Dynamic fusion graph convolutional traffic flow forecasting model with external factors and multi-period features enhanced

Dynamic multi-scale spatial-temporal graph convolutional network for traffic flow prediction

Traffic prediction is a fundamental operation in real-time traffic analysis. A precise prediction of traffic condition can benefit both road users and traffic management agencies. However, since road traffic is decided by multiple static and dynamic factors, traffic prediction is still a challenging task. As the core indicator of traffic condition, many works focus on traffic speed prediction using time-series forecasting approaches. Although current methods take into account the static road topology while modelling, they fail to consider (1) the semantic closeness between road components and (2) congestion caused by upstream/downstream traffic propagation. In this paper, we introduce a Spatial-Temporal Dynamic Graph Network using JS-Graph, which considers both static road features and dynamic traffic flows when forecasting. Specifically, we first propose a data-driven ‘JS-Graph’ method that describes the semantic similarity between road nodes. It models the complex spatial correlations that cannot be captured by the traditional spatial adjacency graph. Secondly, we design a dynamic graph attention network that considers the traffic dynamics that happened in previous time slices when predicting the current one to capture the congestion propagation phenomena. Extensive experiments conducted on real-world datasets show that our proposed method is significantly better than baselines. KeywordsGraph neural networkSpatial-temporal data analysisTime series forecast

To improve the prediction accuracy of traffic flow under the influence of nearby time traffic flow disturbance, a dynamic spatiotemporal graph attention network traffic flow prediction model based on the attention mechanism was proposed. Considering the macroscopic periodic characteristics of traffic flow, the spatiotemporal features are extracted by constructing spatiotemporal blocks with an adjacent period, daily period, and weekly period respectively. The spatiotemporal block is mainly composed of a two-layer graph attention network and a gated recurrent unit to capture the hidden features of space and time. In space, based on considering adjacent road segments, the Pearson correlation coefficient is used to capture the hidden correlation characteristics between non-adjacent road segments according to a certain time step. In terms of time, due to the random disturbance of traffic flow at the micro level, the attention mechanism is introduced to use the adjacent time as the query matrix to weight the output characteristics of daily cycle and weekly cycle, and the three are connected in series to output the prediction results through the linear layer. Finally, the experimental results on the public data sets show that the proposed model is superior to the six baseline models.

Recent research has investigated the effects of var iable speed limits (VSL) on traffic flow processes. Findings show that VSL profoundly impact lane flow distribution (LFD) and increase the use of the slow lane by reducing the speed difference between the slow lane and the overtaking lanes. Finally, VSL reduce the capacity drop phenomenon on multi-lane sections and significantly increase the proportion of flow on the slow lane. These results highlight the major impact of LFD on the ca pacity of multi-lane sections. The paper presents a new analytical model that esti mates the capacity of multi-lane sections and repro duces the capacity-drop caused by LFD characteristics. It can be applied to estimate the capacity of motorway se ctions with two or three lanes under common traffic condit ions; it can also estimate the impact of VSL on the capacity of multi-lane sections. The multi-lane capacity model is coupled with a dyn amic mesoscopic traffic flow model initially design ed for network applications: the meso-LWR model. The resulting dynamic traffic flow model is relevant for mot orway network applications and VSL evaluation. The dynamic model is implemented on a French motorway site. The results show the ability of the model to reproduce actual traffic situations (capacity, congestion per iod, etc.) without VSL, including the amplitude of capacity dr op phenomenon caused by the under-use of the slow lane. It also shows the ability of the model to estimate the effect of VSL on future traffic flow conditions.

Recent advances in traffic monitoring systems have made real-time traffic velocity data ubiquitously accessible for drivers. This paper develops a traffic data-enabled predictive energy management framework for a power-split plug-in hybrid electric vehicle (PHEV). Compared with conventional model predictive control (MPC), an additional supervisory state of charge (SoC) planning level is constructed based on real-time traffic data. A power balance-based PHEV model is developed for this upper level to rapidly generate battery SoC trajectories that are utilized as final-state constraints in the MPC level. This PHEV energy management framework is evaluated under three different scenarios: 1) without traffic flow information; 2) with static traffic flow information; and 3) with dynamic traffic flow information. Numerical results using real-world traffic data illustrate that the proposed strategy successfully incorporates dynamic traffic flow data into the PHEV energy management algorithm to achieve enhanced fuel economy.

Dynamic air traffic flow management (DATFM) aims at flexibly balancing air traffic demand with limited airspace by scheduling aircraft, particularly during unforeseen events, to maintain efficiency in aviation operations. Genetic programming (GP) has shown success in evolving effective heuristics across various domains. However, directly adopting GP to DATFM may be less effective due to the computationally intense simulations required for large-scale aircraft decision-making over broad airspace. To address the issue, we develop a novel multifidelity surrogate-assisted GP. The core idea is that if a computationally efficient low-fidelity surrogate provides enough information to guide the population toward promising areas effectively, then employing more accurate but resource-intensive evaluations would only increase computational effort without enhancing the direction of evolution. A key innovation in our method is a surrogate management strategy that automatically determines when and which surrogate model to use, based on collective information from the evolving population. This approach allows for more effective management of computational resources during the evolutionary process, enabling exploration of a broader range of the heuristic space and increasing the likelihood of identifying promising solutions. The proposed method has been tested on various benchmark instances derived from actual air traffic data. Extensive experimental results demonstrate that the proposed algorithm significantly outperforms current state-of-the-art methods in both effectiveness and efficiency.

Lane reservation strategies are widely used to ensure the right of way for eco-friendly vehicles and encourage people to green their commute. In most lane reservation problems (LRPs), the parameters underlying the system traffic conditions (e.g., vehicle speed and traffic flow) cannot be effectively specified. This paper addresses a new dynamic lane reservation problem (DLRP), which aims to optimize lanes that need to be reserved in different time periods based on trajectory data for existing transit network optimization. For passengers in reserved lanes, their travel time is minimized to guarantee traffic priority. Considering that the reserved lanes cause travel time growth on regular lanes, an improved multiobjective mixed integer nonlinear programming (MINLP) is established to minimize the delay. The problem complexity of this paper is NP-hard. This paper applies a preprocessing method for the actual traffic flow data analysis for link travel time calculation. We developed a hybrid evolutionary algorithm decomposing a multiobjective optimization problem (MOP) to a collection of simple MOPs. These subproblems are collaboratively solved. The hybrid crossover strategy exploits the advantages of different crossover operators for a better performance. The experimental results of MOEA/D and NSGA-II with standard test functions show that the proposed algorithm can improve the convergence and distribution of the results. Through numerous analyses and calculations of instances, the proposed algorithm is proven to be effective.

Rear-end and side-impact crash risks are the two principal types of multi-vehicle crash risk on freeways. Most previous studies examine a single crash risk type, limiting understanding of their combined effects. This study employs a multilevel structural equation modelling (SEM) framework to investigate the sequential and joint impacts of roadway geometry, dynamic traffic flow, and driving behaviour on multi-type crash risks. The framework was calibrated using 1,762 rear-end and 1,243 lane-changing conflicts from 14 directional sites. The multilevel SEM accounts for site-level heterogeneity to produce more robust estimates. The path analysis identifies two dominant causal chains: 'Horizontal Curve - Density - Car-following Behaviour - Crash Risk' and 'Vertical Slope - Speed Distribution - Lane-changing Behaviour - Crash Risk'. Low-speed fluctuating traffic flow shows higher crash risks than high-speed stable traffic flow. Car-following behaviour increases both rear-end and side-impact risks, while lane-changing activity raises side-impact risk but reduces rear-end risk.

A method to identify input-dependent jumps from the execution of obfuscated machine code is presented. Input-dependent jumps, which are defined as jumps whose target addresses can be changed depending on the input, correspond to decision points in program execution. By investigating how a target address is calculated, it is possible to pinpoint the triggering conditions of a given behavior, and new execution paths can be discovered by finding input values that change the target address. Obfuscators hinder such analysis by inserting numerous artificial jumps that use opaque predicates with constant values into the code. One important obfuscation approach is virtualization-obfuscation, in which entire blocs of control flow information are replaced with bytecode interpreter code. Using the fact that the semantics of the original program must be preserved under obfuscation, we propose an obfuscation mitigation approach that exploits the relationship between the original and obfuscated executions using dynamic data flow graphs that represent output computation using concrete and symbolic information. These graphs are generated from execution traces that are recorded using dynamic binary instrumentation and simplified using pattern-based rules based on algebraic identities and the general properties of well-behaved programs. To identify input-dependent jumps, a dynamic data flow graph is generated and simplified for each write access to the program counter; if the node for the target address is reachable from a node for an input value in the resulting graph, the jump is input-dependent. Experimental application of the proposed approach to code treated with various obfuscators successfully revealed the relationship between input-dependent jumps in the original and obfuscated executions, confirming that information obtained from dynamic data flow graphs is useful in understanding branch conditions.

DDoS attacks are one of the most dangerous threats to network security. It may easily flood systems and hinder valid services, which will result in a huge amount of damage in terms of operations and money. The proposed paper includes a machine learning-based system, using real-time dynamic flow data for detection and classification of DDoS attacks. The CatBoost model was used for binary classification on whether there was an attack and multi-class classification on the type of one of the attacks-a SYN flood, UDP flood, and HTTP flood. This was attained on several train-test splits: 80- 20, 70-30, and 60-40. In all these instances, the system did well, with its highest accuracy being 92.19%. This would enable real-time detection and classification, along with high-resolution measurement of time-based metrics, packet-level statistics, and behavioral indicators to tailor mitigation measures. The paper demonstrates the use of this method of machine learning in the case of network security by building a scalable and practical means of counteracting the increasingly prevalent DDoS attack threat.

This study explores the impact of driving behaviors, specifically cautious and aggressive, on the performance of human-driven vehicles (HDVs) and autonomous vehicles (AVs) in traffic flow dynamics. It focuses on various metrics, including level of service (LOS), average speed, traffic volume, queue delays, carbon emissions, and fuel consumption, to assess their effects on overall performance. The findings reveal significant differences between cautious and aggressive AVs, particularly at varying market penetration rates (MPRs). Aggressive autonomous vehicles demonstrate greater traffic efficiency compared to their cautious counterparts. They achieve higher levels of service, improving from poor performance at low MPRs to significantly better performance at higher MPRs and in fully autonomous scenarios. In contrast, cautious AVs often experience poor service ratings at low MPRs, with an improvement in performance only at higher MPRs. Regarding environmental performance, aggressive AVs outperform cautious ones in terms of reduced emissions and fuel consumption. The emissions produced by aggressive AVs are significantly lower than those from cautious AVs, and they further decrease as the MPRs increases. Additionally, aggressive AVs show a considerable reduction in fuel usage compared to cautious AVs. While cautious AVs improve slightly at higher MPRs, they continue to generate higher emissions and consume more fuel than their aggressive counterparts. In conclusion, aggressive AVs offer better traffic efficiency and environmental performance than both cautious AVs. Their ability to improve road efficiency and reduce congestion positions them as a valuable asset for sustainable transportation. Strategically incorporating aggressive AVs into transportation systems could lead to significant advancements in traffic management and environmental sustainability.

Road vulnerability is crucial for enhancing the robustness of urban road networks and urban resilience. In medium or large cities, road failures in the face of unexpected events, such as heavy rainfall, can affect regional traffic efficiency and operational stability, which can cause high economic losses in severe cases. Conventional studies of road cascading failures under unexpected events focus on dynamic traffic flow, but the significant drop in traffic flow caused by urban flooding does not accurately reflect road load changes. Meanwhile, limited studies analyze the spatiotemporal pattern of cascading failure of urban road networks under real rainstorms and the correlation of this pattern with road vulnerability. In this study, road vulnerability is calculated using a network’s global efficiency measures to identify locations of high and low road vulnerability. Using the between centrality as a measure of road load, the spatiotemporal patterns of road network cascading failure during a real rainstorm are analyzed. The spatial association between road network vulnerability and cascading failure is then investigated. It has been determined that 90.09% of the roads in Zhengzhou city have a vulnerability of less than one, indicating a substantial degree of spatial heterogeneity. The vulnerability of roads adjacent to the city ring roads and city center is often lower, which has a significant impact on the global network’s efficiency. In contrast, road vulnerability is greater in areas located on the urban periphery, which has little effect on the global network’s efficiency. Five hot spots and three cold spots of road vulnerability are identified by using spatial autocorrelation analysis. The cascading failure of a road network exhibits varied associational characteristics in distinct clusters of road vulnerability. Road cascading failure has a very minor influence on the network in hot spots but is more likely to cause widespread traffic congestion or disruption in cold spots. These findings can help stakeholders adopt more targeted policies and strategies in urban planning and disaster emergency management to build more resilient cities and promote sustainable urban development.

Refining the classification of combined alignment sections on mountainous freeways and analyzing the spatio-temporal effects on crash frequency.

The coupling between electrified highway traffic network and distribution network makes the traffic flow and power flow more complicated. The dynamic traffic flow in highway traffic network presents the characteristics of multi-temporal distribution and congestion. Dynamic vehicle-flow coupled distribution network has strong randomness, uncertainty and dispersion. Therefore, a method to divide highway energy community based on the density and correlation of highway traffic network and power network is proposed herein. Firstly, the coupling model of highway traffic network and power network is established based on the electric-hydrogen integrated energy supply station. Secondly, the highway energy community is defined, and its measurement index is given based on synergetic theory. Finally, a method of dividing highway energy community structure based on density and correlation is proposed. Simulation results show that the proposed method is effective.

SummaryTraditional macroscopic traffic flow modeling framework adopts the spatial–temporal coordinate system to analyze traffic flow dynamics. With such modeling and analysis paradigm, complications arise for traffic flow data collected from mobile sensors such as probe vehicles equipped with mobile phones, Bluetooth, and Global Positioning System devices. The vehicle‐based measurement technologies call for new modeling thoughts that address the unique features of moving measurements and explore their full potential. In this paper, we look into the concept of vehicular fundamental diagram (VFD) and discuss its engineering implications. VFD corresponds to a conventional fundamental diagram (FD) in the kinematic wave (KW) theory that adopts space–time coordinates. Similar to the regular FD in the KW theory, VFD encapsulates all traffic flow dynamics. In this paper, to demonstrate the full potential of VFD in interpreting multilane traffic flow dynamics, we generalize the classical Edie's formula and propose a direct approach of reconstructing VFD from traffic measurements in the vehicular coordinates. A smoothing algorithm is proposed to effectively reduce the nonphysical fluctuation of traffic states calculated from multilane vehicle trajectories. As an example, we apply the proposed methodology to explore the next‐generation simulation datasets and identify the existence and forms of shock waves in different coordinate systems. Our findings provide empirical justifications and further insight for the Lagrangian traffic flow theory and models when applied in practice. Copyright © 2015 John Wiley & Sons, Ltd.