The imminent integration of autonomous vehicles (AVs) with human-driven vehicles (HVs) presents a significant transport paradigm shift, making it imperative to understand traffic flow dynamics in a mixed autonomy scenario for effective traffic management and harnessing the benefits of AVs. This study focuses on investigating the mixed autonomy of day-to-day traffic flow dynamics, leveraging its unique advantages to capture the mixed traffic flow evolution over multiple days, considering individual travelers’ route choices based on past experiences and current conditions. Although some studies have explored the day-to-day dynamics with mixed autonomy, the evolutionary process of the link-level AV penetration rate is usually ignored in the existing literature. In this study, we investigate the mixed autonomy of day-to-day dynamics considering the link-level penetration rate evolution of AVs. In our proposed approach, HV-based travelers use predictions from the advanced traveler information system (ATIS) along with their past route travel time (RTT) data to make route predictions for the current day and subsequently select their routes. Conversely, AV-based travelers employ advanced communication and coordination techniques to predict HV-based travelers’ route choices when ATIS broadcasts its prediction, and then make system-optimal route choices. We develop an optimization model and design a customized Armijo rule-based improved gradient projection (IGP) algorithm to obtain the optimal AV flows. Through numerical experiments, we validate the effectiveness of our approach. The systematic analysis of day-to-day dynamics in mixed traffic environments enhances our understanding of the intricate interactions between AVs and HVs, especially for long-term transport systems planning and management with day-to-day dynamics.

Anomaly detection is a crucial application of traffic management to intelligent transportation systems (ITSs). An ITS utilizes predictive models to identify potential problems and improves system operation reliability by analyzing real-time data streams from millions of sensors, actuators, and other recording devices. However, the collected data typically contain anomalies that can lead to inaccurate traffic system estimates. To evaluate the effectiveness of anomaly detection algorithms on ITS data, comparative analyses of different algorithms are conducted using both labeled and unlabeled data collected from the California Department of Transportation Performance Measurement System. The results highlight the robustness of certain algorithms, such as CBLOF (Cluster-Based Local Outlier Factor), in detecting both point and collective anomalies. Additionally, this study introduces an enhanced dynamic CBLOF algorithm that integrates adaptive windowing and incremental learning, enabling real-time updates and enhanced responsiveness to evolve traffic conditions. Experimental findings demonstrate that this enhancement significantly boosts anomaly detection accuracy while maintaining computational efficiency, making it well-suited for dynamic traffic scenarios. Furthermore, we explored the fusion of traffic domain knowledge with data features in anomaly detection. Our findings emphasize the importance of selecting appropriate algorithms that consider specific problems, evaluation indicators, and data characteristics. This research provides valuable insights for transportation authorities seeking to improve their ability to detect anomalous traffic events, ultimately leading to more effective traffic management and congestion reduction strategies.

Accurate and refined passenger flow prediction is essential for optimizing the collaborative management of multiple collection and distribution modes in large-scale transportation hubs. Traditional methods often focus only on the overall passenger volume, neglecting the interdependence between different modes within the hub. To address this limitation, we propose MM-STFlowNet, a comprehensive multi-mode prediction framework grounded in dynamic spatial-temporal graph modeling. Initially, an integrated temporal feature processing strategy is implemented using signal decomposition and convolution techniques to address data spikes and high volatility. Subsequently, we introduce the Spatial-Temporal Dynamic Graph Convolutional Recurrent Network (STDGCRN) to capture detailed spatial-temporal dependencies across multiple traffic modes, enhanced by an adaptive channel attention mechanism. Finally, the self-attention mechanism is applied to incorporate various external factors, further enhancing prediction accuracy. Experiments on a real-world dataset from Guangzhounan Railway Station in China demonstrate that MM-STFlowNet achieves state-of-the-art performance, with an average improvement of 52.56% in MSE and 36.38% in MAE. Especially during peak hours, it demonstrates excellent forecasting performance, providing valuable insights for transportation hub management. Our model is also demonstrated strong generalization in low-resource scenarios and different traffic scenarios. Our code is available at https://github.com/BMRETURN/MM-STFlowNet

Label Distribution Learning (LDL) offers a promising solution to label ambiguity by employing Label Distributions (LDs) instead of traditional logical labels. However, acquiring LDs for real-world data is both expensive and challenging. To address this issue, Label Enhancement (LE) techniques have been proposed to derive LDs from readily available logical labels. While much of the prior work has focused on enhancing LE for better recovery performance, the ultimate objective remains improving LDL’s overall effectiveness. In this paper, we introduce a novel LE method, Topological Label Enhancement via Optimal Transport (TLEOT), which integrates Optimal Transport (OT) theory with topological space analysis. This method goes beyond improving LE, targeting the enhancement of LDL performance by aligning the feature and label distributions within a unified topological framework. Additionally, we present two innovative topological techniques designed to further improve LDL. Extensive experimental evaluations on real-world datasets demonstrate that TLEOT consistently outperforms nine state-of-the-art methods in predictive tasks. Furthermore, the proposed topological techniques significantly enhance LDL’s performance, validating their practical utility in real-world applications.

Detecting spatiotemporal anomalies is imperative for addressing critical societal and engineering challenges, including public safety assurance, environmental hazard identification, epidemic surveillance, and transportation system optimization. Existing methodologies, however, face persistent limitations due to sparse labeled datasets and the inherent complexity of dynamic spatiotemporal systems. In order to bridge this gap, we present unsupervised-semi-supervised stacking (USemiS), a novel framework that synergizes semi-supervised learning with ensemble meta-learning. USemiS introduces three core innovations: 1) unsupervised component learners that extract low-level representations of heterogeneous anomalies, 2) a consensus-based tuning mechanism that dynamically weights robust learners via stability metrics, and 3) spatiotemporal MixUp (ST-MixUp), a tailored augmentation strategy that interpolates anomalies across spatial and temporal dimensions to enhance decision boundaries. By integrating these components, USemiS effectively disentangles latent anomaly patterns while mitigating label scarcity. Evaluated on large-scale traffic anomaly and crowd fall detection datasets, USemiS achieves state-of-the-art performance, outperforming existing methods by 1.3% and 2.1% in AUC under extreme low-label regimes (0.4% and 0.8% labeled data, respectively). These results underscore USemiS's capacity to generalize across diverse spatiotemporal contexts, offering a scalable and robust solution for real-world applications where labeled anomalies are scarce yet critical.

Traffic flow prediction in spatio-temporal networks is a crucial aspect of Intelligent Transportation Systems (ITS). Existing traffic flow forecasting methods, particularly those utilizing graph neural networks, encounter limitations. When processing large-scale graph data, the depth of these models can restrict their ability to effectively capture complex relationships and patterns. Additionally, these methods often focus mainly on local neighborhood information, which can limit their capability to recognize and analyze global relationships and patterns within the graph data. Therefore, we proposed a deep knowledge distillation model, tailored to effectively capture spatio-temporal patterns in traffic flow prediction. This model incorporates a bidirectional random walk process on a directed graph, enabling it to effectively capture both spatial and temporal dependencies. Utilizing a blend of mutual learning and self-distillation, our approach enhances the detection of spatio-temporal relationships within traffic data and improves the feature perception ability at both local and global levels. We tested our model on two real-world datasets, achieving notable improvements in prediction accuracy, especially for predictions within a one-hour timeframe. In comparison to the baseline model, our proposed model achieved accuracy improvements of 0.19 and 0.18 on the respective datasets. These results highlight the success of using mutual learning and self-distillation to transfer knowledge effectively within and between models and to improve the model’s capability in identifying and extracting features.

The efficient operation of urban bus networks largely depends on optimised scheduling conducted before the one-day operation, crucially relying on reliable origin-destination (OD) information. Passengers travel on direct and transfer trips due to complex infrastructure and services in bus networks. These two differential behaviours necessitate a model that captures topological differences to accurately predict the OD matrix. Responding to this need, we propose a graph-based deep learning model, termed the Direct-Transfer Heterogeneous Graph Network (DT-HGN). This model is designed to predict the OD matrix whilst expressly distinguishing direct and transfer passenger behaviour. DT-HGN articulates direct and transfer trips as distinct graphs, each characterised by its unique adjacency matrix. The model’s architecture embraces two principal blocks: a Spatio-Temporal (ST) construct and an Auto-Encoder (AE) component. The ST-block applies a Gated Recurrent Unit model and a Graph Convolutional Network to discern features of direct and transfer trips, considering both temporal and spatial dimensions. Conversely, the AE-block utilises a heterogeneous graph convolutional network to transmute the two heterogeneous graphs into latent features. Our real-world validation process, executed over a two-month period on an urban bus network, attests to DT-HGN’s robust ability in accurate OD matrix prediction, outperforming contemporaneous state-of-the-art models. This study addresses the crucial need for a comprehensive network-level OD matrix and provides a new perspective for optimising the entire public transport network by accurately depicting station-to-station demand. The approach extends beyond the limitations of traditional bus lines, allowing for a more comprehensive analysis and improvement of urban public transport systems.

Accurate transfer time prediction and future transfer time information are important for both public transport operators and passengers. However, existing studies cannot effectively manage high-dimensional transfer time data, capture the complex nonlinearity of transfer time, or provide accurate transfer time information. This study provides a reliable prediction model called low-rank Bayesian temporal matrix factorization (LBTMF) to address these challenges. First, on the basis of a high-dimensional spatiotemporal matrix of transfer time data, we develop a low-rank temporal-regularized matrix factorization-based imputation module to capture spatial and temporal characteristics to replace missing transfer time data. Second, to further predict the transfer time with the imputation of missing data, we propose the spatiotemporal-based Bayesian temporal matrix factorization prediction module to recover hourly and daily regular characteristics to predict the transfer time at different metro stations during various periods. Finally, the comprehensive experimental findings suggest that the LBTMF model outperforms other excellent approaches in terms of imputation efficiency, prediction accuracy, and robustness.

The urban metro system accommodates significant travel demand and alleviates traffic congestion. Improving metro operational efficiency can increase the metro operator revenue and promote the development of robust urban transportation. To achieve this goal, passenger flow forecasting is a crucial and well-recognized task in metro operation. However, passenger flow forecasting is a challenging task as there exist many unquantifiable factors in resident travel. To address this problem, we propose an innovative model named Interaction Graph Network (IG-Net) to perform passenger flow forecasting at the station level, capable of capturing the non-Euclidean relationships between stations. Three kinds of inter-station interaction graphs are developed to model these inter-station interactions: connectivity, similarity, and temporal correlation graphs. Moreover, we apply multiple channels of graph convolutional neural networks to capture interaction representations and develop a multi-task learning architecture across multiple stations. The proposed IG-Net achieved better performance than the benchmark models when forecasting passenger flow over multiple stations, based on experiments with the Suzhou metro. Finally, we identify the significant effects of interaction graph combinations and multi-task loss functions via further experimentation.

To estimate the accurate fundamental relationship in traffic flow, this paper proposes a novel framework that extends classical fundamental diagram (FD) models to incorporate more dimensions of traffic state variables and allow for the impact of the supply-side factors of roads. The proposed framework is suitable for real-time traffic management, especially in urban areas, due to its reliance on minimal assumptions, its flexibility in adapting to various data sources, and its scalability to higher-dimensional data. The Gaussian process (GP) model is adopted as the base model for learning the optimal mapping from these input features to traffic volume. To enhance the GP model, an in-depth analysis of the properties of its kernel and likelihood function is provided. To cope with the hyperparameter optimisation of the GP, a modified Newton method for GP-based traffic flow model is also designed, which can jump over regions with small gradients. Experiments based on simulation data demonstrate the ability of the proposed framework to capture complex relationships between traffic state variables and supply-side factors, and show its value for estimating dynamic road capacity.

Traffic missing data imputation is a fundamental demand and crucial application for real-world intelligent transportation systems. The wide imputation methods in different missing patterns have demonstrated the superiority of tensor learning by effectively characterizing complex spatiotemporal correlations. However, interval-wise missing volume scenarios remain a challenging topic, in particular for long-term continuous missing and high-dimensional data with complex missing mechanisms and patterns. In this paper, we propose a customized tensor decomposition framework, named the data fusion CANDECOMP/PARAFAC (DFCP) tensor decomposition, to combine vehicle license plate recognition (LPR) data and cellphone location (CL) data for the interval-wise missing volume imputation on urban networks. Benefiting from the unique advantages of CL data in the wide spatiotemporal coverage and correlates highly with real-world traffic states, it is fused into vehicle license plate recognition (LPR) data imputation. They are regarded as data types dimension, combined with other dimensions (different segments, time, days), we innovatively design a 4-way low-n-rank tensor decomposition for data reconstruction. Furthermore, to deal with the diverse disturbances in different data dimensions, we derive a regularization penalty coefficient in data imputation. Different from existing regularization schemes, we further introduce Bayesian optimization (BO) to enhance the performance in the non-convexity of the objective function in our regularized hyperparametric solutions during tensor decomposition. Numerical experiments highlight that our proposed method, combining CL and LPR data, significantly outperforms the imputation method using LPR data only. And a sensitivity analysis with varying missing length and rate scenarios demonstrates the robustness of model performance.

Crowd simulation is a convenient method to evaluate pedestrians’ status and their corresponding management strategies in large public spaces. However, the performance of real-time simulation can be limited by the model’s large-scale computational cost. In order to overcome this difficulty, this study proposes PaCS (Parallel Computation for Crowd Simulation), a parallel computation framework for field-based crowd simulation, based on an enhanced status update method and an efficient task assignment strategy. Parallel computing is introduced with synchronous updates, task division and multiprocessing calculation mechanisms. The movement model is split into the smallest and independent computational units. The field model for simulating the crowd movement has also been improved in terms of weighted multi-direction choice and multi-field environment division. The experiments confirmed that the parallel synchronous algorithm has a significant advantage at the computational scale of more than 10,000 pedestrians. The speedup ratio of the parallel approach can be more than 5 times when simulating one million pedestrians. This framework can help to establish the essential methods for multi-modal transportation systems that require fast simulations for a large-scale crowd. It would also help future digital twin systems to evaluate and validate any potential management strategies when applied in metro stations, railway stations, and other transportation hubs.

Transportation simulation is non-trivial due to the co-existence of thousands of heterogeneous decision makers (or vehicles). Such large-scale decision making is intrinsically a complex decentralized problem, the resolution of which is at the forefront of transportation simulation. Despite many physical or mathematical models proposed to date, underlying them is usually a set of universal rules plus some random perturbations to characterize vehicular movements. Inspired by the decision-making mechanism of rational human beings (i.e., learning iteratively from experience), this study proposes a novel two-stage ensemble reinforcement learning (TERL) paradigm for large-scale decentralized decision making in transportation simulation, in order to enhance the computational efficiency and thus scalability of RL to practical applications at scale. After establishing a problem-specific Markov decision process, the first stage utilizes clustering to group heterogeneous vehicles into quasi-homogeneous clusters. A representative RL model (or agent) is employed for each cluster where the included vehicles share and jointly optimize the policy parameters. The second stage develops an ensemble control strategy based on representative RL models for vehicles isolated as noise during clustering. While vehicles are simultaneously simulated, RL models are separately trained with cluster-specific experience replay. Application of TERL to the classical multi-user dynamic route choice problem in a real-world network of the Gusu District in Suzhou, China demonstrates the effectiveness of the proposed approach in deriving desirable simulation results, compared with the classical shortest path model and the dynamic user equilibrium model.

This paper investigates the spatiotemporal characteristics and predictability of the emerging modern traffic behavior, ridesourcing. We collect a comprehensive data set of Didi ridesourcing cars on a large geographical scale of a capital city in China, including both the temporal order information and the GPS-recorded spatial trajectories. To extract the features of this kind of traffic behavior, we construct a large-scale network by considering every traffic flow of the orders. Therein, a driver consecutively visiting different regions of the city connects the relationship of these sites. The weighted ridesourcing network shows a consistency of the distribution of trip orders and the Clark model for population distribution. The network also has spatial and temporal features with power laws, sometimes with exponential truncations and log-normal distributions. Furthermore, we propose a general analytical method to quantify the predictability of this kind of behavior by calculating the entropy at a collective level, which can be extended to quantify other traffic behaviors. Finally, by considering the traffic congestion factor, we propose a better neural network based model for predicting dwelling time of the ridesourcing behavior. We suggest that the traffic behavior of ridesourcing cars indicates specific non-Markovian characteristics, which can be systematically analyzed from the viewpoint of network sciences.

Urban crowd density prediction is essential for transport demand management and public safety monitoring. Existing studies for crowd density prediction only focus on a few transport modes which do not cover all urban crowds. Moreover, most existing models rely on grid-based divisions, which cannot accommodate unevenly distributed populations. This study proposes a novel model for predicting crowd density in urban areas with mobile-phone signaling data (MPSD). MPSD contain diversified travel and activity information that is not limited to specific transport modes and are practical for activity-travel research. We introduce methods for preprocessing MPSD and extracting information on crowd density and activity choice behaviors (such as activity location, activity type, and activity duration). A novel spatial-temporal convolutional model is proposed for urban crowd density prediction. We design an attention-based feature fusion block, enabling the model to comprehensively consider historical crowd density, activity choices, and external factors (such as weather conditions and holidays). With tailored spatial-temporal convolution blocks, our proposed model successfully captures spatial-temporal information related to crowd movement in different scenes. Unlike existing grid-based models, the proposed model can predict crowd density in irregular-shaped regions. Experimental results show that our model outperforms baseline models that do not apply activity information when selecting different prediction offsets, especially for long-term predictions. The correlation between prediction error and the regularity of crowd activities is also analyzed. The subsequent parameter analysis shows that the influence of external factors on urban crowd movement varies according to activity choices and other travel-related factors like commuting distance.

Short-term estimation and prediction of pedestrian density in urban hot spots (e.g., railway station, shopping mall, etc.) is an important topic for traffic management and control in densely populated areas. In this paper, we propose a short-term pedestrian density estimation and prediction method based on mobile phone data. Firstly, pedestrian density in hot spots is estimated using mobile phone data. To decrease the positioning errors of mobile phone data, a modified particle filter method, which considers the movements of pedestrians, is applied for pre-processing the data. An efficient spatial access method (i.e., Hilbert R-tree) is adopted to construct pedestrians’ position indexes for realizing the short-term estimation. Secondly, based on the estimation results, the spatiotemporal extended Kalman filter (SEKF) is proposed for the short-term prediction of pedestrian density. A massive mobile phone dataset collected in Nanjing, China is used in the case study. The estimated pedestrian density from Monday to Thursday is used for pedestrian density prediction on Friday. The results show that the proposed method can estimate and predict pedestrian density in hot spots, especially in small-scale sites of hot spots efficiently in a short time. Comparing with classical prediction methods, the proposed SEKF method predicts short-term pedestrian density in urban hot spots more accurately.

The link performance and capacity are important quantitative features in road link performance assessment, and they play vital roles in many important transportation tasks, e.g., traffic assignment and dynamic routing. However, it remains a challenging task to quantify them, particularly in a dynamic traffic scenario requiring an accurate, fast, and dynamic output. This study proposes a tailored deep learning framework for the addressed problem, which combines important transport domain knowledge reflected by the Bureau of Public Road (BPR) link performance function. In specifics, the calibration of link performance function and the estimation of link travel time are combined in the proposed framework and realized by two neural network modules. Numerical experiments demonstrate the capability of the proposed framework to capture complex relationships between dynamic link capacity and various factors and show its value in estimating link travel time.

Gaussian process regression (GPR) is an emerging machine learning model with potential in a wide range of transportation system estimation and prediction problems, especially those where the uncertainty of estimation needs to be measured, for instance, traffic flow analysis, the transportation infrastructure performance estimation problems and transportation simulation-based optimization problems. The kernel function is the core component of GPR, and the radial basis function (RBF) kernel is the most commonly used one, suitable for tasks without special knowledge about the patterns of data, like trend and periodicity. However, an inappropriate hyperparameter of the kernel function may lead to over-fitting or under-fitting of GPR. During hyperparameter optimization, the usage of the RBF kernel often suffers from the issue of failing to find the optimal hyperparameter. This paper aims to address this problem by promoting the use of the hat kernel, which can reduce the risk of under-fitting. Moreover, we propose the notion of deformation, corresponding to severe over-fitting of a GPR. To further address this issue, we investigate the connection between deformation and the Bayesian generalization error of GPR. Two lower bounds for the hyperparameter of the hat kernel are also proposed to avoid deformation of GPR.

This paper aims to reveal the uncertainty of traffic flow by introducing a new quantity based on the concept of information entropy (IE). We discover the existence and analyze the properties of IE of traffic flows. It is revealed by both real-world trajectory data and simulation data that the IE of traffic flows can be clearly measured and observed. More importantly, the relationships between IE and other key quantities, those are space mean speed and density, in traffic flow analysis can be described by linear and parabolic functions. We also discover that these relationships are not sensitive to traffic volume. With the inspiration from IE, another new quantity termed speed entropy (SE) is then proposed. Tests with aggregated traffic data from Performance Measurement System (PeMS) show that the pattern of the relationship between SE and flow-weighted average speed illustrates different traffic conditions. In general, a key achievement of the IE analysis is that it gives us a new pathway to better capture the intricate traffic flows from the dimension of uncertainty, thus it has the potential to enhance existing models for traffic data analysis.

We investigate how to effectively and efficiently embed users’ personalized travel behaviors to vectors in this paper. Based on an example scenario of travel mode choice in intelligent transportation system, three data structures representing users’ travel behaviors are defined, namely heterogeneous graph of users’ travel behaviors, user travel behavior$k$-partite graph, and personalized user travel behavior sentence set. This paper systematically analyzes the principle of existing methods and provides intuitions for the problem of learning travel behavior representation in intelligent transportation system. Then we propose the Behavior2vector, which is an improved method tailored for embedding users’ personalized travel behaviors to vectors. In our experiments, we design a travel mode choice model based on machine learning, which uses both hand-crafted basic features and embedded vector features. We further quantify the impact of various factors on travel mode choice and use travel big data to test the hypothesis of traffic assignment models, e.g., travelers always choose the path with the shortest path. In addition, we also compared with the existing graph embedding methods and essentially discussed their advantages and disadvantages.

The integrity of bus route information is fundamental to the analysis of operating status and travel pattern of urban public transport system. However, due to the malfunction of bus positioning devices or delayed update of database, the route information stored in positioning devices might be lost or erroneous. To address this issue, this paper designs a framework matching the bus trajectory with a set of predefined bus routes. The trajectories are first partitioned into segments using the spatio-temporal DBSCAN. Then, the curve similarity between trajectories and bus routes is calculated based on the metric of partial Fréchet distance, which searches for a best mapping between curves that minimizes the maximum distance between vertex pairs. A directed-acyclic-graph-based method is also proposed to compute the partial Fréchet distance. Finally, the best match for each trajectory is given based on the relative ranking. The proposed framework is evaluated on the bus trajectory data of Fuyang and Shenzhen in China. The experimental results demonstrate that the framework can well identify the underlying routes according to the recorded bus trajectories and suggest which route information needs updating.

This paper investigates the continuous network design problem (CNDP) and proposes a simulation-based bi-level model and solution framework based on Bayesian optimization. In the bi-level model, the upper level minimizes the total system cost, and the lower level assigns traffic under an approximated dynamic equilibrium condition corresponding to the given network design strategy. A simulation tool integrated micro- and macro- traffic dynamics, namely SUMO, is employed to solve the lower-level problem. The embedded high-fidelity, non-linear understanding of traffic in the simulator oftentimes gains additional complexity due to the lack of tractable mathematical representation. Thus, a Bayesian machine learning technique is utilized to build the link between simulation and optimization. The proposed solution framework takes both the advantages of fine-grained simulations and the efficiency of surrogate-based optimization techniques. Moreover, lane width expansion is innovatively proposed as the decision variable of CNDP to bridge the gap between theory and practice. The relationship between link free-flow speed (FFS) and lane width is also explored based on real data and established to accurately calibrate the simulation input. For demonstrative purposes, numerical experiments on the optimal lane width expansion design were conducted in the inner city of Suzhou, China. The results show that with proper parameter settings, the proposed method is capable to find the global optimal solution within a very tight computational budget, which makes the simulation-based framework an encouraging option for policymakers to enhance transportation network performance.

The accurate prediction of online taxi-hailing demand is challenging but of significant value in the development of the intelligent transportation system. This article focuses on large-scale online taxi-hailing demand prediction and proposes a personalized demand prediction model. A model with two attention blocks is proposed to capture both spatial and temporal perspectives. We also explored the impact of network architecture on taxi-hailing demand prediction accuracy. The proposed method is universal in the sense that it is applicable to problems associated with large-scale spatiotemporal prediction. The experimental results on city-wide online taxi-hailing demand dataset demonstrate that the proposed personalized demand prediction model achieves superior prediction accuracy.

Deep Neural Network (DNN) has been applied in a wide range of fields due to its exceptional predictive power. In this paper, we explore how to use DNN to solve the large-scale bus passenger flow prediction problem. Currently, most existing methods designed for the passenger flow prediction problem are based on a single view, which is insufficient to capture the dynamics in passenger flow fluctuation. Thus, we analyze the passenger flow from scopes on both macroscopic and microscopic levels, in order to take full advantage of the information from a variety of views. To better understand the role of different views, decision-tree-based models are used in modeling and predicting passenger flow. The defects and key features of decision-tree-based models are then analyzed. The results of the analysis can assist the architecture design of the deep learning network. Inspired by the feature engineering of decision-tree-based models, a modular convolutional neural network is designed, which contains automatic feature extraction block, feature importance block, fully-connected block, and data fusion block. The proposed model is evaluated on the city-wide public transport datasets in Nanjing, China, involving 1,091 bus lines in total. The experiment results demonstrate the outstanding performance of the proposed method in real situations.

Accurate demand prediction plays a significant role in online car-hailing platforms. With ensemble learning, several models can be combined into a single demand predictive model, achieving low prediction error. Nevertheless, the existing ensemble methods are not intended for spatio-temporal data and thus cannot deal with it. In this article, a spatio-temporal data ensemble model is proposed to predict car-hailing demands. Treating the prediction results as various channels of an image, the proposed ensemble module first compresses and then restores the results using the fully convolutional network. Additionally, a skip connection is used to preserve both the fine-grained information in the shallow layers and the deep coarse information. Based on the principle of model as a service, any model can be plugged into our framework as base models to improve the prediction accuracy. Experimental results demonstrate the effectiveness of the presented model.

How to effectively ensemble different base models is a challenging but extremely valuable task. This study focuses on the construction of an ensemble framework designed for spatio-temporal data to predict large-scale online taxi-hailing demand, where an attention-based deep ensemble net is designed to enhance the prediction accuracy. We present three attention blocks to model the inter-channel relationship, inter-spatial relationship and position relationship of the feature maps. Then, the attention maps can be multiplied by the input feature map for adaptive feature refinement. The proposed method is a kind of commonly used ensemble method which applies to large-scale spatio-temporal prediction. Experimental results on city-wide online taxi-hailing demand predictions demonstrate that our proposed attention-based ensemble net is superior to the existing ensemble strategy in terms of the prediction accuracy.

Urban travel time estimation is a strategically important task for many levels of traffic management and operation. Although a number of technologies regarding data and model have been developed recently, it has not been well solved yet given the following challenges: effective modeling approach, data sparsity, and traffic condition fluctuation. In this paper, a tensor-based spatial-temporal model is proposed for citywide travel time estimation, using the big and sparse GPS trajectories received from taxicabs. The travel time of different road segments under different traffic conditions in some time slots is modeled with a third-order tensor. Meanwhile, the occurrence probability of different traffic conditions on different road segments in these time slots are modeled with another third-order tensor. Combined with historical knowledge learned from trajectories, missing entries in the two tensors can be estimated by a context-aware tensor factorization approach. Based on the reconstruction results, for any road segment of the urban road network in the current time slot, we can know not only the travel times under different traffic conditions but also the occurrence probabilities of corresponding traffic conditions. The model incorporates both the spatial correlation between different road segments and the deviation between different traffic conditions, as well as the coarse-grain temporal correlation between recent and historical traffic conditions and the fine-grain temporal correlation between different time slots. The model is applied to a case study for the citywide road network of Beijing, China. Empirical results of extensive experiments, based on the GPS trajectories derived from over 32670 taxicabs for a period of two months, demonstrate that the model outperforms the competing methods in terms of both effectiveness and robustness.