This paper presents a novel approach to data-driven time-dependent origin–destination (OD) estimation using a joint origin–destination-path choice formulation, inspired by the well-known equivalence of doubly constraint gravity models and multinomial logit models for joint O–D choice. This new formulation provides a theoretical basis and generalizes an earlier contribution. Although including path choice increases the dimensionality of the problem, it also dramatically improves the quality of the data one can directly use to solve it (e.g. measured path travel times versus coarse centroid-to-centroid travel times); and opens up possibilities to combine different assimilation techniques in a single framework: (1) fast shortest path set computation using static (e.g. road type) and dynamic (speed, travel time) link properties; (2) predicting a “prior OD matrix” using the resulting path-shares and (estimated or measured) production and attraction totals; and (3) scaling/constraining this prior using link flows (informative of demand). If the resulting system of equations has insufficient rank, we use principal component analysis to reduce the dimensionality, solve this reduced problem, and transform that solution back to a full OD matrix. Comprehensive tests and sensitivity analysis on 7 networks with different sizes and characteristics give an empirical underpinning of the extended equivalence principle; demonstrate good accuracy and reliability of the OD estimation method overall; and suggest that the method is robust with respect to major assumptions and contributing factors.@en

To promote urban sustainability, many cities are adopting bicycle-friendly policies, leveraging GPS trajectories as a vital data source. However, the inherent errors in GPS data necessitate a critical preprocessing step known as map-matching. Due to GPS device malfunction, road network ambiguity for cyclists, and inaccuracies in publicly accessible streetmaps, existing map-matching methods face challenges in accurately selecting the best-mapped route. In urban settings, these challenges are exacerbated by high buildings, which tend to attenuate GPS accuracy, and by the increased complexity of the road network. To resolve this issue, this work introduces a map-matching method tailored for cycling travel data in urban areas. The approach introduces two main innovations: a reliable classification of road availability for cyclists, with a particular focus on the main road network, and an extended multi-objective map-matching scoring system. This system integrates penalty, geometric, topology, and temporal scores to optimize the selection of mapped road segments, collectively forming a complete route. Rotterdam, the second-largest city in the Netherlands, is selected as the case study city, and real-world data is used for method implementation and evaluation. Hundred trajectories were manually labelled to assess the model performance and its sensitivity to parameter settings, GPS sampling interval, and travel time. The method is able to unveil variations in cyclist travel behavior, providing municipalities with insights to optimize cycling infrastructure and improve traffic management, such as by identifying high-traffic areas for targeted infrastructure upgrades and optimizing traffic light settings based on cyclist waiting times.

@en

Physical distancing has been an important asset in limiting the SARS-CoV-2 virus spread during the COVID-19 pandemic. This study aims to assess compliance with physical distancing and to evaluate the combination of observed and self-reported data used. This research shows that it is difficult to operationalize new rules, that context affects compliance, that there needs to be a need for compliance, and that rules require upkeep. From a methodological point of view, this study found that the combined methods provide a comprehensive picture of compliance behaviour, that it is challenging but essential to mitigate response fatigue in long-term monitoring studies, and that it would be interesting in future research to learn how actual behaviour is influenced by personal narratives.

@en

To analyze inherent and diverse patterns within line-based public transport daily delay occurrences, we introduce a data-driven exploratory analysis focused on the spatial-temporal distribution of these delays. Our approach relies on the utilization of the image pattern recognition technique and k-means clustering algorithm. We extract daily punctuality information from the automatic vehicle location data for a singular public transport route. This information is then translated into a visual representation through aggregated daily delay distribution profile images, offering insights into the spatial and temporal distribution of delays. The delay distribution finds expression in the arrangement of pixels within these profile images. The essence of these images is further distilled through image pattern recognition using the neural network architecture of ResNet50. Employing the k-means algorithm, we cluster these images based on their similarity, revealing five distinct daily delay patterns. The analysis of these patterns offers insight into their unique characteristics, yielding noteworthy outcomes. These findings hold the potential to provide public transport operators with an enriched comprehension of the dynamics of delays occurring on a specific line.

@en

The relationship between real-world traffic and pavement raveling is unclear and subject to ongoing debates. This research proposes a novel approach that extends beyond traditional correlation analyses to explore causal mechanisms between mixed traffic and raveling. This approach incorporates the causal discovery method, and is applied to five Dutch porous asphalt (PA) highway sites that have substantial data sets. Findings indicate a nonlinear relationship between traffic volume and raveling, with road age emerging as a shared contributor. The results also suggest that the degree to which different vehicle types contribute as a causal factor for raveling varies with carriageway configurations and lane characteristics. This underlines the need for targeted maintenance strategies. Challenges remain due to confounding correlations among traffic variables, necessitating further development of causal discovery models. This study may not conclusively resolve the debate on to what extent traffic contributes to raveling, but we argue we provide sufficient evidence against rejecting this hypothesis.

@en

Accurate and reliable short- term passenger flow prediction can support operations and decision-making of the URT system from multiple perspectives. In this paper, we propose a URT multi- step short- term passenger flow prediction model at the network level based on a Transformer-based LSTM network, Depth-wise Attention Block, and CNN network, named as Spatial- Temporal Integrated Prediction Model (STIPM). The STIPM comprises three branches. The first branch takes time- series inflow data as input, and a Transformer-based LSTM network is selected to extract the temporal correlations. The second one takes timestep- based OD data as input, and many spatial and temporal features are captured using Depth- wise Attention Blocks. Meanwhile, timestep- based OD data can better include inter- station relations and global information. The third branch takes Point of Interest data (POI) as input and CNN network is utilized for spatiotemporal features extraction, which can also become the bridge between spatial and temporal features. Moreover, the“Multi-inputmulti- output Strategy”for multi- step prediction is used to obtain a longer prediction period and more detailed information under a relatively high forecasting accuracy. The STIPM is applied to two large- scale real- world datasets from the URT system, and the obtained prediction results are compared with ten baselines and four variants from itself, in which STIPM model achieves highest prediction accuracy indicated by RMSE, MAE, and WMAPE evaluations, which demonstrates the superiority and robustness of the STIPM.

@en

COVID-19 significantly influenced travel behaviours and public attitudes towards public transport. Various studies have illustrated complicated factors related to long-term travel behaviour, indicating difficulty in understanding and predicting post-pandemic long-term travel behaviour via traditional methods. In these complex circumstances, it is valuable to take advantage of social media data to obtain real-time public opinions to understand dynamic travel behaviour changes from the passenger perspective. The present study provides a means - leveraging Twitter data and state-of-art Natural Language Processing (NLP) technologies - to interpret the underlying associations among public attitude, COVID-19 trends and public travel behaviour. Concretely, New York City has been selected due to its dependence on public transit for daily commuting. More than 500K tweets have been collected from January 2019 to June 2022. Automated text mining, topic modelling, and sentiment analysis have been implemented in these contexts to identify dynamic public reactions. A consistently negative attitude to public transit is detected and five main topics, including derivative topics from COVID-19, are discovered within the COVID-19 duration. Policy makers and transit managers can use these topics to take onboard the public's concerns. The paper thus exemplifies how social media data and NLP technologies can support policy-making progress and can benefit other tasks in the transportation domain.

@en

Predictions on Public Transport (PT) ridership are beneficial as they allow for sufficient and cost-efficient deployment of vehicles. On an operational level, this relates to short-term predictions with lead times of less than an hour. Where conventional data sources on ridership, such as Automatic Fare Collection (AFC) data, may have longer lag times and contain no travel intentions, in contrast, trip planner data are often available in (near) real-time and are used before traveling. In this paper, we investigate how such data from a trip planner app can be utilized for short-term bus ridership predictions. This is combined with AFC data (in this case smart card data) to construct a ground truth on actual ridership. Using informative variables from the trip planner dataset through correlation analysis, we develop 3 supervised Machine Learning (ML) models, including k-nearest neighbors, random forest, and gradient boosting. The best-performing model relies on random forest regression with trip planner requests. Compared with the baseline model that depends on the weekly trend, it reduces the mean absolute error by approximately half. Moreover, using the same model with and without trip planner data, we prove the usefulness of trip planner data by an improved mean absolute error of 8.9% and 21.7% and an increased coefficient of determination from a 5-fold cross-validation of 7.8% and 18.5% for two case study lines, respectively. Lastly, we show that this model performance is maintained even for the trip planner requests with prediction lead times up to 30 min ahead, and for different periods of the day. We expect our methodology to be useful for PT operators to elevate their daily operations and level of service as well as for trip planner companies to facilitate passenger replanning, in particular during peak hours.

@en

Understanding the relationship between pavement raveling and traffic characteristics is important to pavement management and maintenance planning. In this work, we propose a framework to empirically quantify this relationship. It consists of an alignment method to tackle the inconsistent spatial-temporal scales of the raveling and traffic measurements and we propose spatial-temporal maps to qualitatively analyze and compare the data. A non-parametric correlation is done on the aligned raveling and traffic flow data. This framework is applied to five study areas in the Dutch highway network. The correlation analysis of the study areas provides empirical evidence to a commonly held theory that traffic flow has effects on raveling. Categorizing the correlation by lanes indicates that the raveling is homogeneous in the through or auxiliary lanes, and the severe raveled sections are parallel to the road discontinuity, suggesting the potential effect of mandatory lane changing on raveling development. The proposed framework can be employed in empirical raveling models that predict raveling based on traffic and other factors.@en

Predictions on public transport ridership are beneficial as they allow for sufficient and cost-efficient deployment of vehicles. At an operational level, this relates to short-term predictions with lead times of less than an hour. Where conventional data sources on ridership, such as Automatic Fare Collection (AFC) data, may have longer lag times, in contrast, trip planner data is often available in (near) real-time. This paper analyzes how such data from a trip planner app can be utilized for short-term bus ridership predictions. This is combined with AFC data (in this case smart card data) to construct a ground-truth on actual ridership. The trip planner data is studied using correlation analysis to select informative variables, that are then used to develop 4 supervised machine learning models (linear, k-nearest neighbors, random forest, and gradient boosting decision tree). The best performing model relies on random forest regression and reduces the error by approximately half compared to a baseline model based on the weekly trend. We show that this model performance is maintained even for prediction lead times up to 30 minutes ahead, and for different periods of the day.@en

In dit artikel worden de effecten van COVID-19 op de (relevante onderdelen van de) aanbodkant van het mobiliteitssysteem beschreven. Gezien de verwachte impact ligt de focus op voetgangersstromen, fietsstromen, het gebruik van deeldiensten en openbaar vervoer voertuigen. We kijken naar het effect op de capaciteit, via gemeten of theoretisch verwachte effecten op het gedrag. We maken hierbij onderscheid tussen eenvoudige infrastructuur elementen (een voetpad, een fietspad) en netwerken of knooppunten, waarbij meer complexe interacties optreden. Dit is relevant om te kunnen inschatten welke vraag door het mobiliteitssysteem kan worden verwerkt, voordat overbelasting (congestie, vertragingen, verdrukking) ontstaat. Het artikel start met een theoretisch raamwerk waarin we de verschillende aspecten die invloed hebben op de aanbodkant van het mobiliteitssysteem in kaart brengen. Vervolgens gaan we in op de verschillende (exit) scenario’s die internationaal zijn uitgerold en waarop deze impact hebben. De effecten zijn zo goed mogelijk gekwantificeerd op grond van a) bevindingen uit de literatuur, b) theoretische analyses, en - waar mogelijk - c) analyse van beschikbare data.De belangrijkste bevinding van dit onderzoek is dat de doorstroomcapaciteit van enkelvoudige infrastructuur elementen met 60 tot 70 procent kan afnemen, mits mensen zich strikt aan de 1,5 meter afstand houden. Indien we naar netwerken of (multimodale) knooppunten kijken, dan zien we dat deze afname groter wordt, afhankelijk van hoe effectief de ruimte in het knooppunt wordt - of kan worden - benut. We laten zien hoe lokale kenmerken zich doorvertalen naar het gehele knooppunt en we laten zien hoe de spreiding in ruimte leidt tot extra reductie van de doorstroomcapaciteit. De effecten met betrekking tot de opslagcapaciteit – hoeveel verkeersdeelnemers kunnen zich maximaal in een netwerk of knooppunt bevinden - zijn nog groter dan voor de doorstroomcapaciteit, wederom gegeven de opvolging van de 1,5 meter maatregel. Deze bevindingen laten zien dat de 1,5 meter maatregelen bij vergelijkbare vraag zou leiden tot enorme afwikkelingsproblemen. Uit eerste data analyses blijkt echter dat de 1,5 meter afstand vaak met voeten wordt getreden en dat veel interacties plaatsvinden op minder dan 1,5 meter afstand. Wel is onze bevinding dat in de verschillende fasen van de COVID-19 crisis de opvolging van de 1,5 meter regel verandert. @en

Is it possible to use just aggregate carriageway data for the evaluation of congestion warning systems (CWS) in large networks—or any system affecting traffic safety for that matter? In this paper, two hypotheses related to this question are tested. The first hypothesis is that it can be done by comparing large-scale congestion patterns on road stretches with and without CWS. The underlying rationale is that heterogeneous congestion patterns with many disturbances, frequent wide moving jams, and large speed differences result in more potentially unsafe traffic conditions than more homogeneous congestion patterns. The second hypothesis is that it is possible to compare differences in average (maximum) deceleration distributions into congestion waves between road stretches with and without CWS. Both hypotheses have been tested for similar bottlenecks with similar demand patterns and the results suggest the first hypothesis must be rejected. Although the idea seems plausible (CWS result in more homogeneous congestion patterns) there were too many confounding factors in the data to make the case. However, persuasive evidence was found for the second hypothesis. Statistically significant differences were found between (maximum) deceleration distributions on road stretches with and without CWS that suggest CWS do—as expected—contribute positively to traffic safety. It thus seems to be possible to monitor safety effects using just average speeds. However, the method is limited to providing relative comparisons. Furthermore, to fully rule out the effects of unobserved factors, more evidence and validation with microscopic data are needed.

@en

The biggest challenge of analysing network traffic dynamics of large-scale networks is its complexity and pattern interpretability. In this work, we present a new computationally efficient method, inspired by human vision, to reduce the dimensions of a large-scale network and describe the traffic conditions with a compact, scalable and interpretable custom feature vector. This is done by extracting pockets of congestion that encompass connected 3D subnetworks as 3D shapes. We then parameterize these 3D shapes as 2D projections and construct parsimonious feature vectors from these projections. There are various applications of these feature vectors such as revealing the day-to-day regularity of the congestion patterns and building a classification model that allows us to predict travel time from any origin to any destination in the network. We demonstrate that our method achieves a 44% accuracy improvement when compared against the consensus method for travel prediction of an urban network of Amsterdam. Our method also outperforms historical average methods, especially for days with severe congestion. Furthermore, we demonstrate the scalability of the approach by applying the method on the entire Dutch highway network and show that the feature vector was able to encapsulate the network dynamics with a 93% prediction accuracy. There are many paths to further refine and improve the method. The compact form of the feature vector allows us to efficiently enrich it with more information such as context, weather and event without increasing the computational complexity.

@en

The fundamental challenge of the origin-destination (OD) matrix estimation problem is that it is severely under-determined. In this paper we propose a new data driven OD estimation method for cases where a supply pattern in the form of speeds and flows is available. We show that with these input data, we do not require an iterative dynamic network loading procedure that results in an equilibrium assignment, nor do we need an assumption on the kind of equilibrium that emerges from this process. The minimal number of ingredients which are needed are (a) a method to estimate/predict production and attraction time series; (b) a method to compute the N shortest paths from each OD zone to the next; and (c) two—possibly OD-specific—assumptions on the magnitude of N; and on the proportionality of path flows between these origins and destinations, respectively. The latter constitutes the most important behavioral assumption in our method, which relates to how we assume travelers have chosen their routes between OD pairs. We choose a proportionality factor that is inversely proportional to realized travel time, where we incorporate a penalty for path overlap. For large networks, these ingredients may be insufficient to solve the resulting system of equations. We show how additional constraints can be derived directly from the data by using principal component analysis, with which we exploit the fact that temporal patterns of production and attraction are similar across the network. Experimental results on a toy network and a large city network (Santander, Spain) show that our OD estimation method works satisfactorily, given a reasonable choice of N, and the use of so-called 3D supply patterns, which provide a compact representation of the supply dynamics over the entire network. Inclusion of topological information makes the method scalable both in terms of network size and for different topologies. Although we use a neural network to predict production and attraction in our experiments (which implies ground-truth OD data were needed), there are straight-forward paths to improve the method using additional data, such as demographic data, household survey data, social media and or movement traces, which could support estimating such ground-truth baseline production and attraction patterns. The proposed framework would fit very nicely in an online traffic modeling and control framework, and we see many paths to further refine and improve the method.

@en

Cities are complex, dynamic and ever-evolving. We need to understand how these cities work in order to predict, control or optimize their operations. We have identified some open issues related to network and data complexity that need to be solved to build feasible methods for these purposes. To this end, we first build multiscale graphs automatically to address a problem that is becoming increasingly relevant in the age of big data, where reducing the network complexity could easily determine the viability of the research in real-world applications. Next, we propose different methods from different fields to extract the essence of network dynamics from the vast amount of spatiotemporal traffic data. One such method is a new way of looking at traffic patterns, combining the field of pattern recognition - with a focus on computer vision - with the traffic domain. The inspiration comes from the fact that humans are the most sophisticated pattern recognizer in the world and we use specific visual features to recognize different complex patterns and we explore if these features can also be used to recognize traffic patterns. Finally, we explore different applications of such mobility patterns such as revealing the unknown correlation between supply and demand patterns, evaluating the scalability of the proposed approach by applying the method to the entire Dutch highway network and transferability by building similar network patterns for public transport networks. Thus, this thesis develops a series of efficient data-driven methods for extracting the mobility patterns of large-scale metropolitan networks and explore some of their applications. With the increasing availability of data in the transport domain, the Achilles heel is not data scarcity anymore but rather extracting insights from this massive amount of data. This thesis is a step forward in solving this complex problem by leveraging the increased acceptance of using machine learning as a worthy and effective method for network-wide analysis of traffic patterns.@en

In large-scale urban agglomerations, heavy rail in the form of metro and commuter train serves as the backbone of the metropolitan public transport network. The objective of this paper is to investigate whether networks with strikingly different structure and development pattern exhibit different robustness properties in the event of random and targeted attacks. We adopt a complex network theory approach, investigating network performances under alternative sequential disruption scenarios corresponding to the successive closure of either stations or track segments. We also investigate the case where the removal of a network node or link implies the closure of all traversing lines Network performance is measured both in terms of the capacity of the network to function in terms of connectivity as well as the additional impedance induced for those that remain connected. An aggregate robustness indicator based on the integral of the deterioration of network performance is adopted. Three exemplary networks are selected, the urban rail networks of London, Shanghai and Randstad. These three networks offer showcases of short and long development patterns, mono- and polycentric agglomeration structures, including the largest and the oldest metropolitan heavy rail networks. The polycentric network of the Randstad was found the least robust in this analysis when compared to the more monocentric networks of London and Shanghai. The London network is in general more robust than the Shanghai network thanks to the presence of cycles beyond the core. Our findings provide more nuanced evidence on the relation between network structure and development pattern, and its robustness.

@en

Smart card data enables the estimation of passenger delays throughout the public transit network. However, this delay is measured per passenger trajectory and not per network component. The implication is that it is currently not possible to identify the contribution of individual system components – stations and track segments – to overall passenger delay and thus prioritize investments and disruption management measures accordingly. To this end, we propose a novel method for attributing passenger delays to individual transit network elements from individual passenger trajectories. We decompose the delay along a passenger trajectory into its corresponding track segment delay, initial waiting time and transfer delay. Using these delay components, we construct a solvable system of equations, using which the delays on each network component can be computed. The estimation method is demonstrated on one year of data from the Washington DC metro network. Our approach produces promising results by compressing millions of individual trajectories into 3D networks, leading to a dimensionality reduction of 94%. Moreover, the mean slack variable value (that can be interpreted as proxies for estimation errors) is smaller than five seconds per passenger, and has the desired positive sign for almost 90% of all travelers. Applications using the estimation results include revealing network-wide recurrent delay patterns, modeling delay propagation and detecting disruptions.

@en

Classification of congestion patterns is important in many areas in traffic planning and management, ranging from policy appraisal, database design, to prediction and real-time control. One of the key constraints in applying machine learning techniques for classification is the availability of sufficient data (traffic patterns) with clear and undisputed labels, e.g. traffic pattern X or Y. The challenge is that labelling traffic patterns (e.g. combinations of congested and freely flow areas over time and space) is highly subjective. In our view this means that assessment of how well algorithms label the data should also include a qualitative component that focuses on what the found patterns really mean for traffic flow operations and applications. In this study, we investigate the application of clustering analysis to obtain labels automatically from the data, where we indeed first qualitatively assess how meaningful the found labels are, and subsequently test quantitatively how well the labels separate the resulting feature space. By transforming traffic measurements (speeds) into (colored) images, two different approaches are proposed to extract the features of a large number of traffic patterns for clustering: point-based and area-based. The point-based approach is widely applied in the image processing literature, and explores local interest points in images (i.e. where large changes occur in color intensity); whereas a new area-based approach combines domain knowledge with Watershed segmentation to partition the images into different spatial-temporal segments from which domain specific features, such as wide moving jam patterns, are extracted. The results show that the Watershed segmentation separates the traffic (congestion) patterns into more meaningful and separable classes, comparable to those that have been proposed in the literature. Since there is no ground-truth set of labels, the quantitative assessment tests how well both methods are able to separate the respective feature spaces they construct for the (large) database of traffic patterns. We argue that the more crisp this separation is; the better the labelling has turned out. For this quantitative comparison we train a multinomial classifier that maps unseen patterns to the labels discovered by each of the two labeling approaches. The most important result is that the classifier using the area-based feature vector achieves the highest average levels of confidence in its decisions to classify patterns, implying a highly separable feature vector space. We argue this is good news! Not only does the combination of image processing (Watershed) and domain knowledge (traffic flow characteristics) lead to meaningful labels that can be automatically retrieved from large databases of data; this method also leads to a more efficient separation of the resulting feature space. Our next endeavor is to further refine and use this method to develop a search engine for the (rapidly growing) 200 TB historical database of traffic data hosted by the Dutch National Datawarehouse (NDW).

@en

On-demand transport has become a common mode of transport with ride-sourcing companies like Uber, Lyft and Didi transforming the mobility market. Recurrent patterns in prevailing demand patterns can be used by service providers to better anticipate future demand distribution and thus support demand-Anticipatory fleet management strategies. To this end, we propose three steps for extracting such demand patterns from travel requests: (1) constructing the origin-destination zones by spatial clustering, (2) composing the hourly and daily origin-destination matrix, and; (3) temporal clustering to extract the dynamic demand patterns. We demonstrate the three step approach on the open-source Didi ride-sourcing data. The data consists of travel requests data for November 2016 from Chengdu, China amounting to approximately 6 million rides. The analysis reveals pronounced and recurrent and thus predictable daily and weekly patterns with distinct spatial properties pertaining to ride-sourcing production and attraction characteristics.

@en