Jinlei Zhang

With the advancement of ubiquitous connectivity and autonomous intelligence, future urban transportation systems are expected to become increasingly integrated across modes and contexts, necessitating a scalable and generic city-wide urban traffic prediction framework for traffic management. However, most existing studies focus on traffic flow prediction for a specific transport mode under a single scenario or time granularity, limiting their ability to generalize across diverse transport modes and scenarios. To solve these challenges, we propose a large language model (LLM)-based scalable and generic city-wide urban traffic prediction framework (LLM-UTP) for short-term traffic flow prediction. It captures the generic trends and specific fluctuations across different transport modes and scenarios and consists of three parts: a trend data enhancement module, a spatiotemporal feature encoding module, and an LLM module. Extensive experiments conducted on 11 large-scale real-world traffic datasets from 29 cities/areas covering a wide range of transport modes, traffic scenarios, and time granularities to validate the proposed model's complexity, scaling law, scalability, generality, and predictive performance, demonstrating its superiority. These results highlight the practical potential of LLM-UTP as a scalable and generic foundation model for intelligent traffic management and decision-making in future smart cities.

Short-term origin-destination (OD) demand prediction is critical in managing the multimodal transportation system. The joint short-term OD demand prediction for multimodal systems faces three challenges: (1) data availability: real-time OD demand is not available for prediction; (2) sparsity and high-dimensionality of OD demand: the OD demand is spatiotemporal sparse and usually high dimension; (3) impact of different transportation modes: the future OD demand for one mode is affected by others, and extensive studies primarily focus on a single transportation mode, overlooking the influence between different modes. To tackle these challenges, we propose a multitask learning and Partial-Differential-based model to predict the short-term Multimodal Transport Systems OD demand (PD-MTSOD), which includes (1) an OD demand learner to estimate real-time OD demand, (2) data aggregation with hypergraph attention to capture spatiotemporal features, and (3) OD demand decomposition into self-generated increment, other-modes-generated increment, and real-time OD demand, and use partial-differential-based methods to model intermodal correlations. Extensive tests on Beijing and New York city’s multimodal systems show that PD-MTSOD surpasses baseline models. In addition, we prove the benefits of joint considering multiple transportation and explore the correlations of different transportation modes. This paper offers a reliable method for understanding multimodal transportation systems.

Metro stations serve as critical hubs for passenger gathering and scattering. Under disturbing scenarios, a station’s ability to respond to disturbances, named resilience, fundamentally governs the operational stability, sustainability and emergency performance of the metro network. Existing metro network resilience studies typically treated stations merely as topological nodes, making it impossible to account for the internal passenger flow organization and facility capacities of the station. The resilience of the station itself cannot be characterized and quantified. This study focuses on the metro station’s resilience. From the perspective of sustainable operation, considering the passenger flow management of the station, the station’s resilience is defined as the ability of the station to maintain its basic service capabilities and minimize the number of delayed passengers within the station during disturbances. A passenger delay coefficient is introduced to quantify variations in passenger delay volumes within the station. The total number of passengers entering and leaving a station is used to quantify its service capacity. A resilience measurement model for the station is constructed by coupling the passenger delay coefficient and the service capacity. A case study of a transfer station experiencing a sudden passenger surge is conducted for model validation, considering passenger flow control measures and train capacity constraints. The results demonstrate that the model measures the station’s resilience across varying passenger flow management strategies effectively. This study provides a quantitative tool for measuring metro station resilience, enabling emergency responses, operational optimization and policy formulation that support the sustainable and stable operation of metro stations and networks.

Accurate traffic flow prediction is essential for highway operations, especially during holidays when surging traffic poses significant challenges. This study focuses on holiday traffic and introduces a spatiotemporal cross-attention network (ST-Cross-Attn) that combines a bidirectional convolutional LSTM (Bi-ConvLSTM) with a cross-attention module to jointly predict toll station inbound flow and outbound flow. Under the multi-task learning framework, the model shares spatial–temporal features between inbound flow and outbound flow, enhancing their representations and improving multi-step prediction accuracy. Using three years of highway traffic flow data during Labor Day from Shandong, China, ST-Cross-Attn outperformed eight state-of-the-art benchmarks, achieving an average improvement of 4.34% in inbound flow prediction and 2.3% in outbound flow prediction. Extensive ablation studies further confirmed the effectiveness of the model’s components and multi-task learning framework, demonstrating its potential for reliable holiday traffic forecasting.

Due to the rapid development of urban rail transit and the expansion of new lines, accurately predicting the passenger flow of newly expanding subway lines has become increasingly important. The lack of historical data and long forecasting times have led to insufficient accuracy in previous studies when directly predicting inflow at new line stations. To address these challenges, this study proposes a method to decompose inflow features into trend features and scale features, and introduces a model named Semi-Conv-Transformer, based on semi-supervised learning and deep learning neural networks, for prediction of newly expanding subway lines. This innovative approach divides the inflow prediction of newly expanding subway lines into three parts: 1) enhancing the dataset using semi-supervised learning for data augmentation, 2) predicting trend features and scale features with Conv-Transformer deep learning model, and 3) combining the trend features and scale features of station passenger flow to obtain the required inflow data. The proposed method was tested on a new subway line operated in Nanning, China. In this experiment, the model achieved better experimental performance than previous methods and achieved higher accuracy.