Urban flooding caused by extreme rainfall events disrupts transportation systems, yet generating realistic flood-traffic scenarios for disaster preparedness remains a labor-intensive manual process. This study proposes a Knowledge Graph (KG)-driven pipeline that automatically generates domain-specific training data for fine-tuning small language models (sLLMs) to synthesize urban flood-traffic scenarios. A domain KG comprising 58 entities and 285 relationships was constructed for Jinju City, South Korea, integrating empirical flood data from 112 local documents with quantitative rainfall-traffic impact values from 14 international studies. Nine domain constraint rules, including a novel spatial consistency rule, ensure the physical plausibility of generated scenarios. Through constrained weighted graph walks, 800 semi-structured English narrative scenarios were automatically generated in approximately 5 min, substantially reducing the labor required compared to manual creation. Three sLLMs spanning different architectures and parameter scales—Flan-T5-Large (770M), Qwen2.5-3B-Instruct (3B), and Qwen2.5-7B-Instruct (7B)—were fine-tuned using QLoRA on a single GPU with 16 GB VRAM. Evaluation on 78 test samples demonstrated consistent performance improvements with increasing model scale: Qwen2.5-7B achieved BLEU-4 of 0.5524, ROUGE-L of 0.6883, BERTScore F1 of 0.9662, and KG Fact Consistency of 1.0000, representing a 33.8% BLEU-4 improvement over Flan-T5-Large. Both Qwen models achieved KG Fact Consistency of 1.0000. The 3B model achieved 98.6% of the 7B model’s BLEU-4 at 53% of the VRAM cost with identical factual consistency, representing the most cost-effective configuration. All models were trained for 10 epochs on the same GPU, demonstrating practical feasibility for municipal disaster response deployment.

The rapid expansion of expressway open-toll plazas, where high-speed electronic toll collection lanes operate alongside slower manual and single-lane toll lanes, has introduced complex crash risks that are difficult to capture using traditional crash-frequency models alone. Sparse crash records, heterogeneous operating conditions, and evolving tolling configurations call for data-driven approaches that integrate simulation, surrogate safety metrics, and computational learning. This study aims to perform a comprehensive risk assessment of open-toll plaza configurations by combining surrogate safety measure (SSM) estimations with computational learning (CL)-based accident prediction models. First, a broad set of geometric and operational scenarios is generated by systematically varying lane layout, toll-lane composition, merging length, speed limit, traffic volume, and heavy-vehicle share. For each scenario, microscopic traffic simulations are conducted, and SSMs—including time to collision (TTC), post-encroachment time (PET), and conflict counts—are extracted to quantify instantaneous interaction risk at the vehicle level. These SSMs serve both as comparative safety indicators and as target outputs for subsequent CL models. Next, the CL model is trained to predict conflict frequencies from design and traffic features, while a complementary parametric count model is estimated for benchmarking. To enhance interpretability, eXplainable AI (XAI) attribution techniques are used to decompose the CL predictions into feature-level contributions, revealing nonlinear and interaction effects associated with high-risk operating regimes. The results highlight the dominant influence of traffic volume and toll lane ratio on conflict occurrence. By integrating SSM-based simulation outputs with CL and XAI, the proposed approach provides quantitative and interpretable evidence that supports safer design and operation of expressway open-toll plazas within next-generation data-driven transportation systems.

This study presents a novel integrated framework for autonomous vehicle control at unsignalized intersections in mixed traffic environments, addressing the critical challenge of coordinating Society of Automotive Engineers (SAE) level 4 connected and autonomous vehicles (CAVs) and manually driven vehicles (MVs). The combination of driving intention prediction with a Social Long Short-Term Memory (Social LSTM) and a scheduling algorithm with optimization-driven Pass-through Time Windows (PTWs) is adopted to address traffic flow uncertainty. The Social LSTM model with spatial pooling layers to capture complex multi-vehicle interactions and predict surrounding vehicles’ trajectories and maneuver intentions using naturalistic driving data from the CitySim dataset was applied. Unlike conventional approaches that treat prediction and control separately, this framework leverages high-confidence trajectory predictions to inform proactive scheduling decisions for conflict mitigation. The PTW scheduling algorithm formulates intersection management as a constrained optimization problem, dynamically allocating non-overlapping temporal windows for vehicle entering and exiting while considering vehicle dynamics, safety gaps, and deceleration constraints. Comprehensive simulation analysis across varying traffic volumes and CAV market penetration rates reveals significant improvements in both safety and operational efficiency. The scheduling algorithm has notably reduced traffic delay times while maintaining balance with safety measures. This finding provides a fundamental basis for infrastructure-based cooperative driving research, serving as a contributing factor for the development of advanced traffic management systems during the mixed-traffic period.

Korea operates exclusive bus lanes (XBLs) on many of its expressways. As XBLs convert one lane of regular-use traffic into bus-only traffic, they have a large impact on traffic flow, so careful judgment is required to determine if the operation is effective. However, XBLs have been operated based on only political judgment due to the lack of standards for the operation of expressway XBLs. Therefore, we sought to establish standards for the operation of expressway XBLs using the micro-traffic simulation program VISSIM and the multi-criteria value function methodology. Various scenarios were established based on traffic volume changes of passenger vehicles and bus traffic of four- and five-lane expressway networks. Through an expert survey, the average speed and speed-deviation values, which are the criteria for evaluating operational efficiency and safety, were determined. Also, value points were converted using average speed and speed deviation extracted from the simulation. In addition, quantitative operation standards were established using the converted value scores. Using the results of this study, we established standards for the operation of the XBLs and presented guidelines for related agencies such as police, bus groups, and corporations. The National Police have prepared guidelines for the operation of the XBLs. Citing the results of this study, the new guidelines were implemented in February 2021, and sections of some XBLs have been abolished. Through this study, quantitative standards for the operation of XBLs, one of the management lane techniques necessary for sustainable highway operation, were prepared and applied to actual highways. By properly applying the newly applied guidelines according to quantitative standards, there will be effects of reducing traffic congestion, improving travel time, and enhancing environmental characteristics such as exhaust gas emission. It is also expected to have a positive effect on safety.

The factors that affect the severity of crashes must be identified for pedestrian and traffic safety in urban roads. Specifically, in the case of urban road crashes, these crashes occur due to the complex interaction of various factors. Therefore, it is necessary to collect high-quality data that can derive these various factors. Accordingly, this study collected crash data, which included detailed crash factor data on the huge urban and mid-level roads. Using this, various crash factors including driver, vehicle, road, environment, and crash characteristics are constructed to develop a crash severity prediction model. Through this, this study identified more detailed factors affecting the severity of urban road crashes. The crash severity model was developed using both machine learning and statistical models because the insights that can be obtained from the latest technology and traditional methods are different. Therefore, the binary logit model, a support vector machine, and extreme gradient boosting were developed using key variables derived from the multiple correspondence analysis and Boruta-SHapley Additive exPlanations. The main result of this study shows that the crash severity decreased at four-street intersections and when traffic segregation facilities were installed. The findings of this study can be used to establish a traffic safety management strategy to reduce the severity of crashes on urban roads.