Journal of Transportation Engineering, Part A: Systems

Abstract

Crashes involving heavy trucks on rural mountainous freeways are known to result in severe injuries and fatalities, particularly under challenging driving conditions. This study aims to develop a robust model to accurately predict fatal and injury crashes involving heavy trucks on rural mountainous freeways. The crash database of Interstate-80 in Wyoming was used to extract a wide range of variables related to environmental, roadway, crash, occupant, and vehicle characteristics. This study employed a state-of-the-art deep neural network architecture named ResNet-50 using transfer learning to develop crash severity prediction models. The numeric crash data were converted to images utilizing DeepInsight to facilitate the application of the proposed deep learning model. Due to the imbalanced nature of the crash severity data, this study employed random undersampling (RUS) and synthetic minority oversampling technique (SMOTE) data balancing techniques and investigated several data sampling ratios. A ratio of $1:2:2$ (Fatal: Injury: PDO) combined with both RUS and SMOTE produced the best performance with recall values of 99.7%, 79.7%, and 79.3% for fatal, injury, and PDO crashes, respectively. This study also employed Boruta and extreme gradient boosting (XGBoost) to examine the significance of variables on crash severity. The findings revealed that the deployment of airbags, use of seatbelts, driver distraction, and driver conditions such as inattentiveness and fatigue, vehicle type, vertical grades, weather, and road surface conditions were the most critical variables contributing to the severity of crashes involving heavy trucks. Furthermore, this study developed several reduced truck-involved crash severity prediction models without significantly compromising the prediction performance. The proposed deep neural network model can provide accurate and timely prediction of fatal and injury crashes involving heavy trucks, which is beneficial for ensuring efficient collision management, avoiding secondary pile-up crashes, and facilitating prompt medical assistance.

Practical Applications

The proposed method offers a potential solution to predict crash severity, particularly for fatal and injury crashes, in challenging mountainous regions. This is crucial for effective collision management, prevention of secondary crashes, and timely medical assistance. Key contributing factors to crash severity, such as steep grades, sharp turns, driver distraction, fatigue, weather conditions, and road surface conditions, were identified. This information can guide safety practitioners, emergency services, traffic management centers, and manufacturers in improving their countermeasures and safety applications to provide accurate warnings to road users. For example, connected and automated vehicles can use this information to warn drivers about difficult roads with lots of trucks or upcoming bad weather on mountainous roads. Smart cameras inside the car can also be used to detect if the driver is distracted, not paying attention, or tired and can give them physical or sound alerts to help prevent serious accidents. The proposed method can be applied in various safety-related scenarios using numerical or vision-based data to gain a deeper understanding of driver behavior and its association with safety events, ultimately improving road safety for everyone.

With the advent of the Internet of Things, bike-sharing systems have seen widespread adoption globally, whereas they often grapple with an uneven spatiotemporal distribution of vehicles. This issue is particularly acute in the wake of electronic fences, with some areas often faced with the predicament of inadequate supply. To tackle this challenge, accurate prediction of borrowing and returning demands at different parking spots and varying times is necessary. In this study, we used a comprehensive data set from Yancheng, Jiangsu, China, covering shared bicycle usage across 394 parking spots. These data enabled us to delve deep into urban travel patterns and discern the various factors influencing these behaviors. To enhance the prediction accuracy, we propose the time-series weighted regression (TSWR) model, a long-term multistep forecasting method, which adeptly addresses issues associated with sparse statistical data and long-term prediction inaccuracies, outperforming other machine learning models in our experiments. Further recognizing the considerable impact of geographical location and weather conditions on shared bicycle demand, we incorporated the rule-based adjustment optimization (RAO) method into our approach, which refines nonlinear components by accounting for various factors. The implementation of RAO resulted in a 10.34% increase in accuracy compared to TSWR alone and an improvement of over 35% in comparison to other approaches. Overall, this study illuminates the underlying influences on urban travel patterns and offers valuable suggestions for bike dispatching to those enterprises, contributing significantly to the research in this field.

The traffic checkpoint data collected by video detection and license plate recognition technologies contain abundant traffic information such as vehicle travel time and path. However, due to various factors, there are often many outliers in the travel-time data, which may lead to inaccurate research results. How to extract effective travel-time data from the traffic checkpoint data is one of the key questions for urban traffic evolution. Based on the large-scale checkpoint data from a city in Guangdong Province, China, this study proposes a statistics-based filtering algorithm for travel-time outliers and applies it to urban vehicle travel path and origin-destination (OD) identification. The sampling rate and reading rate of checkpoint data are verified. The sampling rate and reading rate are 99.46% and 93.7%, respectively, and the reliability of the data is high. A statistics-based filtering algorithm for travel-time outliers is proposed and compared with three existing clustering algorithms. The results showed that for a load section of 376.54 m, the maximum travel time is reached during the period 18:00–18:15. The average travel time is 5.55 min, and the maximum travel time is 8.08 min, which is consistent with the actual operation of the road section. The travel time obtained by the three clustering methods does not reflect the actual operation of the road section well. This means that the effective travel time extracted by this method is more practical and can reflect the fluctuation of travel time on urban roads. Then, the sensitivity of the proposed method to the time window and road length is tested. The results showed that the time window setting will affect the outlier-filtering effect, and the road length has less effect on the proposed method. Finally, the proposed algorithm is applied to single travel identification of urban vehicles to extract single travel paths and OD information of vehicles, which can provide accurate data for urban traffic research.

Autonomous vehicles (AVs) are revolutionizing the transportation system and necessitating a reevaluation of infrastructure capacity. Apart from their shorter headway, AVs exhibit distinct behavioral characteristics compared to human-driven vehicles (HDVs) in two key aspects: (1) significantly reduced noise levels, and (2) diminished bidirectional lateral influence. Unlike HDVs, AVs maintain closer alignment at the lane middle line and are not subject to the influence of the lateral wandering of neighboring vehicles. This is attributed to their remarkable ability to accurately predict HDV behavior. As the market penetration rate (MPR) of AVs increases, the interaction patterns of HDV flow between adjacent lanes, which traditionally lead to lateral friction and capacity loss, undergo gradual transformation. The current evaluation of capacity and adjustment factors (CAFs) based on theoretical models or simulation packages may overlook the stochastic lateral friction exhibited by mixed AV flow, resulting in potentially biased results. In this study, we develop a two-dimensional stochastic model that captures the random lateral dynamics and investigate the capacity and adjustment factors for multilane mixed AV flow. The findings demonstrate that the model effectively reproduces the lateral friction phenomenon, and as the number of lanes increases, the capacity experiences a decline of up to 20% due to lateral interactions. Moreover, the study clarifies the impact of dedicated lane assignment on the total capacity, as it hinders the propagation of lateral friction.

This paper proposes a recommender in multimodal transportation-as-a-service (MMTaaS) system that offers personalized travel planning in multimodal transportation network. The framework focuses on three key aspects: (1) multimodal path-planning based on individual travel demands within a large-scale road network; (2) determination of traveler-specific travel itineraries, taking into account various information sources such as the topology of the road network and the supply of each transportation mode; and (3) personalization of travel plan recommendations using stacked autoencoder based on individual traveler attributes. The proposed recommender adopts a data and model-driven approach, leveraging data from various sources to inform decision-making and model the problem. The effectiveness and feasibility of the MMTaaS framework are demonstrated through a case study in Jiaxing City, Zhejiang Province, China, which highlights the framework’s ability to handle single and multimodal traffic trips and customized individual trips. The results of this study demonstrate the effectiveness and feasibility of the proposed MMTaaS recommender and provide valuable insights for the development of future transportation-as-a-service systems.

Estimating pedestrian exposure for all the intersections in a jurisdiction is crucial for developing strategies with a focus on pedestrians. Some engineering applications require the annual average daily pedestrian traffic (AADPT) to be disaggregated per crosswalk. When continuous counts are available at the intersection, this indicator can be calculated directly. However, when only short-term counts (STCs) or no information on pedestrian volume is available, the AADPT per crosswalk cannot be calculated and must be estimated using other means. This work (1) evaluated the degree of confidence for estimating the pedestrian volume in each crosswalk based on point estimates of percentage shares per crosswalk obtained from STCs; and (2) developed models to estimate the percentage share of pedestrian volume per crosswalk as a function of attributes of the intersection that commonly are available for jurisdictions, referred to as the land-use (LU) model. The two methods were evaluated using continuous count data from three different jurisdictions, and a naive estimate assuming equal shares per crosswalk was used as a benchmark. The performance of each method was measured as the fraction of the intersection AADPT that was allocated wrongly to each crosswalk. The use of the LU model generated an average wrong allocation of 0.301, a statistically significant improvement of 11.4% compared with the naive estimate. The use of a STC from a single day produced an average wrong allocation of 0.153, an improvement of 54.9% from the naive estimate. Increasing the number of days of STCs to two or three resulted in average performance indicators of 0.117 and 0.106, respectively. The benefits of using STCs for more than three days are minimal. The STC method was developed using STCs from the same 1-year period in which the observed share was averaged. In practice, STCs are likely to be between 1 and 5 years old. Analysis using STCs from previous years showed that estimation error in practice may be as much as twice as large as the aforementioned errors.

Work zone safety, as an issue requiring complex solutions and tech-driven investment, has been a major concern for transportation agencies and policymakers. To help decision-makers prioritize investments, there is a need for tools enabling them to assess policies designed to reduce the cost and count of work zone crashes at the same time. Traditional statistical methods, however, are unable to consider two or more target variables when they are intrinsically different (e.g., when a mixture of continuous and dichotomous/discrete variables is of interest). As a result, the dependencies among the outputs of interest may be neglected, which potentially causes biased decision-making. In this study, we used two machine learning models, i.e., group method of data handling (GMDH) and multilayer perceptron (MLP), to simultaneously predict count and cost of work zone crashes in the state of Tennessee in the United States. We further used these tools to assess four policies that could help improve work zone safety. We compared predictability of these models in individual prediction of cost and count with the three statistical models of Poisson regression, negative binomial regression, and multivariate regression. Also, we compared the importance of different variables to evaluate the role that each input variable plays in prediction and recognized the most effective factors. Our results indicated that reduction of annual average daily traffic (AADT) and speed limit are the most effective policies to simultaneously lower work zone crash count and costs, ensued by reducing the percentage of trucks and prohibiting commercial vehicles on segments. We observed that MLP and GMDH perform almost equally in the simultaneous prediction, while these models provide significantly greater degrees of accuracy in crash count and cost prediction, compared with statistical models. The results depicted that the major predictors of cost and count prediction of work zone crashes are speed limit, illumination, involving commercial vehicles, AADT, type of road segment, and percentage of trucks.

A quantitative evaluation of the traffic flow order is necessary to improve the operational level at intersections. To date, how to assess the subjective perceptions of the traffic flow order at intersections is unclear. This study develops a visual analog scale (VAS) and an artificial intelligence algorithm to explore the subjective and objective evaluation methods of the traffic flow order at signalized intersections, respectively. First, a subjective measurement method is developed based on the VAS by evaluating 100 video clips from 24 intersections in Shanghai. Then, an objective estimation model for the vehicular traffic flow order evaluation is constructed based on a multilayer perceptron neural network (MLP). A model parameter-hyperparameter joint optimization method is proposed. The results show that the developed method for the traffic flow order at intersections has a coefficient of determination (R2) of 0.83 and an average absolute error of 5.78 compared with the subjective evaluation.

Optimization of train operation is crucial in rail transport networks where large amounts of energy are used. The energy-efficient operation of trains is seen to be a good approach to saving operating costs in metro systems. There are two main techniques for efficient train operation. These are decreasing traction energy and increasing the usage of regeneration energy. In order to make maximum use of the regenerative energy of trains, an integrated optimization method that calculates the optimum train speed trajectory and timetable is proposed in this study. The suggested model aimed to increase the overlapping time of braking and accelerating train groups by improving the timetable. In addition, it is intended that two trains in the same electric zone do not accelerate or brake simultaneously so that the catenary voltage does not rise and decrease too much. The Istanbul M3 subway system with its real operating data is modeled and simulated. The genetic algorithm and simulated annealing algorithm were employed in the optimization process. The outputs of the study show that the consumption of traction energy can be substantially decreased by employing the best train speed profile and timetable.

Abstract

The unbalanced distribution of parking demand is a primary source of parking problems. The autonomous driving system and parking assist system of fully autonomous vehicles provide a possibility to alleviate the uneven distribution of parking demand in the city. Exploring the parking behavior of autonomous vehicle (AV) users is necessary for assessing parking pricing policies. In turn, the distribution of parking demand and people’s preference for parking lots may be impacted by the change in the AV parking lot’s parking charge. Thus, this paper aims to provide an effective optimization method for AV parking lot price strategy, based on the analysis of the parking behavior of AV users. A stated choice experiment is designed to understand AV users’ parking behavior. To estimate the impact of attributes and social-demographic factors, the multinomial logit model is adopted. From the results, we find a farther but cheaper parking lot is more attractive than a closer but more expensive parking lot to AV users. Based on the estimated results, a bilayer optimization model based on parking behavior is built to optimize AV parking lot prices. The Lagrangian relaxation method is employed to solve the model. By empirical testing, the AV parking price optimization model based on parking behavior is proven to be feasible and effective. According to the case study’s findings, the parking lot pricing optimization model and algorithm proposed in this paper can not only meet the benefits of parking lots and the utility of autonomous vehicle users, but also alleviate the demand for parking concentrated in the city center, which means the parking difficulties caused by the distribution of parking demand can be alleviated through the optimization of AV parking lots’ parking price.

Practical Applications

Parking difficulties in the city center are a major problem in urban traffic. Higher demand for parking in the city center may cause a series of problems, such as traffic congestion, emissions, and traffic safety. The autonomous driving system and parking assist system of fully autonomous vehicles provide a possibility to alleviate these problems. by exploring of AV users’ choice preference on parking lots, we find a farther but cheaper parking lot is more attractive than a closer but more expensive parking lot. This indicates that people may change their parking lot choice behavior. However, it is not enough to balance parking demand by relying on autonomous vehicles and user choice preferences alone. Without the guidance of reasonable parking lot prices, parking lots far away from the city center cannot attract users who originally choose parking lots in the city center. As a result, by the research on the parking lot selection preferences of autonomous vehicle users, the pricing optimization of autonomous vehicle parking lots can effectively adjust the distribution of urban parking demand, thus alleviating urban parking difficulties.

This study employs the mixed logit model to examine the statistically significant risk factors in influencing driver injury severity in four crash configurations in two-vehicle rear-end crashes on urban state roads based on seven-year data from the Washington State Department of Transportation. The data set only includes collisions with passenger cars and pickup trucks involved, as these two vehicle types are the most common in these crashes, and these two types typically have different masses and kinetic features. Four crash configurations are examined concerning the type of vehicles and their relative positions in a crash. Four models for these configurations and one model for the overall data set are estimated. The result shows discrepancies and similarities in risk factors for injury and fatality in different crash configurations, and each model has its unique parameters with heterogeneity of different distributions. Contributing factors to driver fatality include (1) alcohol-impaired drivers and older drivers when a passenger car strikes another passenger car, (2) curved and not level roads, and the vehicle in front when a passenger car is stricken by a pickup truck, (3) male drivers and being in winter season when a pickup truck is stricken by a passenger car, and (4) seatbelts not being in use for all configurations. Research findings in this work help understand the risk factors for driver injury severity in each configuration and develop injury and fatality prevention strategies for two-vehicle rear-end crashes involving passenger cars and pickup trucks.