Road closures resulting from construction activities significantly disrupt traffic flow and often divert heavy truck traffic onto alternative routes. These detours, if not properly planned, can funnel trucks onto roads not designed for high axle loads, accelerating pavement deterioration and increasing maintenance needs. To address this challenge, this research proposes a comprehensive bilevel modeling framework that integrates vehicle-type-specific traffic flow prediction with detour signage placement optimization. Employing a user equilibrium traffic assignment model solved by paired alternative segments algorithm, the study predicts traffic distributions before and after road closures, identifying road segments vulnerable to increased truck traffic. A genetic algorithm is utilized to solve the bilevel optimization framework to strategically determine optimal detour signage placement. The primary objective is to minimize a fitness function that incorporates both infrastructure considerations and network efficiency measures, prioritizing the restriction of heavy trucks from vulnerable or prohibited road segments while accounting for network total travel times. The model is demonstrated through application to the entire Minneapolis transportation network, illustrating its capability to handle complex, large-scale real-world scenarios effectively. Results show that the optimized signage placement significantly reduces truck usage of vulnerable roads while maintaining efficient traffic operations, offering valuable insights for policymakers and planners aiming to enhance resilience and sustainability in detour management strategies.

Self-driving vehicles are envisioned as automated and safety-focused vehicles facilitating smooth movement on roads. This research proposes a novel, robust, and intelligent navigation framework for such vehicles through an integrated fusion of advanced technologies like predictive analytics with remote sensing and detection for accurate obstacle/object detection. TaskTrek, ViewVerse, and RuleRise form the core of the essential model governing vehicle-environment interaction. TaskTrek handles kinematic trajectory synthesis and space-time traffic modeling, ViewVerse provides LiDAR-based volumetric perception and radar-assisted navigational intelligence, and RuleRise manages topological localization, vehicle actuation, and autonomous decision-making through multimodal sensory fusion. The model applies an iterative Multi-FacBiNet method, which uses the cognitive Fully Convolutional Neural Network (FCNN) method to detect and classify obstacles during vehicle movement on the road. Upon stimulation during vehicle movement, the model provided an encouraging outcome. The fusion of predictive intelligence, Radar, and sensing technologies gave 95.3% proficiency. Minimum obstacle detection, processing, and response delays of 0.116 seconds, 0.105 seconds, and 0.36 seconds, respectively, are recorded. The computed mean obstacle detection accuracy for right, left, front and back camera angles are 88.3%, 83.8%, 91.4%, and 89.9%, respectively. Further, a comprehensive analysis of the model’s performance in different on-road scenarios considering metrics like traffic load, road type, and region density was done. The model generated a very impressive accuracy of obstacle detection on all parameters. The results of this study not only aid in accelerating the development of precise navigation-enabled self-driving vehicles but also in the context of environmentally friendly mobility/motion tracking solutions.

Global cloud service providers handle inference workloads for Large Language Models (LLMs) that span latency-sensitive (e.g., chatbots) and insensitive (e.g., report writing) tasks, resulting in diverse and often conflicting Service Level Agreement (SLA) requirements. Managing such mixed workloads is challenging due to the complexity of the inference serving stack, which encompasses multiple models, GPU hardware, and global data centers. Existing solutions often silo such fast and slow tasks onto separate GPU resource pools with different SLAs, but this leads to significant under-utilization of expensive accelerators due to load mismatch. In this article, we characterize the LLM serving workloads at Microsoft Office 365, one of the largest users of LLMs within Microsoft Azure cloud with over 10 million requests per day, and highlight key observations across workloads in different data center regions and across time. This is one of the first such public studies of Internet-scale LLM workloads. We use these insights to propose SageServe , a comprehensive LLM serving framework that dynamically adapts to workload demands using multi-timescale control knobs. It combines short-term request routing to data centers with long-term scaling of GPU VMs and model placement with higher lead times, and co-optimizes the routing and resource allocation problem using a traffic forecast model and an Integer Linear Programming (ILP) solution. We evaluate SageServe through real runs and realistic simulations on 10 million production requests across three regions and four open-source models. We achieve up to 25% savings in GPU-hours compared to the current baseline deployment and reduce GPU-hour wastage due to inefficient auto-scaling by 80%, resulting in a potential monthly cost savings of up to $2.5 million, while maintaining tail latency and meeting SLAs. The workload traces, our simulator harness and the SageServe scheduler are available at https://github.com/shashwatj07/SageServe.

LLT: A lane-level traffic flow prediction model optimized by multiple strategies for complex road network environments

Hopf bifurcation analysis and control of traffic flow model based on speed limit and lane change information of networked vehicles

Bifurcation Analysis and Control of Traffic Flow Models Considering Driver Perception of Conflict Factors

Adaptive traffic prediction model using Graph Neural Networks optimized by reinforcement learning

A systematic review of machine learning-based microscopic traffic flow models and simulations

Abstract The concept of contraflow left‐turn lane (CLL) design has been proposed for nearly 10 years, which provides a novel approach to alleviate traffic congestion in urban areas, especially for those signalized intersections with heavy left‐turn traffic. While putting forward the practical application of CLL design to area‐wide signalized intersections, whether the control scheme of large‐scale signalized intersections with CLLs could be more sustainable remains an open question. This paper analyzes spatiotemporal characteristics of the CLL system and proposes analytical traffic and emission models based on the finite capacity queuing model, in which the special queuing behavior brought by the pre‐signal is explicitly considered. A sustainable traffic signal control framework is built to optimize signal timings of intersections with CLLs. A real‐world case study based on a road network in Yangon is conducted, and the results illustrate the proposed method's efficiency and sustainability in managing signalized intersections with CLLs in road networks.

Phase transitions and Lyapunov analysis of a heterogeneous traffic model involving human-driven and connected autonomous vehicles integrating overtaking effect

Abstract This study proposed a deep reinforcement learning framework based on the Proximal Policy Optimization (PPO) algorithm to address highway on-ramp merging control in mixed traffic environment. Firstly, a mixed traffic flow model incorporating the dynamic characteristics of both Automated Vehicles (AVs) and Human-Driven Vehicles (HDVs) on the SUMO simulation platform is established. By integrating the classical Gipps model into the reward function, the framework enables dynamic evaluation of safe gaps for AVs while simultaneously incorporating efficiency and comfort metrics, thereby constructing a multi-objective optimization model. Then, the PPO algorithm is applied to derive the optimal on-ramp merging solution. Finally, experiments under different traffic densities of the main road are conducted and the evaluation results demonstrate that the proposed framework can significantly reduce rear-vehicle collision risk, enhance driving comfort, and improve on-ramp merging efficiency comparing to the rule-based methods. The superiority of the proposed framework is more remarkable under high-density traffic condition of the main road. The results provide an effective solution for a coordinated scheme for on-ramp merging control in mixed traffic environment.

Reservoir computing (RC) has gained attention as an efficient machine learning method for time series prediction because of its low computational costs and simple learning process. Herein, we propose the Harvested Reservoir Computing (HRC) framework which treats complex real-world dynamics as spontaneously emerging physical reservoirs. As an instance of HRC, we introduce Road Traffic Reservoir Computing (RTRC), whereby dynamical traffic flow patterns are harnessed as natural computational resources to predict future traffic states in experiments. Unlike conventional reservoir computing, this approach requires no explicit reservoir design, but instead "harvests" the intrinsic dynamics of traffic as a physical reservoir. Experiments using a scaled traffic model and numerical simulations on a grid road network demonstrate that the framework's prediction accuracy is highly dependent on traffic density. An optimal density range is identified within which prediction performance is maximized because of a tradeoff between nonlinearity and short-term memory. These findings highlight the potential of complex real-world dynamics as viable components within computational frameworks.

AGBi-Mamba: a traffic flow prediction model based on adaptive graph convolution and bidirectional Mamba networks

Traffic state forecasting plays a critical role in developing effective traffic control and management strategies. While machine learning (ML) approaches have become popular, owing to automatic spatio-temporal feature extraction, traditional ML approaches fail to generalize and adapt to unseen datasets, limiting their practical applicability. To boost generalization, researchers are increasingly turning to few-shot adaptation techniques, such as meta-learning, which focuses on learning how to learn and enabling rapid adaptation to unseen datasets using limited data. This study applies the Model-Agnostic Meta-Learning framework to a multi-dimensional spatio-temporal graph attention-based traffic prediction model (M-STGAT), producing a new model, called Meta M-STGAT. The goal is to improve forecasting performance through faster adaptation to unseen time periods. This study uses open-access traffic speed and lane closure data from the California Department of Transportation Performance Measurement System and corresponding weather data from the National Oceanic and Atmospheric Administration’s Automated Surface Observing System. Meta M-STGAT is compared against state-of-the-art traffic state forecasting models, including traditional M-STGAT, a multi-dimensional graph attention network, and a multi-dimensional long short-term memory network. Model performance is evaluated for 30-, 45-, and 60-min prediction horizons on one primary and three transfer datasets. Results show that Meta M-STGAT consistently outperforms all alternative state-of-the-art models across all transfer datasets and prediction horizons. The findings underscore the potential of meta-learning in enhancing traffic state forecasting and its practical implications for traffic management systems.

Data collected over networks arise in a number of scientific, engineering and industrial applications, in which the datapoints are noisy observations relating to a process of interest over the graph structure. In this article we propose a novel multiscale representation of data on the edges of a network. In contrast to other methods in the literature which employ expensive node to edge data transformations, our decomposition acts directly on the network edges. Using our method, we propose an efficient edge denoising algorithm, termed E-LOCAAT, which displays good performance across a range of data scenarios, particularly when the number of edges is large. The proposed method is illustrated using extensive simulations and we demonstrate its applicability on a real-world dataset arising in road traffic modeling.

In this paper we extend the Aw–Rascle–Zhang (ARZ) non-equilibrium traffic flow model to take into account the look-ahead capability of connected and autonomous vehicles (CAVs), and the mixed flow dynamics of human-driven and autonomous vehicles. The look-ahead effect of CAVs is captured by a non-local averaged density within a certain distance (the look-ahead distance). We show, using wave-perturbation analysis, that increased look-ahead distance loosens the stability criteria. Our numerical experiments, however, showed that a longer look-ahead distance does not necessarily lead to faster convergence to equilibrium states. We also examined the impact of spatial distributions and the market penetrations of CAVs and showed that increased market penetration helps to stabilize mixed traffic while the spatial distribution of CAVs has less effect on stability. The results revealed the potential to use CAVs to stabilize traffic and may provide qualitative insights into speed control in the mixed autonomy environment.

Abstract Traffic assignment serves as an important component in modeling flow distribution across infrastructure networks and supporting intelligent traffic management and urban planning. Fast algorithms for solving the stochastic user equilibrium (SUE) model are essential for enhancing computational performance and scalability of traffic assignment models applied to complex infrastructure networks. We augment the gradient projection (GP) algorithm for the SUE models through Barzilai–Borwein (BB) step size adaptation. For further optimizing computational performance, we explore iteration strategies within the GP algorithm: Jacobi parallelization, Gauss–Seidel sequential updating, and successive over‐relaxation (SOR) with dynamic relaxation. Global convergence of these iterative methods in solving the SUE problems is theoretically established, with convergence conditions for the SOR derived. The results demonstrate the BB step size's superior performance, compared to alternative step size methods, across all network scales, and it achieves the best stability when demand factors and the dispersion parameter increase.

Traffic flow prediction is an important part of intelligent transportation system, which can provide important support for traffic management optimization, resource scheduling and urban planning. At its core, it uses historical and real-time data to predict future traffic conditions. The main challenge of traffic flow forecasting is how to effectively model complex spatio-temporal dependencies in traffic data. However, with the continuous development of traffic forecasting technology, traffic forecasting models still have obvious shortcomings in dynamic spatial dependence and periodicity modeling, which limits their forecasting effect. Therefore, this paper proposes an improved traffic flow prediction model DASTFormer (Dynamic Adaptive Spatio-Temporal Transformer). Firstly, an adaptive adjacency matrix is added to the embedding layer to enable the model to dynamically adjust the relationship weights between nodes, thereby enhancing the adaptability of the model to the spatial dependence of real-time changes. Secondly, an Adaptive Weighted Adapter is designed and combined with semantic spatial attention to further optimize the model’s semantic dependency modeling in complex traffic scenarios. Finally, adding the Auto-Correlation and then combining the self-attention mechanism into the time prediction module can effectively capture the periodic features in the data, and also improve the modeling ability of the model for complex temporal relationships, thus further enhancing the accuracy and stability of the prediction. Experimental results show that the proposed method achieves better prediction accuracy and robustness on multiple traffic datasets.

Abstract Rapid urbanization, increased motorization, and industrialization have led to ever-increasing levels of Traffic-Related Air Pollution (TRAP), which has significant implications for public health and urban sustainability. This systematic review assesses the application of Geographic Information Systems (GIS) to model vehicle emissions and the related health impacts in urban areas. This review is based on literature published between 1990 and 2024. We screened 4,780 peer-reviewed articles and 780 met inclusion criteria. We examined the computational methods used in impact studies, including data from spatial datasets, pollutant variables, and epidemiological data. The most common methods were geo-statistical interpolation (Kriging, Geographically Weighted Regression), Land-Use Regression (LUR), and machine learning (Support Vector Regression, Neural Networks), typically with California Line Source Dispersion Model (CALINE) and Community Multiscale Air Quality Model (CMAQ). To pull multiple analytical perspectives, we purposefully combined systematic review methods with techniques of bibliometric analysis using VOS-viewer and R-software, allowing us to the research output and trends, collaborative networks and research themes. Ultimately our mixed-methods approach demonstrated important differences between developed and developing contexts regarding data availability, exposure modeling approaches and the integration of health co-benefits from active transport. Building on these findings, we introduce a GIS-based decision-support framework integrating traffic data, remote sensing, pollution modeling and health monitoring into a real-time, open-access platform to assist with evidence-based urban planning. This review, emphasizing the computational tools to create high-resolution exposure maps and better translate policy into practice, advances the field of computational urban science and provides a reproducible framework for ameliorating pollution-related health impacts in their best-case scenario rapidly urbanizing cities.

In a connected vehicle environment, connected and autonomous vehicles (CAVs) can coordinate with each other to navigate intersections without the need for traffic signals. To enhance the traffic efficiency of CAVs at unsignalized intersections, this paper proposes a distributed strategy equilibrium algorithm based on game-theoretic grid coordination. First, the intersection is divided into multiple grids, with vehicles in each grid forming a game group. These groups are connected through virtual logic lines, creating a virtual logical network. Each game group locally optimizes its strategy while engaging in cross-grid games with adjacent groups, ultimately forming a distributed coordination game strategy that optimizes utility functions toward Nash equilibrium. Subsequently, a genetic algorithm is employed to search for the optimal strategy set in a multi-objective optimization problem. This study utilizes Python and SUMO for co-simulation of the game strategies, with comparative experiments set up to verify the model and algorithm’s effectiveness. The results demonstrate that, compared with the traditional First Come, First Serve algorithm and the Mixed Integer Linear Programming algorithm, the proposed algorithm significantly improves average delay time, vehicle passing speed, and energy consumption. Specifically, the proposed algorithm reduces the average delay time by 46.94%, increases vehicle passing speed by 15.5%, and lowers energy consumption by 20%, highlighting its potential to enhance traffic efficiency and reduce delays at unsignalized intersections.