The rapid development of the Internet of Medical Things (IoMT) enables automatic recording of surgical reports via interconnected medical devices. However, the reliability of these data is often compromised by missing data, frequently stemming from intermittent IoT communication issues like network disruptions or device malfunctions. This incomplete data critically hinders downstream medical applications and violates implicit network intents related to data integrity and timeliness within an Intent-based Networking (IBN), essential for supporting proactive resource allocation in operating rooms and optimized surgical scheduling. While existing studies focus on addressing missing event types, event time imputation remains a significant, underexplored challenge due to the need to capture implicit temporal contexts and complex cross surgical procedures dependencies. To tackle this for IoMT, we propose a novel Surgical event time IMputation in Intent-Based Networking(SIM-IBN) model. SIM-IBN employs continuous-time LSTMs with attention mechanisms to learn intra-and inter-sequence correlations, effectively recovering missing timestamps. By enhancing data reliability at the source, SIM-IBN serves as a crucial component enabling IBN systems to better fulfill intents for dependable IoMT operations. Rigorous evaluation on real-world surgical event datasets demonstrates SIM-IBN’s superiority over state-of-the-art baselines by up to 11.88% across various missing data scenarios, validating its potential to enable more reliable IoMT systems and enhance operational efficiency in smart healthcare environments.

Vehicle trajectory prediction serves as a critical enabler for autonomous navigation and intelligent transportation systems. While existing approaches predominantly focus on temporal pattern extraction and vehicle-environment interaction modeling, they exhibit a fundamental limitation in addressing trajectory heterogeneity originating from human driving styles. This oversight constrains prediction reliability in complex real-world scenarios. To bridge this gap, we propose the Driving-Style-Adaptive (\underline{\textbf{DSA}}) framework, which establishes the first systematic integration of heterogeneous driving behaviors into trajectory prediction models. Specifically, our framework employs a set of basis functions tailored to each driving style to approximate the trajectory patterns. By dynamically combining and adaptively adjusting the degree of these basis functions, DSA not only enhances prediction accuracy but also provides \textbf{explanations} insights into the prediction process. Extensive experiments on public real-world datasets demonstrate that the DSA framework outperforms state-of-the-art methods.

Coordinated ramp metering (CRM) is one effective measure to alleviate urban expressway congestion. Traditional model-based methods generally concentrate on single-bottleneck scenarios, while ignoring the case of multiple bottlenecks. In addition, the fixed-sensor fails to fully capture the dynamic traffic characteristics. The rapid development of traffic detection technology has made available a large amount of automatic vehicle identification (AVI) data, which can record detailed individual trajectories. Taking advantage of the AVI data, CRM can be improved. Besides, multi-agent deep reinforcement learning (MADRL) and game theory have been proven to be effective for traffic signal control. These methods can address the challenges faced by CRM, such as solving nonlinear and high-dimensional optimization problems. This paper proposes a distributed CRM strategy with multi-bottleneck to minimize the total travel time and balance the multiple on-ramps equity, using the individual trajectory information from AVI data. Firstly, the paper defines road segment units, road segment groups, and bottlenecks. Next, the problem is formulated as a potential game that captures the interaction among multiple bottlenecks. The controllers utilize the MADDPG algorithm to determine the green duration of the on-ramps. Finally, the proposed strategy is tested on a real-world urban expressway in a microsimulation platform SUMO. Experimental results demonstrate that the proposed strategy performs better than the baseline methods in eliminating mainline congestion and improving the multiple on-ramps equity. Compared to the no-control scenario, the proposed strategy has improved the performance of the system throughput, average travel time, and average mainline speed by 1.31%, 44.36%, and 115.23%.

Origin-destination flow prediction traditionally depends on intricate numerical simulations of human mobility. Data-driven approaches, including transformers, have disrupted this paradigm with sophisticated predictive models. However, purely data-driven approaches often act as closed box models, failing to reveal the underlying mechanisms of human mobility. We address these limitations with physics-informed mobility perception networks (PI-MPN), a continuous-time process that implements a key principle of diffusion from human mobility, and introduces the principle into data-driven deep learning networks. PI-MPN consists of data-driven mobility perception networks and physics-informed neural diffusion networks PI-MPN models precise spatio-temporal movement by data-driven networks, learning urban human mobility as a neural flow by physics-informed networks. We conduct comprehensive evaluations on two real-world traffic datasets, revealing the superiority of PI-MPN over existing data-driven benchmarks. Moreover, case analyses demonstrate that the diffusion process can describe the underlying urban dynamics, thus enhancing the model interpretability of human mobility. Model implementation is available at https://github.com/xuesong-wu/PI-MPN

Multi-agent trajectory prediction is essential in autonomous driving, risk avoidance, and traffic flow control. However, the heterogeneous traffic density on interactions, which caused by physical laws, social norms and so on, is often overlooked in existing methods. When the density varies, the number of agents involved in interactions and the corresponding interaction probability change dynami-cally. To tackle this issue, we propose a new method, called Density-Adaptive Model based on Motif Matrix for Multi-Agent Trajectory Prediction (DAMM), to gain insights into multi-agent systems. Here we leverage the motif matrix to represent dynamic connectivity in a higher-order pattern, and distill the interaction information from the perspectives of the spatial and the temporal dimensions. Specifically, in spatial dimension, we utilize multi-scale feature fusion to adaptively select the optimal range of neighbors participating in interactions for each time slot. In temporal dimension, we extract the temporal interaction features and adapt a pyramidal pooling layer to generate the interaction probability for each agent. Experimental results demonstrate that our approach surpasses state-of-the-art methods on autonomous driving dataset.

Connected and autonomous vehicles offer the possibility to carry out control strategies, thus having great potential to improve traffic efficiency and road safety. The efficient passage of an emergency vehicle calls for the collaborative driving decision-making among multiple vehicles in a dynamically changing local area. However, existing work fails to efficiently adapt to dynamic and complex traffic conditions, thus cannot well solve the task. For better solution, we propose a flexible cooperative multi-agent reinforcement learning approach based on value function factorization, called Q-LSTM. Since the traffic environment is partially observable, the centralized training and decentralized execution paradigm is adopted to learn effective cooperative strategies for individual agents. To flexibly adapt to the changing neighborhood condition around the emergency vehicle, we introduce a long short-term memory network to decompose the learned global value function into local value function of each agent within the neighborhood, whose quantity and entities vary over time. To address the credit assignment problem and realize different roles of the emergency and regular vehicles, reward mechanism and the way agent-wise Q-networks update are well-designed. Extensive experiments are conducted on the Simulation of Urban MObility platform. Results show that our Q-LSTM outperforms state-of-the-art value-based MARL methods. Moreover, the robustness and adaptability of the Q-LSTM are verified in the cases of increased traffic density.

It is important to achieve the real-time and reliable vehicular ad hoc networks (VANETs) for the safety and efficiency of the autonomous intersection management (AIM) system. However, there are challenges in establishing the low-delay and robustness VANETs topology in a horizon due to the high mobility, intersection shelters and malicious attacks. This paper investigates the long-horizon VANETs networking problem with the objective of producing the robust topology in the process of routing paths planning. To achieve this goal, a centralized periodic intervention networking (CPIN) approach is proposed, which consists of three main modules: long-horizon information mapping, full-path routing modeling and proactive optimization mechanism. The first module aims to provide a spatio-temporal delay tensor, considering the mobility and the quality of service (QoS). The second module converts routing path planning into a combinatorial optimization problem to achieve long-horizon full-path routing. The third module introduces a proactive optimization mechanism based on node degree centrality to evolve the VANETs networking process toward high robustness. To find the optimal solution efficiently, we propose an improved Monte Carlo Tree Search (MCTS) algorithm based on the operational characteristics of the AIM system. A series of simulation experiments are carried out to verify the effectiveness of CPIN. The results show that the CPIN outperforms the benchmark method in terms of time delay, wireless hops, and network robustness. Specifically, the CPIN reduces the average time delay by at most 44.4%-48.3% while ensuring relative high robustness.

Since mobility needs grow rapidly, the metro system of modern cities is suffering from the oversaturated situation in the rush hour, which makes the metro system vulnerable and inefficient. Thus, passenger inflow control is proposed and implemented. This paper investigates the station-based passenger inflow control problem with the objective of maximizing overall transport efficiency and fairness. To solve the formulated nonlinear and nonconvex programming model, a novel framework integrating unsupervised subgoal discovery and hierarchical reinforcement learning, named SD-HIC, is proposed. With a series of subgoals, the complicated and long-horizon original task is decomposed and can be solved by reinforcement learning responsively and far-sightedly. A real-world case with operational data in the Guangzhou metro is presented to demonstrate the performance of the proposed model and framework. According to the results, the overall transport utility is improved by 28.87% compared with the benchmark inflow control strategy that is frequently used in daily operations. Through solution algorithm studies and critical parameter analyses, the performance of the proposed passenger inflow control framework is further verified. Notably, the revealed subgoals are also distinguishable and interpretable, which is helpful for the operation staff in practice.

Network-wide traffic signal control (NTSG) is one of the most important factors that impact the transportation network efficiency. Nevertheless, the NTSG problem still faces some issues impeding its development: (1) the traditional flow-based data has poor applicability to the evaluation of the network-wide traffic state, and the identification of the most delayed bottleneck intersections; (2) the existing control methods are difficult to capture the global optimal solution, owing to the lack of collaboration among multiple controllers. In this paper, to overcome the aforementioned challenges, we collect the trip-based data to evaluate the network-wide traffic state, and identify the most delayed trip-based bottlenecks as the controllers. Then, a bi-hierarchical game-theoretic (BHGT) method is proposed to solve the NTSG problem. At the lower layer, the NTSG problem is decomposed into several sub-problems of bottleneck control. A coalition game of intersections is formulated to solve the optimal signal control strategy of each single bottleneck. Furthermore, at the upper layer, a potential game is formulated to collaborate the multiple bottlenecks’ strategies solved from the lower-layer. After several iterations between two layers, the BHGT method will converge to a solution which minimizes the total travel delay of the network. Experimental results on a real-world dataset in Xuancheng City prove that the BHGT method outperforms other baseline methods in reducing the network-wide travel delay, both in low-traffic and high-traffic scenarios.

Missing data is a common and critical problem in the stage of traffic data collection and processing. How to impute the missing values in spatio-temporal traffic data has been a challenging topic for a long time. Recently, a variety of methods have been proposed to impute the missing values. Among them, the tensor-based methods show higher competence in multi-dimensional traffic data imputation. However, the previous studies of tensor factorization rarely considered the joint imputation of multiple correlative data such as traffic speed and traffic volume. In this paper, a novel method called Multi-Task Neural Tensor Factorization (MTNTF) is proposed to learn the non-linear correlation patterns of traffic speed-volume, and then address the joint imputation of traffic speed and traffic volume. Extensive experiments on a real dataset show our MTNTF significantly outperforms the state-of-the-art methods in element-wise missing and fiber-wise missing cases. In addition, our method can impute the slice-wise missing values of traffic volume based on incomplete traffic speed.

As a crucial component in intelligent transportation systems, traffic flow prediction has recently attracted widespread research interest in the field of artificial intelligence (AI) with the increasing availability of massive traffic mobility data. Its key challenge lies in how to integrate diverse factors (such as temporal rules and spatial dependencies) to infer the evolution trend of traffic flow. To address this problem, we propose a unified neural network called Attentive Traffic Flow Machine (ATFM), which can effectively learn the spatial-temporal feature representations of traffic flow with an attention mechanism. In particular, our ATFM is composed of two progressive Convolutional Long Short-Term Memory (ConvLSTM[1]) units connected with a convolutional layer. Specifically, the first ConvLSTM unit takes normal traffic flow features as input and generates a hidden state at each time-step, which is further fed into the connected convolutional layer for spatial attention map inference. The second ConvLSTM unit aims at learning the dynamic spatial-temporal representations from the attentionally weighted traffic flow features. Further, we develop two deep learning frameworks based on ATFM to predict citywide short-term/long-term traffic flow by adaptively incorporating the sequential and periodic data as well as other external influences. Extensive experiments on two standard benchmarks well demonstrate the superiority of the proposed method for traffic flow prediction. Moreover, to verify the generalization of our method, we also apply the customized framework to forecast the passenger pickup/dropoff demands in traffic prediction and show its superior performance. Our code and data are available athttps://github.com/liulingbo918/ATFM.

With the continuous expansion of transportation networks in the urbanization process, monitoring network-wide traffic dynamics with the link resolution has become a significant challenge. Moreover, applying the traditional link-based data into traffic states identification over large-scale networks requires expensive costs. Fortunately, nowadays, we can collect large amounts of trip-based data which makes it possible to tackle this problem. In this paper, we propose a probabilistic framework to infer network-wide link travel time with trip-based data from automatic vehicle identification (AVI) detectors. Different from the link-based data, the AVI data contains multiple attributes, such as trip origin, destination, time stamps, and vehicle license plate. Thus, it is more possible to infer individual traces at a network-wide scale with the AVI detectors. In the proposed framework, we first develop the probabilistic trip travel time allocation (PTTA) model to assign the trip travel time of each vehicle into its traversed links. Especially, the combination impacts of signal control and traffic flow on link travel time are introduced in this model for each allocation. Then, we extract available allocations covered by divided intervals for real-time estimation of network-wide average link travel time. A real-world data set was collected from Xuancheng, Anhui, China. The empirical results show that the performance of PTTA model has a superior level on allocation accuracy over similar methods, and the proposed framework is proved to be reliable at a network-wide scale when the AVI detection sparsity of network is within a certain range.

Destination prediction is crucial for many location based services such as sightseeing places recommendation and targeted advertisements push. Most existing techniques utilize the historical trajectories to predict destinations, but they fail to well describe the spatio-temporal characteristics of trajectories and suffer the trajectory sparsity problem, i.e., the available historical trajectories are hard to cover all probable trajectories. The temporal sensitivity of historical trajectories highlights the sparsity problem even more. In this paper, we address this problem by building a probabilistic generative model to capture the spatio-temporal features of trajectories. We develop an extended Latent Dirichlet Allocation (LDA) model to characterize the generative mechanism of track points in each trajectory. In this model, trajectory, track point of trajectory and destination are regarded as document, word and response respectively. To address the trajectory sparsity problem, each trajectory is expressed by the distribution of trajectory patterns which are the topics discovered from historical trajectories. Then, the most likely destination is predicted through the trajectory patterns. The experiments performed on a real-world taxi trajectory dataset from Guangzhou confirm the advantage of the probabilistic generative model in destination prediction, achieving remarkable accuracy and strong interpretability.

This paper presents a collaborative route discovery method that leverages the experience and preferences of taxi drivers in urban areas. The proposed method is mainly comprised of two phases: collaborative preference discovery (CPD) and intelligent driver network generation (IDNG). In the first phase, given an origin-destination (O-D) pair and provided that the cluster is a road segment set within a time-reachable range, we propose CPD which involves cluster-to-cluster retrieval to capture the top-k routes that are not only frequently traversed by taxis but also neighboring to the O-D pair. In the second phase, to support route computation, an IDNG algorithm is devised to generate an experiential graph for each specific O-D pair. In empirical studies, using the period-based experiential route database, sensitivity analysis is employed to select optimal parameters of intelligent driver networks. The results demonstrate that the routes recommended by our collaborative method are much more reliable than those of the shortest-path method with respect to the variance of travel time. Moreover, the recommended routes are traversed more frequently than those of the fastest-path and the shortest-path methods, while the travel time and route lengths of our routes are approximately equal to those of the conventional methods.