Traffic forecasting is a key task in the field of Intelligent Transportation Systems. Recent research on traffic forecasting has mainly focused on combining graph neural networks (GNNs) with other models. However, GNNs only consider short-range spatial information. In this study, we present a novel model termed LSTAN-GERPE (Lightweight Spatio-Temporal Attention Network with Graph Embedding and Rotational Position Encoding). This model leverages both Temporal and Spatial Attention mechanisms to effectively capture long-range traffic dynamics. Additionally, the optimal frequency for rotational position encoding is determined through a grid search approach in both the spatial and temporal attention mechanisms. This systematic optimization enables the model to effectively capture complex traffic patterns. The model also enhances feature representation by incorporating geographical location maps into the spatio-temporal embeddings. Without extensive feature engineering, the proposed method in this paper achieves advanced accuracy on the real-world traffic forecasting datasets PeMS04 and PeMS08.
During the gradual deployment of connected and autonomous vehicle (CAV) technology, lane management strategies are regarded as potential solutions in the complex traffic flow environment, which consists of connected and autonomous vehicles (CAVs) and human-driven vehicles (HDVs). In this paper, we specify the vehicles as cars and buses to simulate a more realistic mixed traffic flow environment, not limited to CAVs and HDVs. Considering the different driving behavior for various types of vehicles, the effects of lane management strategies considering the real mixed traffic flow environment are explored. First, the driving behaviors under mixed traffic flows are analyzed. Four lane management strategies utilizing the existing bus lane and high occupancy (HOV) lane are designed. A simulation platform is then built using Python and SUMO to obtain vehicle trajectory data. Finally, the emission and energy consumption calculation model based on vehicle-specific power (VSP) is used to quantify the environmental effects of various lane management strategies. The results show that when the penetration rate of CAVs is lower than 30%, four lane management strategies can reduce fuel consumption and emissions by up to 20%. Among them, the strategy of using bus lanes to grant priority to CAVs and buses yields the lowest fuel consumption and emissions. When the penetration rate is lower than 60%, vehicles with priority experience a significant decrease in energy consumption and emissions. (Abstract)
Numerous cities offer a diverse array of funding services to scholars. This study introduces data-driven methodologies to investigate the effects of city-level funding initiatives on annual publication counts and annual cumulative impact factors-two metrics of academic impact. We build a local dataset according to scholars' information from the Scopus database and design a complex network-based Difference-in-Differences (DID) methodology. In addition, we utilize data from the Clarivate's Journal Citation Reports to measure academic achievements and influence. A novel aspect of our research is the data-based scholar-journal publication network construction, which aids in the meticulous selection of a control group, enhancing the robustness and efficiency of our research. Our results demonstrate that the city-level funding service significantly improves the academic output of distinguished scholars. Notably, the study uncovers a temporal lag in the impact of funding on academic output, typically ranging from one to two years. Further analysis reveals that the effectiveness of the project in enhancing scholars' academic impact varies among individuals. Based on these findings, recommendations for the rational adjustment of funding and evaluation criteria for the funding support service are proposed. This paper underscores the importance of a data-driven approach in policy design and assessment, particularly in the realm of academic funding services.
Chatbots, based on artificial intelligence power, has been common across many industries and deeply embedded in customers' daily lives. However, it's been reported that chat bots fail when interacting with customers. This research investi gates the impact of chatbot-made failure on customers' purcha se intention and the underlying mechanisms. Based on expectan cy violation theory and social identity theory, the authors prop ose that erring chatbots (vs. humans) trigger a greater extent of negative expectancy violation, which can be explained by custo mers' intergroup biases, which in turn negatively impact custo mers' purchase intention. The results of our experiment (N=13 2) validate our hypotheses. By exploring mental models and int ergroup bias held by customers, these findings deepen our unde rstandings of customers' perceptions of AI, and provide insight s for alleviating algorithm aversion in a human-machine symbi otic society.
Out-of-hospital cardiac arrest (OHCA) events, with a survival rate of approximately 10%, pose a substantial threat to public safety. While the condition is severe, timely intervention can offer a ray of hope for patients. The emergency first responder service system can harness the potential of social communities to mitigate rescue time delays, not relying solely on professional medical teams. This study develops a simulation framework that integrates agent-based simulation and geographic information systems (GIS), making the simulation environment more realistic and thus the experimental results more reliable. Authorities can utilize our designed framework for comprehensive simulation and strategy validation of emergency incidents, essentially providing a decision support tool. Based on this framework, we propose an ABS-GIS-enabled approach, namely a Screening Assistance strategy, to enhance the efficiency of the current responder system, serving as the initial step in validating the usability of the framework. The ABS-GIS-enabled approach considers factors such as the number of dispatched responders, individual task decline rates, and skill proficiency levels to more precisely target qualified responders for rescue tasks. Numerical experiments reveal that the proposed approach significantly outperforms the current strategy, achieving an improvement of 15% in patient survival rate to improve dispatch robustness. Finally, this study provides valuable insights and guidance for dispatch strategy optimization and outlines future directions for expanding the proposed novel simulation framework to improve the performance of the emergency first responder service system.
Fatigue driving detection technology plays a pivotal role in ensuring road safety, and electroencephalography (EEG) signals can be employed as an objective measure of driver fatigue in intelligent vehicles. However, current EEG-based fatigue driving detection methods encounter certain limitations. Firstly, the restricted receptive field of convolutional neural networks struggles to effectively handle the non-stationary nature of fatigue EEG signals for feature extraction. Secondly, real-world training data often suffers from noisy labels, leading to model overfitting on mislabeled data and consequent degradation in the fatigue detection performance. In this paper, we propose the IE conformer ensemble, a robust EEG-based fatigue driving detection model. The IEconformer architecture integrates multi-scale convolutional layers for local feature extraction and the multi-head attention mechanism to capture global feature correlations. To tackle the challenge of noisy data during training, we introduce the co-teaching plus mechanism into our training scheme. This mechanism facilitates cross-updating each IEconformer using disagreement data that yields minimal loss on the respective IEconformer. Experimental results demonstrate the superiority of our proposed IEconformer ensemble over baseline models in fatigue detection. Particularly, the IEconformer ensemble demonstrates high performance even in the presence of noisy data during the training stage, underscoring the practicality of our approach in fatigue driving detection applications for intelligent vehicles.
Batch processing of small-batch customer orders can effectively reduce operating costs. Manufacturers need third-party logistics (3PLs) to provide services in the production process to quickly deliver finished products to customers. However, the two have very different batch processing of cus-tomer orders: manufacturers merge customer orders with identical demands, while 3PLs merge customer orders in prox-imate regions. Consequently, when both manufacturers and 3PLs are involved in decision-making, the efficiency and economy of production-logistics operations are severely compromised under batch processing. Therefore, the key issue ad-dressed in this paper is how to achieve production-delivery synchronization under batch processing mode, as well as synchronized control in a highly dynamic environment. To ad-dress this issue, a digital twin-based production-delivery synchronized decision framework is proposed, centered on the concept of “cloud-edge-terminal”. A synchronization control mechanism based on digital twin is designed. A collaborative optimization-based synchronized mathematical model and optimization algorithm are developed. Finally, a comparison between two synchronization modes, demonstrates that the proposed solution effectively achieves optimal production-logistics operational states and reduces total costs.
This study explores the appointment scheduling problem for telemedicine consultation services within the context of telemedicine. With the objective of cost minimization, it considers uncertainties in stochastic service times and the availability of doctors. The problem is modeled using a distributionally robust optimization framework, where scenarios are depicted based on relevant uncertain events, and partial distribution information of random variables is extracted from these scenarios to construct scenario-wise ambiguity set. The model is reformulated as a mixed-integer linear programming problem, which can be directly solved using existing solvers. Numerical experiments using real data reveal that the solutions provided by this model are not overly conservative, offering reasonable scheduling solutions for different numbers of patients over a period., and with shorter solution times compared to stochastic programming models. Additionally, sensitivity analyses are conducted on model parameters, investigating the impact of fixed doctor costs and ambiguity set parameters on the results.
In the Philippines, the process of coconut milk extraction (including coconut grating) is commonly aided by a semiautomated machine but due to long working hours, workers performing manual operation often report physical discomfort and fatigue. This study aims to address the issue by optimizing the layout and processes involved in coconut grating. Results of several postural assessments, including Rapid Entire Body Assessment (REBA) and Rapid Upper Limb Assessment (RULA), show that the workers operating in the existing process are exposed to high ergonomic risk. Meanwhile, the body areas that are critically affected were identified using the CMDQ which revealed that majority of workers felt discomfort mostly on their forearm, wrist, upper arm, shoulder, lower back, and upper back. A moderate to strong correlation between RULA total scores and CMDQ scores was found using the Spearman Rho's correlation analysis with a p-value of 0.037 and beta coefficient of 0.457. To mitigate the ergonomic risks in the coconut grating process, a new workstation design was created and proposed by the researchers using Siemens Tecnomatix Process Simulate, employing ergonomic principles that are applicable for improvement.
Developing doctor recommendation techniques has the potential to enhance the efficiency of telemedicine service with the increasing demand for telemedicine. We propose a novel recommendation method tailored for more sparser and more professional telemedicine contexts than online healthcare. Firstly, we construct a knowledge graph based on the expertise of physicians to extract the feature of disease relevance, so as to make up for the sparsity of data. Subsequently, coarse and fine granularity semantic feature is extracted from historical diagnostic data to calculate text similarity between doctors and patients. Then, the features of gender, age, title and scheduling activity are considered to improve model performance doctor modeling. Finally, we input the extracted features into a neural network to generate recommendation results that are both effective and interpretable. Experimental results demonstrate that, compared to traditional methods, our approach significantly improves the accuracy and robustness of telemedicine doctor recommendations. Additionally, interpretability analysis shows text similarity and disease relevance (obtained from doctors' professional expertise and consultation text) contribute mostly to the recommendation system, which reconfirms our efforts are meaningful.
The COVID-19 pandemic, particularly the highly transmissible Omicron variant, presents significant challenges for hospitals in preventing nosocomial transmission. This study investigates the effectiveness of various intervention strategies in mitigating Omicron spread within a tertiary hospital setting in Sichuan Province, China. We employ a novel hybrid simulation model combining Agent-Based Modeling (ABM) and Discrete Event Simulation (DES). This model incorporates factors such as patient demographics, service facility settings in hospitals, social distancing, personal protective equipment (PPE) use, and patient flow patterns. The model evaluates the impact of different intervention scenarios, including adjustments to patient flow, facility arrangements, social distancing protocols, and PPE mandates. This research offers valuable insights for hospital administrators to optimize infection control measures and minimize the risk of Omicron transmission among patients and healthcare workers.
Intelligent logistics with autonomous vehicles is promising to improve the efficiency of rapidly expanding delivery systems. However, complex environments, diverse packages, and irregular human interaction have imposed challenges to the application of unmanned delivery vehicles. Combining various forms of vehicles and developing knowledge-driven delivery services have been the key to further facilitating the upgrading of logistics systems. In this paper, we propose an integrated framework named LogisticsVISTA (Logistics on Vehicles with Intelligent Systems for Transport Automation) comprised of UAVs (Unmanned Aerial Vehicles), UGVs (Unmanned Ground Vehicles), and USVs (Unmanned Surface Vehicles) to provide timely and flexible 3D delivery services. In LogisticsVISTA, we design Large language models (LLMs) and large multimodal models (LMMs) for task planning, active perception, autonomous decision, and human-machine interaction. Besides, virtual-real integrated parallel scenarios are constructed to provide the playing ground for embodied agents efficiently and flexibly.
The problem of service system design aims to determine the optimal design for the structure and parameters of the service system among a set of competitive alternatives. In practice, service systems can be highly complex and large-scale, which sometimes renders analytical approaches not applicable. As a result, stochastic simulation has emerged as a popular choice for the design of service systems. In this research, we consider simulation-based service system design, and focus on the application to the design of electric vehicle charging stations. This problem is formulated as the ranking and selection (R&S) model in simulation optimization, and multi-fidelity simulation is adopted to further improve the efficiency of finding the best system design. We derive the asymptotic optimal sample allocation rule that determines the number of samples allocated to each design-fidelity pair, and develop a selection algorithm for implementation. Numerical results demonstrate that the proposed algorithm has superior performance compared to state-of-the-art competitors.
The collection and application of Medical Service Data (MSD) are crucial for service scientific research and medical decision-making. However, limitations in sparse data availability for rare conditions impede the utilization of MSD. Insufficient data can compromise the rigor and integrity of analytical research. To tackle the issue, we propose a novel Tabular Random Column Rearrangement method based on Diffusion model (TRamCol-Diff). TRamCol allows the model to learn implicit relationships between columns step by step, thereby improving the expressiveness of the data and enhancing accuracy in data generation. TRamCol-Diff employs a diffusion process to automate the input modeling of MSD, simulating the real-world generation process of medical service data. This approach enhances the authenticity and reliability of the synthesized data, providing robust support for downstream analytical tasks. We offer a novel solution for generating authentic, reliable, and diverse medical service data, which is crucial for advancing the quality of medical services and promoting medical scientific research.
Time series prediction poses a significant challenge in statistics and artificial intelligence, drawing considerable attention and sparking extensive research. The complexity of patterns within such data makes it challenging for many models to learn effectively. Additionally, the limited availability of time series data hinders the use of high-capacity models like deep neural networks, which are prone to overfitting with sparse data. To overcome this limitation, simpler models such as ARIMA or SARIMA are often employed, proving effective despite their straightforward nature. Another data category with similarities to time series is textual data, where neural network models have recently shown promise. In this realm, models pretrained for specific tasks and fine-tuned for others have demonstrated excellent results even with limited data or computational resources. This study addresses the time series prediction challenge by introducing llama-time, leveraging the capabilities of the llama 2 language model. Through fine-tuning for time series prediction, our approach yields favorable results on previously unseen data. Consequently, our model functions as a zero-shot learner, proficient in predicting various time series data without prior exposure or assumptions about underlying patterns.
In medical settings, a significant disparity often exists between the rates of positive and negative health outcomes, exemplified by diseases such as cancer and instances of bone fractures. Predictive modeling for these outcomes can be adversely affected by data imbalance. Sample ratio imbalance is still a challenging and critical issue. While many existing studies have explored the imbalance between majority and minority class samples, the internal imbalance within minority class samples has not yet been well addressed. This study proposed and developed a novel algorithm, called Multi-label Random Undersampling and Synthetic Minority Oversampling Technique (ML-RUSMOTE), to address such complex multilabel sample imbalances. We then developed multi-label predictive models by combining the proposed ML-RUSMOTE algorithm with three representative classifiers: Binary Relevance, Multi-label k-Nearest Neighbors, and Multi-label Deep Neural Networks, respectively. The approach was benchmarked against various traditional methods for handling multi-label imbalances. We utilized a substantial cohort of 6,374 patients from three hospital centers in Western Ireland for model validation. Our findings demonstrate that the proposed ML-RUSMOTE algorithm, particularly when integrated with deep learning techniques, significantly outperforms conventional methods in managing multi-label imbalances. Promisingly, the proposed approach can help address common imbalance issues in disease risk predictions, particularly for those patient subgroups whose numbers or outcomes are underrepresented.
The dragged attention from the sustainability triangle has steered up manifold supply chain costs and discarded food. Additionally, the configuration of dynamic temperatures with multimodal transport has been neglected by existing food network designs and most optimisation models. Therefore, this paper proposes a muti-objective, multi-product, and multi-echelon mathematical model for “from-farm-to-fork” logistics networks while considering multi-dimensional sustainability, multimodal transport, and time-dynamic temperature-linked perishability. Due to the trait of the multi-objective optimisation problem, the epsilon constraint approach is employed to obtain separate Pareto-optimal solutions, yielding a crossbreed of trade-offs between total supply chain costs, carbon emissions, and job opportunities. To credit the model, real-life data is collected from the stakeholders and entities involved in the tropical fruit traders in Thailand. This investigation elaborates on the compromised impact in which emphasising one aspect of sustainability on unstable temperature circumstances can simultaneously influence the other sustainable dimensions, aiding policymakers in enclosing the optimal non-dominated solution to company directions for sustainability with the eye of fruit ripeness.
In China, the predominant mode of telemedicine service delivery is Business-to-Business (B2B), with primary care physicians (PCPs) as key users. However, the effective utilization of B2B telemedicine is relatively low. This study used a research model to analyze the “hygiene factors” and “motivators” affecting PCPs' continuous usage intention for B2B telemedicine, considering exogenous, technology, and endogenous dimensions. A valid sample of 421 PCPs from China was analyzed by using SEM and fsQCA. The SEM results revealed that within the individual layer (endogenous dimension), attitude, perceived behavioral control, and intrinsic motivation positively influence PCPs' continuous usage intention, while perceived risk has a negative impact. Additionally, in the technology dimension, assurance and reliability directly impact continuous usage intention, while tangibility indirectly influences continuous intention through the mediation of attitude. However, factors from the environment layer (exogenous dimension) do not significantly influence continuous usage intention. The fsQCA results offer insights into three configurations associated with high continuous usage intention and three configurations triggering low continuous usage intention. And base on these results, factors influencing PCPs' continuous usage intention are classified into “hygiene factors” and “motivators”. Our findings provide theoretical support for policymakers aiming to promote PCPs' engagement in B2B telemedicine adoption.
Applying healthcare data can optimize healthcare services and improve public health management. Data transaction accelerates the flow of healthcare data and facilitates its application. In healthcare data transaction, data brokers take incentives to collect healthcare data from the public and sell the data to healthcare providers. However, choosing the optimal incentives becomes a challenge for data brokers. Meanwhile, how healthcare providers apply healthcare data is critical considering the heterogeneity of consumers. To address these challenges, we study a game-theoretic model with a data broker, two healthcare providers offering homogeneous services, and the public. Our study examines healthcare data transaction under different incentives that data broker adopts for the public (i.e., fixed-subsidy model and revenue-sharing model). Meanwhile, we consider the impact of consumer heterogeneity on healthcare providers. We then analyzed the optimal decisions of each subject. We also compare the benefits of each subject under different incentives. The research findings suggest that revenue-sharing between the public and data broker enhances social welfare optimization. Healthcare providers prefer fixed-subsidy model for low-value data, and revenue-sharing model for high-value data. Data broker has the opposite preference. Besides, an increase in the scale of strategic consumers can benefit healthcare providers and data broker to a certain extent, but beyond a certain point, it can harm public utility. Lower scale of strategic consumers can stimulate healthcare providers' healthcare data application.
Energy storage, pivotal for addressing the challenges of renewable energy's intermittent output, has significantly enhanced the power grid's flexibility, stability, and efficiency. This paper delves into energy storage's economic and societal impacts, exploring five key areas: optimal operational strategies, capacity issues, market impact, social welfare enhancement, and emissions reduction through a systematic review of recent literature. By reviewing the economic impact of energy storage, we recognize the advantages of artificial intelligence (AI) technology and the imperfect market mechanisms. Studies on its social impact indicate that, while driven by commercial interests, energy storage can also enhance social welfare and reduce emissions. This paper underscores the developmental potential of energy storage and advocates for positive engagement from both the market players and government agencies.
