Recent years have seen a significant amount of transportation data collected from multiple sources including road sensors, probe, GPS, CCTV and incident reports. Similar to many other industries, transportation has entered the generation of big data. With a rich volume of traffic data, it is challenging to build reliable prediction models based on traditional shallow machine learning methods. Deep learning is a new state-of-the-art machine learning approach which has been of great interest in both academic research and industrial applications. This study reviews recent studies of deep learning for popular topics in processing traffic data including transportation network representation, traffic flow forecasting, traffic signal control, automatic vehicle detection, traffic incident processing, travel demand prediction, autonomous driving and driver behaviours. In general, the use of deep learning systems in transportation is still limited and there are potential limitations for utilising this advanced approach to improve prediction models.

Under efficiency improvement of road networks by utilizing advanced traffic signal control methods, intelligent transportation systems intend to characterize a smart city. Recently, due to significant progress in artificial intelligence, machine learning-based framework of adaptive traffic signal control has been highly concentrated. In particular, deep Q-learning neural network is a model-free technique and can be applied to optimal action selection problems. However, setting variable green time is a key mechanism to reflect traffic fluctuations such that time steps need not be fixed intervals in reinforcement learning framework. In this study, the authors proposed a dynamic discount factor embedded in the iterative Bellman equation to prevent from a biased estimation of action-value function due to the effects of inconstant time step interval. Moreover, action is added to the input layer of the neural network in the training process, and the output layer is the estimated action-value for the denoted action. Then, the trained neural network can be used to generate action that leads to an optimal estimated value within a finite set as the agents' policy. The preliminary results show that the trained agent outperforms a fixed timing plan in all testing cases with reducing system total delay by 20%..

In this study, the authors employ machine learning to develop a new solution method for solving a tri-level network protection problem. In the upper-level, the planner aims to minimise the impact of the interdictor's attempt to disrupt a road network through protection activities. At the middle-level, however, the interdictor seeks to maximise the network's cost function, that is total travel time while the user equilibrium assignment models the road users behaviour at the lower-level. The proposed solution algorithm combines implicit enumeration with machine learning for faster performance. In so doing, four machine learning methods are evaluated among which the artificial neural network model shows the best performance and thereby to be exploited. Principal component analysis is also employed as part of the data pre-processing to perform dimensionality reduction. The proposed solution algorithm exhibits a reasonable level of tractability when employed to solve large problems in which a real-world network is under investigation. Although it cannot guarantee global optimality, it is argued that this is an essential compromise for the application of the network optimisation problems on extensive real-world networks and the large solution space that they generate.

A critical issue in origin–destination (O–D) demand estimation is under-determination: the number of O–D pairs to be estimated is often much greater than the number of monitored links. In real world, some centroids tend to be more popular than others, and only few trips are made for intro-zonal travel. Consequently, a large portion of trips will be made for a small portion of O–D pairs, meaning many O–D pairs have only a few or even zero trips. Mathematically, this implies that the O–D matrix is sparse. Also, the correlation between link flows is often neglected in the O–D estimation problem, which can be obtained from day-to-day loop detector count data. Thus, sparsity regularisation is combined with link flow correlation to provide additional inputs for the O–D estimation process to mitigate the issue of under-determination and thereby improve estimation quality. In addition, a novel strategic user equilibrium model is implemented to provide route choice of users for the O–D estimation problem, which explicitly accounts for demand and link flow volatility. The model is formulated as a convex generalised least squares problem with regularisation, the usefulness of sparsity assumption, and link flow correlation is presented in the numerical analysis.

It is essential to understand how transit passengers arrive at stops, as it enables transit operators and researchers to anticipate the number of waiting passengers at stops and their waiting time. However, the literature focuses more on predicting the total passenger demand, rather than simulating individual passenger arrivals to transit stops. When an arrival process is required especially in public transport planning and operational control, existing studies often assume a deterministic uniform arrival or a homogeneous Poisson process to model this passenger arrival process. This study generalises the homogeneous Poisson process (HPP) to a more general non-HPP (NHPP) in which the arrival rate varies as a function of time. The proposed collective NHPP (cNHPP) simulates the passenger arrival using less time regions than the HPP, takes less time to compute, while providing more accurate simulations of passenger arrivals to transit stops. The authors first propose a new time-varying intensity function of the transit passenger arrival process and then a maximum likelihood estimation method to estimate the process. A comparison study shows that the proposed cNHPP is capable of capturing the continuous and stochastic fluctuations of passenger arrivals over time.

As connectivity and automation make their way in to transportation systems, they are expected to have a forceful impact, drastically changing road transportation. The introduction of autonomous vehicles (AVs) and connected autonomous vehicles (CAVs) is expected to advance safety and comfort. But, they can also affect characteristics of road networks, such as capacities, delays and efficiency. To foresee important challenges, reinforce potential benefits and reduce potential disadvantages of this new disruptive technology, its impacts should be well studied and understood before their anticipated introduction. In this paper, a microscopic simulation framework to estimate these impacts is developed. Simulation experiments are conducted for various traffic mixtures of manually driven vehicles, AVs and CAVs, different desired time headways settings and traffic demand levels, to evaluate the sensitivity of the network performance to these factors. The ring road of Antwerp is used for the case study. Thus, the results and conclusions refer to a large real-world network. The consequences of the introduction of AVS and CAVs on traffic flow and pollutant emissions are evaluated. The results show that depending on the demand, AVs introduction can have negative effects on traffic flow, while CAVs may benefit the network performance, depending on their market penetration.

Vehicle automation opens opportunities toward the improvement of people, planet, profit value in applications on distribution centres (DCs). Despite vision and drive for innovation in logistics, the lack of knowledge prevents the application of autonomous applications. Intelligent Truck Applications in Logistics (INTRALOG) contributes to this deficit and generates valuable insights for a future application in public environments. Current operational automated guided truck applications are bound to fixed infrastructure and do neither operate in the public domain nor offer opportunities to do so. Nowadays, in-vehicle intelligent systems are focused on driver support, opening opportunities such as truck platooning. INTRALOG cross-borders on DCs in relatively low-complex traffic environment, bridging the gap between autonomous driving in the public domain. The multi-agent system developed within INTRALOG aligns logistical movements on DCs, controlling single or double articulated container trailers (longer and heavier vehicle) between the public parking area and cross-docks. This study elaborates on the experiments on automated manoeuvring on a DC with single (SAVs) and double articulated vehicles (DAVs). The experiments comply with business requirements, e.g. manoeuvrability, time to dock and positioning accuracy. The research focuses on the effects of these aspects and control strategies of SAVs and DAVs.

This study presents an innovative hierarchical Bayesian model for mapping of county level truck traffic in New Jersey. First, the model is estimated using truck counts. Then, using overweight truck counts from weigh-in-motion data as the response variable, the model is re-estimated. The goal in using the overweight trucks in the spatial model is to demonstrate the importance of representing their spatial variation due to their impact on the life of the roadway network elements. Finally, truck count maps are developed based on modelling results to visualise the effects of spatial covariates. The results of the study indicate that the most influential covariate for the truck traffic is the length of interstate roadways, followed by employment and population. The developed truck count maps can help transportation professionals on identifying and ranking the locations at an aggregate level, which requires closer attention.

In order to minimise the total cost of the logistics part of the network design, it is necessary to use the hub-and-spoke network structure to achieve economies of scale in transportation. Most of the models used to design hub networks have considered a single level of hub type, but they do not reflect the hub scale by freight volume in the network. The aim of this study was to design a hub-and-spoke network with hierarchical hubs and develop a model for the evaluation of logistics systems. The developed model determined the number of each level of hubs to be opened, identified the locations of these hubs, and allocated the demand nodes to the hubs. In this study, the development of algorithms for solving problems was also investigated. This model provides new insights and approaches into these research areas. The proposed algorithm contributes to the more generalised field of combinatorial optimisation problems, particularly for the problems associated with the design of a hierarchical hub network.

Transport network infrastructure including the use of Intelligent Transport Systems (ITS) is fundamental to the mobility of people and shipments. This study aims at understanding how different ITS measures may impact multiple traffic engineering goals with respect to a set of Key Performance Indicators (KPIs). The ordering of KPIs is determined by the preferences of the decision maker. A Multi-Criteria Decision Analysis framework is proposed for impact assessments of ITS measures. Data about KPIs is derived with the help of traffic data gathered from the Gothenburg Region. Comparing the contributions of ITS measures to different goals suggest that the management of flow and speed in a road network is crucial for improving capacity utilisation. The goal assessments are then projected onto different use-cases in terms of socio-economic impacts. The result indicates that corridor section traffic management, transport management (focusing on transit traffic) and urban gate-way will generate socio-economic effects, respectively, in a decreasing order. Urban gateway is particularly interesting because it is the least intrusive and less costly use-case.

The automotive companies nowadays are working on the development and deployment of the future intelligent transportation systems and connected-autonomous driving addressing any security and privacy concerns of users. In the field of connectivity, vehicular networks are constructed to manage communications among nearby vehicles and between vehicles and road-side infrastructure units and trusted agencies or certificate authorities. The above cooperative systems applications bring the promise of improved road safety and optimised road traffic. For the successful deployment of vehicular communications (VCs), it is essential to make sure that ‘life-critical safety’ information cannot be modified by external or internal within the network attackers. On this basis, lack of such security and privacy in vehicular networks is one of the key hindrances to the widespread implementations of connected-autonomous-driving applications. Moreover, specific operational parameters (highly moving vehicles, frequently and fast changed connectivity) make the problem very novel and challenging. For the present study, a protocol for data privacy and authenticated VC interactions, based on the Diffie–Hellman algorithm, is applied. The above analysis aims to improve the security and enhance the efficiency of VCs toward the deployment of Internet of vehicles technology in the transport area.

Research and development in the field of autonomous vehicles has increased along with related work on automated driving (AD) software. Thorough testing of AD software using simulations must be conducted in advance of testing AD cars on the road. Parameters of the many objects around an AD car, such as other cars, traffic lanes and pedestrians are required as inputs of the simulation. Therefore, the number of parameter combinations becomes extremely large. A combination of parameters is called a test case; hence, the challenge is to search collision test cases from the extremely large number of combinations. A rule-based method is the main focus because an explicit method of searching test cases is required in certain industries in the real world. In this study, a method of rule-based searching for collision test cases of autonomous vehicles simulations is proposed. Simulation models that have rules between an AD car and other cars are defined. Algorithms were also developed to search collision test cases that generate test cases incrementally. Experiments on AD simulations involving the simulation models of a three-lane highway and a signalised intersection were conducted. The results indicate the efficiency of the method.

Worldwide intelligent transportation systems (ITSs) are at the edge from research and development to deployment and commercial operation. Japan has already introduced ITS technology in the market and the USA and the European Union have elaborated planes for deployment. ITS in this context means communication and services based on communication technology between vehicles and other traffic participants on one hand and traffic and road infrastructure, vehicle manufacturer, and other mobility service provider on the other hand. For a successful introduction, a reliable and secure exchange of mobility-related information is a key factor. To provide such an exchange for an overall architecture for ITS and for all participants and users is necessary. With the introduction of the market, the assessment of an ITS architecture without an existing reference system arises. Since most assessment methods currently existing, measure how ‘good’ an architecture is in comparison with another architecture. In this study, the authors describe ways and approaches how existing methods can be extended and combined to provide means for the assessment of ITS architectures in the pre-deployment phase. As a result, deploying parties should be enabled to assess an architecture before introducing it to the public.

When developing cyber-physical systems such as automated vehicles, safety and cybersecurity analyses are often conducted separately. However, unlike in the IT world, safety hazards and cybersecurity threats converge in cyber-physical systems; a malicious party can exploit cyber-threats to create extremely hazardous situations, whether in autonomous vehicles or nuclear plants. The authors propose a framework for integrated system-level analyses for functional safety and cyber security. They present a generic model named Threat Identification and Refinement for Cyber-Physical Systems (TIRCPS) extending Microsoft's six classes of threat modelling including Spoofing, Tampering, Repudiation, Information Disclosure, Denial-of-Service and Elevation Privilege. TIRCPS introduces three benefits of developing complex systems: first, it allows the refinement of abstract threats into specific ones as physical design information becomes available. Second, the approach provides support for constructing attack trees with traceability from high-level goals and hazardous events (HEs) to threats. Third, TIRCPS formalises the definition of threats such that intelligent tools can be built to automatically detect most of a system's vulnerable components requiring protection. They present a case study on an automated-driving system to illustrate the proposed approach. The analysis results of a hierarchical attack tree with cyber-threats traceable to high-level HEs are used to design mitigation solutions.

This study summarises the implementation of software system architecture and relevant modules to enable cooperative adaptive cruise control (CACC) functionalities as an extension of adaptive cruise control (ACC), thereby leveraging the lessons learned from prototype ACC vehicle testing as well as ideas from prior research. These activities were conducted in the United States under a cooperative agreement between the Crash Avoidance Metrics Partners, LLC and the Federal Highway Administration. A key outcome of this project was to understand the implementation of advanced capabilities for the CACC algorithm in a very structured manner. With the introduction of each CACC module, the impacts on the behaviours of vehicles following in a string (or string stability) were quantified to establish potential performance enhancements to automated following systems.

Data collection on motorways for traffic management operations is traditionally based on local measurement points and camera monitoring systems. This work looks into social media as an additional data source for the Austrian motorway operator ASFINAG. A system called Driver's Dashboard was developed collecting incident descriptions from Facebook and RSS feeds, filtering relevant messages, and fusing them with traffic data. All collected texts were analysed for concepts describing road situations linking the texts from the web and social media with traffic messages and traffic data. Driver's Dashboard was designed to examine the potential of social media for traffic monitoring with respect to Austrian characteristics in social media use and road transportation with only very few messages available compared with other studies. Of 3586 messages collected within a five-week period, only 7.1% were automatically annotated as traffic relevant. Furthermore, the traffic relevant messages for the motorway operator were analysed more in detail to identify correlations between message text and traffic data characteristics. A correlation between message text and traffic data was found in nine of 11 messages by comparing the speed profiles and traffic state data with the message text.

Traffic diversion is an effective measure to solve the incidental traffic congestion in urban expressway traffic system. By adopting the macroscopic traffic flow model METANET, this study analyses the state change of traffic flow on the road network and establishes the dynamic traffic diversion model, inducing the redistribution of traffic demand. Considering the changes in the amount of origin–destination (O–D) demand, diversion rate is introduced into the basic theory of dynamic O–D model, and then established a dynamic traffic flow model based on dynamic demand change. The genetic algorithm is used to solve the non-linearity problem of the objective function in the traffic diversion model. This study sets up five cases for numerical analyses, and gets the optimal diversion scheme.

This study presents an innovative set of models for accident prediction which are at the core of a web-based platform for road safety simulation and predictions. Specifically, insights into road hazard prediction are given comparing the latest developments of machine learning research to econometric modelling. The paper provides an overview of the above web-platform as well as the description of its built-in models and the early findings of comparing machine learning and econometric methods with respect to crash severity prediction. The original specification of the proposed predictive models embeds, on top of traditional predictors, complex inputs, sporadically or never encountered in previous studies, related to demographics, land use, roadway geometry, traffic control and accident circumstances (special conditions on the road, etc.). The final outcome reveals high accuracy at national level in forecasting the number of casualties from a road crash and its severity. The related models have proved less effective, instead, in those contexts where road collision phenomena turn out exceptional, thus moving away from the national mean behaviour. Finally, the comparison between statistical parametric and machine learning methods, at this early stage is limited to crash severity classification and has pointed out a clear superiority of the parametric approach.

Vehicle platooning, a beneficial feature of automated driving environments, is likely to affect the lane change behaviour of manually driven vehicles (MVs) because identifying proper gaps in vehicle platoons in the target lane will be more difficult than in the current non-platooning environment. These interactions between MVs and automated vehicles (AVs) could lead to a higher potential for unstable traffic flow, which is closely associated with traffic safety issues. The objective of this study was to investigate the lane change characteristics of MVs in AV platooning environments using driving simulator experiments. This research found that MV drivers tend to show more radical driving behaviour, as indicated by greater steering magnitude and steering velocity when they change lanes into the platooning vehicles’ lane. Therefore, novel traffic operations strategies to manage MVs and AVs effectively are required to ensure traffic safety, which will be conducted by a future study.

This study presents a new virtual evaluation approach of preventive safety applications and advanced driver assistance functions. The approach identifies the worst-case scenarios for a given advanced driver assistance function, AEB system in this study, based on field operational tests (FOT) [safety pilot model deployment (SPMD), in this study]. The authors begin with a description of the studied AEB system and a synthesis of the most relevant tests scenarios. Then, they model the distribution of each test parameter retrieved from the SPMD database by applying two estimation methods (kernel method and expectation-maximisation algorithm). A comparison was made between the two methods to choose the best one. These distributions are then sampled using the proposed sampling strategy based on Metropolis-Hastings algorithm. Then, the idea is to take the samples of each parameter retrieved with this sampler, simulate them on a vehicular software simulator (PreScan) and to get their simulation results. For each test and in case of impact, a proportional score to the speed of impact reduction is attributed. Finally, a risk classification is done based on the scoring results which allows to recover high and very high-risk cases to build a set of worst-case scenarios.

Driving is important for older people to maintain mobility. To reduce age-related functional decline, older drivers may adjust their driving by avoiding difficult situations. One of these situations is driving in adverse weather conditions such as in the rain, snow and fog which reduce the visual clarity of the road ahead. The upcoming highly automated vehicle (HAV) has the potential of supporting older people. However, only limited work has been done to study older drivers’ interaction with HAV, especially in adverse weather conditions. This study investigates the effect of age and weather on takeover control performance among drivers from HAV. A driving simulation study with 76 drivers has been implemented. The participants took over the vehicle control from HAV under four weather conditions clear weather, rain, snow and fog, where the time and quality of the takeover control are quantified and measured. Results show age did affect the takeover time (TOT) and quality. Moreover, adverse weather conditions, especially snow and fog, lead to a longer TOT and worst takeover quality. The results highlighted that a user-centred design of human–machine interaction would have the potential to facilitate a safe interaction with HAV under the adverse weather for older drivers.

Both in the US and abroad, various transit agencies have started to implement mobile ticketing applications, which save commuters time during ticket purchase and allows them to avoid surcharges typically incurred when purchasing tickets on-board. Mobile ticketing applications also have a high potential to help transit agencies reduce costs, give them a better understanding of commuter behaviour, and allow them to make their services more efficient. However, it is not a straightforward task to predict the customers’ reaction and gain their trust and to increase the adoption rate of this relatively new technology. A comprehensive evaluation plan is required to detect and fix critical usability problems during both the application development and system wide implementation phases. To address this need, a four-stage evaluation framework is proposed in this study for mobile ticketing technologies in public transit to improve their usability and enhance their adoptability by the potential users. The proposed evaluation framework was employed when New Jersey transit's mobile ticketing application, MyTix, introduced in 2013, was being developed.

Recommending suitable charging spots to drivers on expressways for both charging equipment and electric vehicles (EVs) is an important issue for the spread of EVs. Therefore, the authors developed a recommendation system based on the prediction of the driving ranges of multiple EVs running on expressways. Recommendations are calculated from the energy consumption predicted by data-driven models constructed by actual data on EV trips. In authors’ system, prediction models for popular EV models were constructed with high accuracy. However, the accuracy of prediction is lower for new EV models than for the popular EV models, because the number of trips of new EV models running on the expressway is limited. To solve this problem, the authors propose a new transfer learning method, a type of machine learning that constructs prediction models using other sufficient data on popular EV models. They also evaluated their proposed method using the data on actual EV trips. As a result, the rate of prediction error of authors’ proposed method was reduced by about 30% from that the conventional method. The authors’ proposed method has the potential to predict the energy consumption for new EV models with higher accuracy.

Five UK cities will implement clean air zones to improve air quality (AQ), specifically in relation to oxides of nitrogen, and 253 UK local authorities have declared AQ management areas. Extended range electric vehicles can offer zero emission (ZE) operation but cities lack the ability to monitor and control when vehicles can switch from engine to battery. Project Autonomous and Connected vehicles for CleaneR Air (ACCRA) aims to develop and demonstrate a system allowing hybrid vehicle to become part of a city's urban traffic management control system; monitoring the vehicles’ location and operational state, and controlling the strategy to ensure ZE operation through geofenced zones with poor AQ. This study demonstrates a new methodology to model emissions using state-of-the-art instantaneous vehicle emissions modelling, geographic information system (GIS) and a new method for capturing vehicle behaviour using global positioning system. Probe vehicles were used for the pilot study in Leeds, UK, to capture drive-cycle data and the emissions are modelled. A GIS interface is used to match these emissions to the network. Finally, a high-resolution emissions map is generated to be used as input for AQ dispersion models and the ACCRA decision-making engine for the generation of geofencing AQ zones.