Recent advances in traffic incident detection have increasingly adopted decentralized learning and inference based on locally generated data from distributed traffic sensors. A key challenge in this paradigm lies in the non-independent and identically distributed (non-IID) nature of traffic data across different regions. Moreover, most existing approaches rely on federated learning (FL) with deep neural networks (DNNs) as backbone models. Although effective, these approaches typically require long look-back windows to capture meaningful traffic patterns, often resulting in high detection latency, especially when early-stage abnormal signals are overshadowed by dominant normal patterns within the detection window. In this work, we revisit a classical yet powerful unsupervised anomaly detection method, the one-class support vector machine (OCSVM), and extend it to decentralized learning settings through network lasso (NL), a mathematically elegant but relatively underexplored distributed optimization framework capable of handling data heterogeneity. The proposed NL-OCSVM method enables low-latency unsupervised incident detection in the local traffic region by leveraging data from other regions with similar characteristics, while mitigating the adverse effects of non-IID data. Experiments on several real-world traffic datasets demonstrate that the proposed method achieves low detection latency and competitive accuracy compared to existing DNN-based FL methods.

Growing levels of urbanisation and associated congestion as well as increasing demand from eCommerce present major challenges for carriers in many cities. Disruption due to incidents, extreme weather and construction projects is creating more unreliable urban freight systems. City Logistics aims to minimise the social, environmental as well as maximise the economic benefits of urban freight transport. Hyperconnected City Logistics (HCL) involves more open, integrated and shared systems that rely on a range of connected and automated technologies.International standards provide a framework for developing and implementing City Logistics solutions based on quality management systems. Facility management standards also contain important principles for improving deliveries at major activity hubs. Information and communication technologies such as RFID, dedicated short-range communication (DSRC) and GPS offer a standardised means of charging freight vehicles for using road infrastructure as well as cost effectively monitoring the performance of road networks and managing access for heavy vehicles. International standards are also vital for identifying and tracking loads, consignments and containers for HCL. The Australian Integrated Multimodal EcoSystem (AIMES) provides a unique testbed for examining emerging technologies based on International Standards Organisation (ISO) standards for improving the safety, efficiency and sustainability of urban freight systems.

The prediction of road traffic attributes, such as flow, speed, and density, plays a crucial role in traffic management systems. Deep learning (DL) techniques provide an effective way to predict the future based on historical data. However, it is often impractical to measure all road traffic attributes for DL-based predictions. Given that traffic flow is the most frequently measured traffic attribute, there is a large body of research on DL-based traffic flow prediction. Nevertheless, using only traffic flow data is insufficient to comprehensively depict road traffic conditions. Traffic fundamental diagrams (TFDs) offer a way to estimate other traffic attributes, based on traffic flow. However, the inherent non-monotonic relationship between flow and density poses a significant challenge because a specific flow may correspond to two different densities, reflecting uncongested and congested road conditions. To address this issue, we propose a novel framework for predicting road traffic attributes by hybridizing DL with TFDs. The proposed framework comprises two streams. In the supplementary stream, traffic data (flow, density, and speed) are used to calibrate TFDs and generate congestion labels, which are then used to train a congestion predictor. The main stream relies solely on traffic flow data, aligning with real-world scenarios. One DL model predicts future flow, while another predicts congestion labels based on historical flow data. These labels, combined with the predicted flow, enable the calibrated TFDs to determine density and speed values. Experiments based on a case study of a freeway in Melbourne, Australia, demonstrate the effectiveness of the proposed framework.

Traffic flow forecasting is a crucial task in intelligent transport systems. Deep learning offers an effective solution, capturing complex patterns in time-series traffic flow data to enable the accurate prediction. However, deep learning models are prone to overfitting the intricate details of flow data, leading to poor generalisation. Recent studies suggest that decomposition-based deep ensemble learning methods may address this issue by breaking down a time series into multiple simpler signals, upon which deep learning models are built and ensembled to generate the final prediction. However, few studies have compared the performance of decomposition-based ensemble methods with non-decomposition-based ones which directly utilise raw time-series data. This work compares several decomposition-based and non-decomposition-based deep ensemble learning methods. Experimental results on three traffic datasets demonstrate the superiority of decomposition-based ensemble methods, while also revealing their sensitivity to aggregation strategies and forecasting horizons.

Estimating the large-scale Origin–Destination (OD) matrices for multi-modal public transport (PT) in different cities can vary largely based on the network itself, what modes exist, and what traffic data is available. In this study, to overcome the issue of traffic data unavailability and effectively estimate the demand matrix, we employ several data sets like the total boarding and alighting, smart card as well as the General Transit Feed Specification (GTFS) in order to capture the PT dynamic patronage patterns. First, we propose a new method to model the dynamic large-scale stop-by-stop OD matrix for PT networks by developing a new enhancement of the Gravity Model via graph theory and Shannon's entropy. Second, we introduce a method entitled "Entropy-weighted Ensemble Cost Features" that incorporates diverse sources of costs extracted from traffic states and the topological information in the network, scaled appropriately. Last, we compare the efficiency of a single travel cost versus various combinations of travel costs when using traditional methods like the Traverse Searching and the Hyman's method, alongside our proposed "Entropy-weighted" method; we demonstrate the advantages of using topological features as travel costs and prove that our method, coupled with multi-modal PT OD matrix modelling, is superior to traditional methods in improving estimation accuracy, as evidenced by lower MAE, MAPE and RMSE, and reducing computing time.

This research presents a comprehensive machine learning approach to predicting the duration of traffic incidents, classifying them as short-term or long-term, and understanding what are the factors that affect the duration the most. Our modelling methodology is using a dataset from the Sydney Metropolitan Area that includes detailed records of traffic incidents, road network characteristics, and socio-economic indicators, we train and evaluate a variety of advanced machine learning models including Gradient Boosted Decision Trees (GBDT), Random Forest, LightGBM, and XGBoost. The models are assessed using Root Mean Square Error (RMSE) for regression tasks and F1 score for classification tasks. Our experimental results demonstrate that XGBoost and LightGBM outperform conventional models with XGBoost achieving the lowest RMSE of 33.7 for predicting incident duration and highest classification F1 score of 0.62 for a 30-minute duration threshold. For classification, the 30-minute threshold balances performance with 70.84 https://github.com/Future-Mobility-Lab/SydneyIncidents

Traffic incident detection plays a key role in intelligent transportation systems, which has gained great attention in transport engineering. In the past, traditional machine learning (ML) based detection methods achieved good performance under a centralised computing paradigm, where all data are transmitted to a central server for building ML models therein. Nowadays, deep neural networks based federated learning (FL) has become a mainstream detection approach to enable the model training in a decentralised manner while warranting local data governance. Such neural networks-centred techniques, however, have overshadowed the utility of well-established ML-based detection methods. In this work, we aim to explore the potential of potent conventional ML-based detection models in modern traffic scenarios featured by distributed data. We leverage an elegant but less explored distributed optimisation framework named Network Lasso, with guaranteed global convergence for convex problem formulations, integrate the potent convex ML model with it, and compare it with centralised learning, local learning, and federated learning methods atop a well-known traffic incident detection dataset. Experimental results show that the proposed network lasso-based approach provides a promising alternative to the FL-based approach in data-decentralised traffic scenarios, with a strong convergence guarantee while rekindling the significance of conventional ML-based detection methods.

We study a practical aspect of variants of the Vehicle Routing Problem (VRP).Classically the VRP objective is to find routes for a fleet of vehicles to visitcustomers at minimum cost, subject to capacity and other constraints. Themain contribution of this work is to directly optimise for an important secondaryobjective: the "visual attractiveness" of routes.Optimising for visual attractiveness using an Adaptive Large NeighbourhoodSearch (ALNS), we consider two auxiliary costs: (a) A centroidal cost thatmeasures route compactness, and (b) A novel cost measuring the bending energyof routes. We are the first to optimise the bending energy of routes. We describehow both costs can be evaluated efficiently for the purposes of ALNS.We performed experimental evaluations using problems from our own workwith industry and government, and a large corpus of synthetic problems. Plansoptimised solely with respect to compactness tend to be long (i.e., expensive),and exhibit unattractive shape. Optimising for both compactness and bendingenergy produces compact, short and attractive plans. In a number of realworldcase studies, we also relate visual attractiveness and robustness, bothqualitatively and quantitatively.

The way in which autonomous transport will be adopted is likely to determine the net social benefits delivered by the technology and the sustainability of the transport system. Autonomous vehicles (AVs) will change travel behavior due to reduction in the effort needed for humans to drive a vehicle, the need for them to find a parking space, and the costs related to vehicle operation. The AVs’ benefits are likely to increase their adoption compared to conventional human-driven vehicles, possibly leading to more vehicle kilometers travelled (VKT) and consequently weakening their benefits in large cities particularly if they are used in competition with public transport (PT). This paper evaluates the interplay between AVs and PT, and how road network pricing can be used to influence behavioral changes when personal autonomous vehicles (PAVs) are highly available. An agent-based demand model framework is proposed to estimate the mode share of PAVs and PT based on their perceived travelling costs on a real transport network in Melbourne, Australia. The modelling results suggest that convenient and affordable PAVs could compete with traditional PT and reduce overall PT patronage by up to 10%. However, through considering road network pricing schemes, the role of PAVs could be shifted from competing with PT to a complementary first-and-last-mile service that increases PT share by almost 17%. The results also show that road pricing policies can be used as effective interventions to manage PAV operations by reducing empty vehicle trips by 20%.

Accurate travel time prediction for major freeways and corridors is crucial but challenging when road incidents happen. Data-driven models require a large set of historical data to estimate the spatial and temporal correlations between road incidents and traffic dynamics. More often than not, the amount of historical data under non-recurring conditions is limited when it comes to training the models. This paper investigates the application of data-driven models on an enriched database with simulated travel times. A well-calibrated traffic simulation is used to capture the artificial incident’s impact on a major urban corridor in Sydney, Australia. This procedure is repeated for multiple created incidents, resulting in a synthetic dataset validated by the available actual historical data. Several machine learning models, such as Regression Tree, Support Vector Regression, Extreme Gradient Boosting, and Recurrent Neural Networks are trained and tested based on the simulated travel time and incident information. As a baseline model for comparison, the measured travel time at the prediction time is considered equal to multi-step ahead travel time. Based on the results, the data-driven models developed with the simulated data outperformed the baseline, indicating that our approach can be effectively employed in the travel time prediction.

Evaluating disruptions in public transport (PT) utilisation is challenging due to often stochastic traveller behaviour and missing data information on affected services. This paper proposes a new approach for modelling PT patronage and disruption impact using integrated data-driven modelling and the Fourier transform technique. Firstly, using tap-on and off information of smart-card data, we estimate in-vehicle passenger numbers to integrate as well as trips passing through the incident area. Secondly, considering the PT patronage pattern as a periodic function, we employ the Fourier transform to convert it into a sum of simpler trigonometric functions to filter out the one representing common data noise successfully and generate an accurate profile for a typical day. Thirdly, we introduce an enhanced sensitivity test to improve the model's ability to identify the impact of the disruption. Finally, multiple impact measurement methods are compared to capture the disruption impact. The findings demonstrate the effectiveness of leveraging in-vehicle count to maximise data volume and enhance impact identification. The PT patronage pattern can be effectively modelled using the Fourier transform. The utilisation of the enhanced sensitivity test can effectively filter out unnecessary trigonometric components, resulting in a refined model capable of accurately identifying the impact of disruption.

Physics-informed neural networks (PINNs) are a newly emerging research frontier in machine learning, which incorporate certain physical laws that govern a given data set, e.g., those described by partial differential equations (PDEs), into the training of the neural network (NN) based on such a data set. In PINNs, the NN acts as the solution approximator for the PDE while the PDE acts as the prior knowledge to guide the NN training, leading to the desired generalization performance of the NN when facing the limited availability of training data. However, training PINNs is a non-trivial task largely due to the complexity of the loss composed of both NN and physical law parts. In this work, we propose a new PINN training framework based on the multi-task optimization (MTO) paradigm. Under this framework, multiple auxiliary tasks are created and solved together with the given (main) task, where the useful knowledge from solving one task is transferred in an adaptive mode to assist in solving some other tasks, aiming to uplift the performance of solving the main task. We implement the proposed framework and apply it to train the PINN for addressing the traffic density prediction problem. Experimental results demonstrate that our proposed training framework leads to significant performance improvement in comparison to the traditional way of training the PINN.

Estimating the large-scale demand matrices for multi-modal public transport in different cities can vary depending on the public transport networks, public transport modes, and traffic data. To overcome the issue of traffic data shortage and effectively estimate the Origin-Destination (OD) matrix, we employ several data sets such as the total boarding and alighting as well as the public transport timetable, in order to capture the public transport dynamic patronage when establishing a dynamic and microscopic demand matrix for public transport. In this paper, we propose a new method to model the dynamic large-scale stop-by-stop OD demand for public transport by developing a boosting of the gravity model via graph theory and Shannon's entropy. First, we propose a novel cost matrix estimation method that considers various sources of travel cost features extracted from both the traffic flow information and the topological information in the graph network. Second, we develop a method entitled ``Ensemble Cost Matrix Weighted by Entropy'' to estimate the best weights of importance for each feature using Shannon's Entropy in order to maximise the performance of the cost matrix in the OD matrix estimation. Third, we propose a method for integrating multiple OD matrices across various modes, which was validated by using real smart-card data from the city of Sydney, Australia. Last, by comparing the effectiveness with the traditional deterrence function-oriented methods, we prove that our proposed cost matrix estimation method coupled with a multi-modal OD matrix modelling is superior to traditional methods by approximately 54.46% according to RMSE, 84.44% according to MAPE, and 85.09% according to MAE.

Ramp metering (RM) is a traffic management technique that aims at controlling the flow of traffic entering specific roadways tailored for fast-moving traffic containing separate multilane divided carriageways (such as motorways, highways, expressways, freeways, and turnpikes). The objective of RM is to minimize congestion on the main thoroughfare of the roadway. RM algorithms have evolved significantly since the 1960s and will continue to do so into the future. While the functionalities of the algorithms remain valid through time, the applications of the RM strategies are continually being updated. Unlike previous reviews that focused on the RM methodological aspect, this study details the recent literature regarding the implementation of RM strategies. The aim of this paper is to provide a global perspective on existing RM applications and the algorithms used, for future reference for both academics and practitioners. The paper provides an indicative historical context and characteristics for each reported project, as well as an overview of the evaluation of these schemes. Based on the current understanding of RM strategies, the paper discusses challenges and the potential future of RM technology.

Almost half of the world population lives in rural and remote areas facing more challenging and hazardous life conditions than the rest of the population. This paper and proposed method contribute to addressing an apparent gap in modelling travel-related decisions of rural and remote populations to enhance equitable access to basic amenities (e.g., education, health services, retail) for these populations. The application of these models is highly relevant for planning and policy purposes in remote communities. This paper uses data obtained from a mobility survey of remote Aboriginal communities in Central Australia to estimate purpose-specific destination choice models for access to basic services. The analysis is based on stated preference data to measure the attractiveness of destinations. Then, the models are used to analyse the impact of increasing the number of services provided in the remote communities. A better understanding of the impact of local development on travel patterns in remote communities is provided by analysing three aspects: attracted trips to the destinations, reduction of distance travelled by travellers in all the communities and improved equity in the distribution of travel distance among the communities. The case study illustrates the method successfully determines the most relevant parameters in the decisions of the surveyed individuals and determines potential destinations for local development, with the final objective of contributing to the alleviation of accessibility and inclusion problems in remote areas.

COVID-19 has had a substantial impact globally. It spreads readily, particularly in enclosed and crowded spaces, such as public transport carriages, yet there are limited studies on how this risk can be reduced. We developed a tool for exploring the potential impacts of mitigation strategies on public transport networks, called the Systems Analytics for Epidemiology in Transport (SAfE Transport). SAfE Transport combines an agent-based transit assignment model, a community-wide transmission model, and a transit disease spread model to support strategic and operational decision-making. For this simulated COVID-19 case study, the transit disease spread model incorporates both direct (person-to-person) and fomite (person-to-surface-to-person) transmission modes. We determine the probable impact of wearing face masks on trains over a seven day simulation horizon, showing substantial and statistically significant reductions in new cases when passenger mask wearing proportions are greater than 80%. The higher the level of mask coverage, the greater the reduction in the number of new infections. Also, the higher levels of mask coverage result in an earlier reduction in disease spread risk. These results can be used by decision makers to guide policy on face mask use for public transport networks.

A multi-modal transport system is acknowledged to have robust failure tolerance and can effectively relieve urban congestion issues. However, estimating the impact of disruptions across multi-transport modes is a challenging problem due to a dis-aggregated modelling approach applied to only individual modes at a time. To fill this gap, this paper proposes a new integrated modelling framework for a multi-modal traffic state estimation and evaluation of the disruption impact across all modes under various traffic conditions. First, we propose an iterative trip assignment model to elucidate the association between travel demand and travel behaviour, including a multi-modal origin-to-destination estimation for private and public transport. Secondly, we provide a practical multi-modal travel demand re-adjustment that takes the mode shift of the affected travellers into consideration. The pros and cons of the mode shift strategy are showcased via several scenario-based transport simulating experiments. The results show that a well-balanced mode shift with flexible routing and early announcements of detours so that travellers can plan ahead can significantly benefit all travellers by a delay time reduction of 46%, while a stable route assignment maintains a higher average traffic flow and the inactive mode-route choice help relief density under the traffic disruptions.

This paper assesses the adequacy of the BPR volume delay function for the strategic modelling of Connected and Autonomous Vehicles (CAVs). Three testbed environments are simulated at 10% increments of CAV penetration rates (CPR) to observe network performance in mixed fleet environments. The microsimulation dataset is compared with the BPR travel time predictions to evaluate the need for recalibration. Where appropriate, the BPR modelling parameters are redefined as a function of the CPR. The predictive quality of the recalibrated model is then validated by comparing it against the BPR function on synthetic data. The numerical results indicate an overall improvement in travel time prediction using the recalibrated model, with a significant reduction in root mean square error from 15.16 to 8.86. The recalibrated model also outperformed the traditional BPR model in 67% of the 4620 cases used for validation, and better-predicted travel time by 5.43 times.

Connected and autonomous vehicles (CAVs) are expected to save lives by reducing the frequency and severity of traffic accidents and offer additional benefits to the world by reducing travel time and congestion. Challenges exist preventing widespread implementation of CAV technology; notably CAV susceptibility to rear-end collisions, suspected to be due to poor handling of longitudinal distance control in mixed-fleet traffic. This study aimed to determine if it was possible to improve this longitudinal distance control; increasing safety without sacrificing the performance benefits conferred by CAVs. A fixed coupling distance in a custom CAV control protocol was replaced with a linear dynamic model by incorporating a distance and relative velocity modifier. A sensitivity analysis of the modifiers was performed, comparing network performance statistics and conflict and collision rates. Accident frequency and network performance were observed to be more sensitive to the distance modifier than the relative velocity modifier. An increase in the number of CAVs on the road resulted in a faster network, generally at the expense of safety for conventional vehicles (MVs). A select few combinations of velocity and distance modifiers produced networks that outperformed the base case benchmark in both safety and network performance.

The advent of autonomous vehicles (AVs) is likely to introduce new mobility experiences for travelers. In particular, AVs would allow travelers to get off at the destinations and then drive themselves elsewhere to park rather than cruise for parking or park at a location with a high parking fee. The self-parking capability is likely to increase the utility of private-owned AVs (PAVs) and make this mobility option more attractive than human-driven vehicles. The present study investigates the dynamics of travelers shifting to PAVs from other transport modes and its negative impact on road traffic congestion. To this end, we propose an agent-based demand model which considers different travel cost components depending on crucial travel attributes such as trip purpose and activity duration. The estimated demand is then fed into a mesoscopic traffic simulation model to examine the resulting road traffic conditions. As charging private vehicles for the congestion they cause is an effective tool for demand management and congestion alleviation, we also integrate a distance-based pricing scheme into the overall modeling framework to investigate its impact on mode choice and transport network performance. A case study is conducted in Melbourne, Australia to demonstrate the proposed methodology. The results indicate that the distance-based pricing scheme can effectively limit the usage of PAVs and reduce traffic congestion, especially in the city center and peripheral suburbs.