Journal of Transportation Engineering, Part A: Systems

Accurate traffic forecasting on citywide networks is one of the crucial urban data mining applications that accurately provide congestion warning and transportation scheduling. While previous work has made significant efforts to learn traffic temporal dynamics and spatial dependencies, two key limitations exist in current models: (1) most existing approaches solely capture spatial correlations among neighbors on predefined graph structure, and genuine relation may be missing due to the incomplete graph connections; and (2) existing methods are defective to capture the temporal trends because the recurrent and stacking structure employed in these methods suffers from the long-range temporal dependency vanquish problem. To overcome the difficulty in multistep prediction and further capture the dynamic spatial–temporal dependencies of traffic flows, we propose a new traffic speed prediction framework for multiscale graph attention networks (MS-GATNs). In particular, MS-GATNs is a hierarchically structured graph neural architecture that learns not only the local region-wise geographical dependencies but also the spatial semantics from a global perspective. Furthermore, a multiheads attention mechanism is introduced to empower our model with the capability of capturing complex nonstationary temporal dynamics. Experiments on real-world traffic data sets demonstrate that MS-GATNs outperforms the state-of-the-art baselines in long-term forecasting.

To build more accurate and realistic freeway car-following models, driving characteristics specific to freeway car following should be considered. This study, therefore, analyzed three car-following models calibrated for different driving styles. A total of 5,900 freeway car-following events were extracted from 161,055 km of driving data collected in the Shanghai Naturalistic Driving Study (SH-NDS) database. Based on the fuzzy inference system built in this study, these car-following events were categorized as representing one of two styles: nonaggressive or aggressive. The two driving styles were visualized by using the t-distributed stochastic neighbor embedding (t-SNE) algorithm. The Gipps, Wiedemann, and intelligent driver model (IDM) car-following models were calibrated and validated for each driving style group. Using a genetic algorithm to analyze the calibrated parameters of the investigated car-following models, it was found that the model parameter values were related to driving style. When their performances were evaluated, results showed that the IDM performed best. The nonaggressive IDM and the aggressive IDM were used to simulate the car-following scenarios based on the same leading vehicle trajectories. The t-test and the F-test results showed that regarding both time gap and spacing gap, the differences of mean and variance are significant between aggressive and nonaggressive styles in nearly all of the simulated car-following scenarios. The mean spacing gap (nonaggressive: 38 m; aggressive: 30 m) and time gap (nonaggressive: 1.7 s; aggressive: 1.4 s) obtained from modeled car-following scenarios could be used directly in simulation software to show the characteristics of different driving styles.

This study aims to model vehicle headways comprising non-free flow headways (i.e., headways under non-free flow traffic conditions) and free flow headways (i.e., headways under free flow traffic conditions) on port collector-distributor roads using the finite mixture distribution method with two mixture components. Three combinations of leader-follower vehicles were found to be statistically different. Six factors, consisting of truck percentage, traffic flow rate, weather, curbside parking, light condition, and road pavement condition were considered as influencing factors. Model comparison results indicate that the finite mixture distribution model outperforms the single distribution model. Results show that rainy days, curbside parking, poor light condition, and poor pavement performance are associated with an increased mean headway for car-car, car/truck, and truck-truck types. The headways of these three types are generally higher for the bigger traffic flow rate, while a larger truck percentage can decrease the mean headway for car-truck and truck-truck types.

The Highway Capacity Manual 6th Edition (HCM6) provides methods for evaluating freeway segments and facilities, including weaving segments. Recent research has cast doubt on the validity of some elements of the HCM6 weaving methodology. The core objectives of this research were to identify, document, and address key deficiencies in the current HCM6 weaving method through an improved modeling framework and its calibration. This new framework starts with the basic freeway segment speed model and incorporates a speed impedance factor that represents the effect of weaving turbulence. This approach proved to be simpler and more accurate than the current HCM6 models when tested with field measurements. The speed and capacity prediction models were developed for ramp weaves, avoiding the use of intermediate models to predict the number of lane changes and speeds by weaving maneuvers. The study found that the new speed prediction model is more accurate, while the HCM6 model underestimates the speed within the weaving section compared to field data. A sensitivity analysis showed that the new speed and capacity models behaved properly to changes in segment length, weaving flows, and overall demand. The proposed framework is applicable to the analysis of merge and diverge segments and can ensure continuity in predicting the operational performance across different freeway segment types.

Wrong-Way Driving (WWD) crashes have higher fatality rate than other crashes. In contrast to much of the existing literature, which has focused on collision data available after a WWD crash, this paper evaluates WWD events that did not result in a collision. The aim is to see if there are factors beyond those identified on collision-based studies leading to such events. The focus is on WWD events for highway exit ramps and evaluating the efficacy of in-situ countermeasures to mitigate such events. Data were collected by continuous monitoring of 12 exit ramps with potential WWD problems. Eight of the ramps received WWD countermeasures for mitigation purposes, and data were collected for these before and after mitigation. The results indicate that a major factor in a WWD event is driver confusion related to exit ramp geometry and signage, and that in-situ mitigation can be an effective countermeasure. This also suggests that increasing conspicuity of wrong-way entrance can potentially reduce the number of WWD crashes and fatalities.

This study develops a new methodology based on computational metrology techniques to measure rail surface roughness from vertical axle-box accelerations. A VIA rail passenger rail car, operating in eastern Ontario, Canada, was instrumented with accelerometers on the car body, trucks, and axle-boxes. The rail surface was measured by a heavy track geometry inspection car. A Gaussian filter was applied to the measured rail surface data and the rail surface roughness was quantified by the root mean square (RMS), as recommended by ISO 11562 for evaluating engineering surfaces. Values of rail surface RMS roughness calculated from axle-box accelerometer data are verified with the measured surface RMS roughness. The overlap ratio and length of the moving window over which the rail surface RMS roughness is calculated are studied with respect to roughness wavelengths, statistical considerations, and maintenance planning purposes. The effect of rail car operating speed as well as the difference between axle-box accelerations measured at two axles of the instrumented rail car on the accuracy of estimating the rail surface RMS roughness are assessed. Filtering techniques and application limitations for calculating roughness are also discussed. The results of this study suggest axle-box acceleration data provide a useful assessment of rail surface roughness for the typical wavelengths between 3 and 25 m and a complementary technique to light and heavy track geometry inspections.

Travel time to a destination is a key piece of information that motorists desire. Traditionally, transportation agencies collect traffic state information with costly measurement sensors. In the last decade, private sector data, as an emerging data source, has been gradually adopted by many public agencies for performance measurement and travel time provision, mostly on freeways. Agencies have also used data from infrastructure sensors and floating cars to validate the quality of the private sector data on both freeways and arterials under different conditions. Although freeway data quality has been validated by many studies, nonfreeway travel time data have been considered unreliable, particularly when the segment is congested and high-density signals exist. In recent years, data vendors started to provide high-resolution travel time data, which are provided on much shorter segments than conventional traffic message channels (TMC). This study investigates the data quality of this new data source for providing travel times on nonfreeways. It evaluates the selected private sector data on nonfreeways by using floating car and Wi-Fi travel time data as the ground truth. The results show that the new private sector data on nonfreeways are generally acceptable for segments with low and moderate congestion. With higher congestion, the data quality is site-dependent. A neural network model is proposed to implicitly identify the error of a site and provide enhanced travel times. It is recommended that these new data should be validated for the deployment sites due to the site-dependent errors, and advanced data modeling techniques can be used to compensate for such errors.

Ramps are a critical component of an interchange, which seek maximum driver attention to satisfy the deceleration and acceleration requirements. To improve safety and articulate significant safety schemes, it is important to analyze driver behavior on ramp interchanges at a microscopic level. Existing studies on highway geometry are limited to horizontal curves, and only a few studies were found to deal with ramp interchanges. In this study, a model based on support vector regression (SVR) was developed to estimate lateral acceleration ($L_a$) experienced by the drivers on diagonal, loop, and semidirect ramps of service interchanges. To establish these models, the continuous lateral acceleration profiles for 83 drivers were collected using an instrumented vehicle. The developed SVR models exhibited higher accuracy, measured by the values of coefficient of determination and the root-mean square error. Further, a sensitivity analysis was performed to measure the relative importance of input features. The results revealed that ramp curvature and ramp length are the two most significant variables that impact lateral acceleration on diagonal and semidirect ramps. However, for loop ramp connectors, operating speed ($V_{85}$) (85th percentile speed) displayed the highest association with lateral acceleration. The models developed in this study can be used as a tool to estimate the lateral acceleration experienced by the drivers during the design process, thus enhancing the further understanding of driver safety on ramp interchanges.

This paper studied driver behavior on freeway entrance ramp terminals using trip data from the SHRP-2 Naturalistic Driving Study (NDS) database from two different US states. The study included a qualitative assessment of driver speed behavior as drivers navigate the ramp and speed change lane (SCL). A trend of continuous vehicle acceleration was evident from the beginning of the ramp controlling curve that continued after merging onto the freeway right lane (FRL). It also was found that a portion of drivers tended to merge onto the FRL on the taper after the SCL had ended; this behavior was dominant on the taper-type SCL. It was shown that the speed and acceleration behavior of drivers differed on the different sites, indicating that it depends on the complete set of geometric characteristics at the site. Linear-mixed models were developed for the speed and acceleration measures and for lead, lag, total accepted merging gaps to account for the repeated measures caused by repeated trips by the same drivers.

Control strategies with pre and main signal have been increasingly researched and practiced. With the transformative effect of connected and automated vehicles (CAVs), more possibilities have been created, and the long-time mix of CAVs with manual vehicles (MVs) has inspired radical change to the unconventional intersection with double stop lines. Specifically, the study is oriented with CAV controllability to facilitate vehicle passage throughout the intersection, by allowing CAVs to make use of yellow light at the presignal under in-vehicle control. To avoid spill-over and better clear vehicles on the approach, the distance from pre to main signal is constrained so that vehicles can queue upstream and pass the main stop line in time, under the objective of maximizing intersection reserve capacity. The proposed model is solved with iteration, and is found to reduce signal cycle, avoid spill-over, and enhance reserve capacity by larger margin as CAV penetration rate increases. This study provides insights to the design and control of intersections with double stop lines under mixed traffic flow, laying foundation to traffic conversion from MV to CAV.

Lane changing is a fundamental driving task and is closely related to traffic operation. The safety performance of vehicle driving and traffic flow is supposed to be substantially improved if lane-changing behavior can be precisely predicted. To this end, a model based on the Gated Recurrent Unit (GRU) is proposed in this study for freeway on-ramp lane-changing behavior forecasting. One specific feature of the model is that it enables the filtering out of the lateral oscillation behavior and helps enhance forecast accuracy. The experiment results show that the model achieves an accuracy of 96.85% for lane-changing behavior forecasting, and outperforms the GRU model without lateral acceleration input and the LSTM model by 5.12% and 4.51%, respectively.

Among mass transportation facilities, high-speed rail systems not only connect major cities but they also drastically shorten the transportation time. The high-speed train maintenance works are an important part of the safety operation system. Also, the reasonable planning and maintenance works execution can ensure the safe running of trains, improve the quality of operation, and reduce company costs. The regular maintenance is a critical element for providing an excellent transportation service to the public. The present study uses data envelopment analysis (DEA) and the Malmquist productivity index (MPI) to analyze the multiyear trends in productivity for the yearly and midterm maintenance efficiencies of 12 high-speed trains in Taiwan. These features are evaluated through systems thinking, and suggestions are offered towards improving those factors responsible for the inefficiency. The results show that eight of the trains studied had attained satisfactory maintenance efficiency, whereas four could be improved, yielding a productivity of 66.7%. In addition, by following the trends in the MPI values, it was found that the technical efficiency (TE) and the total factor productivity (TFP) are the main factors responsible for stalling maintenance performance, contributing a decline of 37.1% and 37.7%, respectively. Our study has identified three factors leading to maintenance inefficiency. These three factors are: (1) a change in the scope of the contract, (2) a change of work methods, and (3) a lack of effective communication between the management and the customer. It is hoped that this research can provide useful information that will allow decision-makers to avoid erroneous decisions caused by poor intuitive judgment or inadequate considerations, and to formulate correct policies, to improve maintenance productivity, and to ensure that trains can continue to run safely.

The severe geological and environmental conditions along the Sichuan–Tibet railway easily lead to unfavorable events that place construction safety at risk. To comprehensively identify the risks associated with this project, this paper proposes a novel model-based systems engineering (MBSE) method that employs Department of Defense architecture framework (DoDAF) artifacts. In this method, system modeling is applied to risk management, and model analysis is performed to identify the risks in a project and their related factors. This method is closely related to the engineering process and can be applied to all stages of the entire project life cycle. The Jinjiguan No. 2 tunnel project was taken as a case study to demonstrate the application of this method. The task model, structure view, and activity view of this project were established using Systems Modeling Language (SysML). By analyzing the system behavior deviations that emerged during the engineering activities, the potential major risk events in tunnel construction, including collapse, water gushing, and explosions, were identified. The status and behavior of system components were examined from various perspectives to identify the risk factors that may emerge during tunnel construction. The results showed that the proposed method can efficiently develop system models, capture the structure and behavior of a railway engineering system, and accurately identify the risks faced by the system.

Distance-based fares have been applied widely in practice using automatic fare collection technology. This paper proposes an optimization model for the distance-based transit fare structure. A bilevel model was formulated to obtain the optimal fare functions with the aim of minimizing the Gini coefficient of the entire transit system. Passengers’ travel behavior is modeled by a stochastic transit assignment problem in the lower level. Considering the inherent complexity of the bilevel model, a heuristic algorithm, namely the artificial bee colony algorithm, is applied, which is incorporated with a method of successive averages to solve the transit assignment subproblem. The results show that the Euclidean distance–based fare has better performance in terms of social equity, but it is less equitable for passengers with middistance trips.

Smart cities—being equipped with connected infrastructure—receive significant real-time traffic data. In this paper, a data-driven connected corridor traffic simulation model, i.e., a digital twin, is developed that leverages real-time data streams to model the current traffic state and provide dynamic feedback on traffic and environmental performance measures, e.g., travel time, speed, energy consumption, and vehicular emissions. The developed digital twin model architecture uses real-time 6-min volume aggregate data and 0.1 Hz to 10 Hz signal indication data as input to simulate the connected corridor using a traffic modeling software microscopic simulation. Dynamic data retrieval and transfer from the simulation model are enabled using the software’s COM feature and a Flask web server. The robustness and feasibility of the digital twin architecture and the generated performance measure reasonableness are demonstrated on a smart corridor test bed. Such a model can be used to monitor and provide insights on the impacts of intelligent transportation system technologies on connected corridor traffic and environmental performance.

Mass streetcars operating record data used in a proper way can reveal the operational characteristics of streetcars to the full extent. This paper aims to design the traffic-signal progression for the modern streetcar by mining and analyzing these data. Each unit in the study consists of two adjacent signalized intersections. The $k$-means clustering algorithm is used to cluster the streetcar travel time data between two intersections and to determine whether the streetcar has stopped waiting for a red light at an intersection or not. An offset calculation model for streetcars is proposed to lower the impact of the conventional transit signal priority (TSP) for streetcars on motor vehicles. A real-world case study of Line One of the Nanjing, China, Qilin modern streetcar shows that the traffic-signal progression design proposed in this paper can reduce the conventional TSP’s negative impacts on motor vehicles by about 20%.

Airports provide essential infrastructure to connect travelers and products to destinations across regions, nations, and the world. However, these connections were the focus of restrictions in the midst of a global pandemic because the same potential to provide access to people and products as a hub of long-distance travel can also serve as an opportunity to spread a virus. The objective of this paper is to explore the temporal effects of the COVID-19 pandemic on flight operations at airports in the US, in addition to an examination of the differences among various classifications of airports. For commercial service and general aviation airports, the airports with the quickest and most substantial recoveries were in areas with populations of under 100,000 people [within 48.2 km (30 mi) of the airport]. Airports with higher and lower populations within 48.2 km experienced sharper declines in flights and slower recoveries. Local, state, and national airport infrastructure investments will need to consider these effects and changes in demand spurred by COVID-19 while monitoring the long-term sustainability of the changes.