbalanced and complex datasets. The D2-Informer framework excels in both short-term and long-term sequence processing, ensuring high precision and recall in real-world fault diagnosis applications. This model is computationally efficient, requiring minimal retraining, making it suitable for deployment in various industrial contexts. The experimental results demonstrate that the D2-Informer not only outperforms traditional anomaly detection methods but also adapts seamlessly to diverse datasets, offering a robust solution to high-dimensional data challenges without significant computational overhead.

When making route decisions, travelers may engage in a certain degree of reasoning about what others will do in the upcoming day, rendering yesterday’s shortest routes less attractive. This phenomenon was manifested in a recent virtual experiment that mimicked travelers’ repeated daily trip-making process. Unfortunately, prevailing day-to-day traffic dynamic models failed to faithfully reproduce the collected flow evolution data therein. To this end, we propose a day-to-day traffic behavior modeling framework based on the cognitive hierarchy theory, in which travelers with different levels of strategic reasoning capabilities form their own beliefs about lower step travelers’ capabilities when choosing their routes. Two widely studied day-to-day models, the network tatonnement process dynamic and the logit dynamic, are extended into the framework and studied as examples. Calibration of the virtual experiment is performed using the extended network tatonnement process dynamic, which fits the experimental data reasonably well. Our analysis reveals that both extended dynamics exhibit multiple equilibria, one of which corresponds to the classic user equilibrium. We further analyze and characterize these nonuser equilibrium states. Whereas analyzing global stability is intractable because of the presence of multiple equilibria, local stability criteria near equilibria are developed analytically. General insights on how key parameters affect the stability of user equilibria are unveiled. Funding: This work was supported by the National Natural Science Foundation of China [Grants 72201214, 72025104], the Sichuan Science and Technology Program [Grant 2023NSFSC1035], and the Central Guidance for Local Science and Technology Development Fund Projects [Grant 2024ZYD0108]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/trsc.2024.0890 .

Inefficient and unreliable public transportation systems remain a significant challenge in growing cities, with bus bunching being a key contributor to passenger dissatisfaction. Despite numerous proposed holding strategies to mitigate this issue, there is a lack of a standardized testbed for their comprehensive evaluation. This paper presents an open-source, extensible simulation platform that enables the development and benchmarking of bus holding strategies in a unified environment. It accommodates both model-based and model-free reinforcement learning (RL) control strategies, providing a systematic approach to assess their performance under various operating conditions. Holding control strategies can be customized by users within our platform, provided they create a class that fulfills the basic requirements of the exposed application programming interface (API). The platform is designed to be easily extensible, allowing users to incorporate real-world datasets and customize detailed operational features. We demonstrate the platform’s capabilities by comparing three holding strategies: a model-based forward headway control method and two RL-based approaches. Experimental results highlight the importance of comprehensive evaluations, as the relative performance of different strategies varies under different holding time budgets. The proposed simulation platform aims to facilitate more robust, comparable, and reproducible research in bus operation control strategies, ultimately leading to improved bus service reliability in real-world implementations.

Accurately estimating data in sensor-less areas is crucial for understanding system dynamics, such as traffic state estimation and environmental monitoring. This study addresses challenges posed by sparse sensor deployment and unreliable data by framing the problem as a spatiotemporal kriging task and proposing a novel graph transformer model, Kriformer. This model estimates data at locations without sensors by mining spatial and temporal correlations, even with limited resources. Kriformer utilizes transformer architecture to enhance the model's perceptual range and solve edge information aggregation challenges, capturing spatiotemporal information effectively. A carefully constructed positional encoding module embeds the spatiotemporal features of nodes, while a sophisticated spatiotemporal attention mechanism enhances estimation accuracy. The multi-head spatial interaction attention module captures subtle spatial relationships between observed and unobserved locations. During training, a random masking strategy prompts the model to learn with partial information loss, allowing the spatiotemporal embedding and multi-head attention mechanisms to synergistically capture correlations among locations. Experimental results show that Kriformer excels in representation learning for unobserved locations, validated on two real-world traffic speed datasets, demonstrating its effectiveness in spatiotemporal kriging tasks.

Due to the inherent uncertainties of the bus system, bus bunching remains a challenging problem that degrades bus service reliability and causes passenger dissatisfaction. This paper introduces a novel deep reinforcement learning framework specifically designed to address the bus bunching problem by implementing dynamic holding control in a multi-agent system. We formulate the bus holding problem as a decentralized, partially observable Markov decision process and develop an event-driven simulator to emulate real-world bus operations. An approach based on deep Q-learning with parameter sharing is proposed to train the agents. We conducted extensive experiments to evaluate the proposed framework against multiple baseline strategies. The proposed approach has proven to be adaptable to the uncertainties in bus operations. The results highlight the significant advantages of the deep reinforcement learning framework across various performance metrics, including reduced passenger waiting time, more balanced bus load distribution, decreased occupancy variability, and shorter travel time. The findings demonstrate the potential of the proposed method for practical application in real-world bus systems, offering promising solutions to mitigate bus bunching and enhance overall service quality.