Spatial-temporal Model Research Articles

STC-GraphFormer: Graph Spatial-Temporal Correlation Transformer for In-vehicle Network Intrusion Detection System

The integration of several developing technologies and their applications with Internet of Vehicles (IoVs) techniques has been improved. Utilizing these emerging technologies renders the in-vehicle network more susceptible to intrusions. Furthermore, the utilization of Electronic Control Units (ECUs) in current vehicles has experienced a significant increase, establishing the Controller Area Network (CAN) as the widely used standard in the automotive field. The CAN protocol provides an efficient and broadcast-based protocol for facilitating serial data exchange between ECUs. However, it lacks provisions for security measures such as authentication and encryption. The attackers have exploited these weaknesses to launch various attacks on CAN-based IVN. This paper proposes STC-GraphFormer, an innovative spatial-temporal model that utilizes a Graph Convolutional Network (GCN) and a transformer. The spatial GCN layers are utilized to construct and acquire local spatial features, while the temporal transformer layers are employed to capture the long-term global temporal dependencies. By employing this integrated approach, STC-GraphFormer can learn complex spatial-temporal correlations within the IVN data, enabling it to detect and classify malicious intrusions. The proposed STC-GraphFormer has been validated using five real in-vehicle CAN datasets that cover a wide range of attacks that have not been previously investigated together. The finding results indicate that the STC-GraphFormer is more efficient than the SOTA approaches. It demonstrates excellent performance, with Car-hacking (0.99983), IVN intrusion detection (0.9991), CAN Dataset for intrusion detection “OTIDS” (0.9992), CAR hacking: attack & defense challenge (0.9901), and Survival analysis (0.9982), with a minimal false alarm rate and the highest achievable F1 scores for various types of attacks.

DMS-OGSTV: A Novel Infrared Small Target Detection Based on Spatial-Temporal Tensor Model

DMS-OGSTV: A Novel Infrared Small Target Detection Based on Spatial-Temporal Tensor Model

Enhancing Swift and Socially-Aware Navigation with Continuous Spatial-Temporal Routing

AbstractRouting for autonomous robots in dynamic human environments requires paths that are collision-free, efficient, and socially considerate. This article introduces an optimization-based routing method that operates in continuous space using a spatial-temporal model of crowd dynamics. Our approach anticipates future crowd changes and adjusts routes by considering potential speed variations due to local motion planning. It optimizes navigation speed while avoiding densely crowded areas, ensuring efficient and socially-aware navigation. Simulations in three scenarios demonstrate superior performance compared to benchmark methods in terms of navigation efficiency and adaptability in crowded, dynamic environments.

Short-term forecasting of dissolved oxygen based on spatial-temporal attention mechanism and kernel-based loss function

Dissolved Oxygen (DO) is a fundamental indicator of the health of aquatic ecosystems. This study introduces a novel spatial-temporal attention model combined with Kernel Mean Squared Error loss function (STA-KMSE) for forecasting short-term DO. An attention-based Convolutional Neural Network (A-CNN) is used to incorporate the influence of upstream stations in the spatial module, a factor often overlooked in prior studies. The temporal module captures the influence of both previous hours and days using attention-based Long Short-Term Memory (A-LSTM), offering a more comprehensive temporal context. Our approach addresses the issue of outlier sensitivity and non-linear error patterns by utilizing the KMSE loss function, which improves the model's robustness and more effectively captures intricate variations in DO levels compared to the standard Mean Squared Error (MSE) commonly used in water quality modelling. The model is trained on an extensive dataset consisting of various water quality parameters collected from twenty locations along the Ganga River, its tributary, and nallahs spanning from 1 April 2017 to 30 April 2021. In terms of RMSE, STA-MSE outperforms A-CNN by 62.669 % and A-LSTM by 7.419 %. Compared to STA-MSE, the STA-KMSE shows an average reduction in RMSE by 3.067 % for one-step, 3.197 % for two-step, and 3.428 % for three-step forecasting. Experiments indicate better performance for river locations than for nallahs. SHapley Additive exPlanation (SHAP) analysis highlights pH and temperature as crucial factors for estimating DO in rivers, while level, temperature, conductivity, total suspended solids, pH, chlorine, and chemical oxygen demand are significant for nallahs.

Implications of increased spatial and trophic overlap between juvenile Pacific salmon and Sablefish in the northern California Current

AbstractObjectiveThe study was designed to assess long‐term variability in the distribution of juvenile Pacific salmon Oncorhynchus spp. and Sablefish Anoplopoma fimbria. The study also evaluated whether Sablefish and Pacific salmon shared food resources and looked to characterize Sablefish during an understudied period of their life cycle.MethodsTo meet the objectives, the study used data from 26 years of surface trawls conducted in Oregon and Washington coastal waters (1998–2023). Spatial–temporal models were used to measure changes in abundance and distribution of Pacific salmon and Sablefish along with covariates of ocean temperature. The study evaluated trophic characteristics of Pacific salmon and Sablefish from 2020 for differences. The temporal variation in size and diets of Sablefish were also analyzed, along with energy density of fish caught in 2020.ResultThe spatial–temporal model demonstrated that there has been a nearshore expansion of juvenile Sablefish over the past 26 years that was correlated with increased ocean temperature. The nearshore expansion of Sablefish resulted in increased spatial and trophic overlap with juvenile Pacific salmon. While feeding in nearshore waters, juvenile Sablefish demonstrated competitive feeding advantages over juvenile Pacific salmon during a critical phase of salmonid early marine life history. Juvenile Sablefish exhibited significant ontogenetic diet and energetic shifts, and even the smallest (68–80 mm fork length) were piscivorous.ConclusionsIf juvenile Sablefish numbers continue to increase relative to Pacific salmon, they could exert more competitive pressure, especially if food resources become limited. Pacific salmon may experience adverse effects from competition, regardless of whether or not juvenile Sablefish, which have recently expanded into nearshore waters, successfully recruit to the adult population.

Self-Adaptive Graph With Nonlocal Attention Network for Skeleton-Based Action Recognition.

Graph convolutional networks (GCNs) have achieved encouraging progress in modeling human body skeletons as spatial-temporal graphs. However, existing methods still suffer from two inherent drawbacks. Firstly, these models process the input data based on the physical structure of the human body, which leads to some latent correlations among joints being ignored. Furthermore, the key temporal relationships between nonadjacent frames are overlooked, preventing to fully learn the changes of the body joints along the temporal dimension. To address these issues, we propose an innovative spatial-temporal model by introducing a self-adaptive GCN (SAGCN) with global attention network, collectively termed SAGGAN. Specifically, the SAGCN module is proposed to construct two additional dynamic topological graphs to learn the common characteristics of all data and represent a unique pattern for each sample, respectively. Meanwhile, the global attention module (spatial attention (SA) and temporal attention (TA) modules) is designed to extract the global connections between different joints in a single frame and model temporal relationships between adjacent and nonadjacent frames in temporal sequences. In this manner, our network can capture richer features of actions for accurate action recognition and overcome the defect of the standard graph convolution. Extensive experiments on three benchmark datasets (NTU-60, NTU-120, and Kinetics) have demonstrated the superiority of our proposed method.

Model-enhanced spatial-temporal attention networks for traffic density prediction

Traffic density is a crucial indicator for evaluating the level of service, as it directly reflects the degree of road congestion and driving comfort. However, accurately predicting real-time traffic density has been a significant challenge in Intelligent Transportation Systems (ITS) due to the nonlinear and spatial-temporal dynamic complexity of traffic density. In this paper, we propose a novel Model-enhanced Spatial-Temporal Attention Network (MSTAN), which constructs a spatial-temporal traffic kernel density model using the Kernel Density Estimation (KDE) method to process the spatiotemporal data and calculate the probabilities of various spatiotemporal events. These probabilities are input into the attention mechanism, enabling the model to recognize the inherent connection between dynamic and distant events. Through this fusion, the network can deeply learn and analyze the spatial-temporal properties of traffic features. Furthermore, this paper utilizes the attention mechanism to dynamically model spatial-temporal dependencies, capturing real-time traffic conditions and density, and constructs a spatial-temporal attention module for learning. To validate the performance of the proposed MSTAN model, experiments are conducted on two public datasets of California highways (PeMS04 and PeMS08). The experimental results demonstrate that the MSTAN model outperforms existing state-of-the-art baseline models in terms of prediction accuracy, thus proving the effectiveness of the model both theoretically and practically.

LSTGCN: Inductive Spatial Temporal Imputation Using Long Short-Term Dependencies

Spatial temporal forecasting of urban sensors is essentially important for many urban systems, such as intelligent transportation and smart cities. However, due to the problem of hardware failure or network failure, there are some missing values or missing monitoring sensors that need to be interpolated. Recent research on deep learning has made substantial progress on imputation problem, especially temporal aspect (i.e., time series imputation), while little attention has been paid to spatial aspect (both dynamic and static) and long-term temporal dependencies. In this article, we proposed a spatial temporal imputation model, named Long Short-Term Graph Convolution Networks (LSTGCN), which includes gated temporal extraction (GTE) module, multi-head attention-based temporal capture (MHAT) module, long-term periodic temporal encoding (LPTE) module, and bidirectional spatial graph convolution (BSGC) module. The GTE adopts a gated mechanism to filter short-term temporal information, while the MHAT utilizes position encoding to enhance the difference of each timestamps, then use multi-head attention to capture short-term temporal dependency. The BSGC is adopted to handle with spatial relationships between sensor nodes. And we design a periodic encoding technique to process long-term temporal dependencies. The BSGC handles spatial relationships between sensor nodes, and a periodic encoding technique is used to process long-term temporal dependencies. Our experimental analysis includes completion and forecasting tasks, as well as transfer and ablation analyses. The results show that our proposed model outperforms state-of-the-art baselines on real-world datasets.

HOI-V: One-stage human-object interaction detection based on multi-feature fusion in videos

Effective detection of Human-Object Interaction (HOI) is important for machine understanding of real-world scenarios. Nowadays, image-based HOI detection has been abundantly investigated, and recent one-stage methods strike a balance between accuracy and efficiency. However, it is difficult to predict temporal-aware interaction actions from static images since limited temporal context information is introduced. Meanwhile, due to the lack of early large-scale video HOI datasets and the high computational cost of spatial-temporal HOI model training, recent exploratory studies mostly follow a two-stage paradigm, but independent object detection and interaction recognition still suffer from computational redundancy and independent optimization. Therefore, inspired by the one-stage interaction point detection framework, a one-stage spatial-temporal HOI detection baseline is proposed in this paper, in which the short-term local motion features and long-term temporal context features are obtained by the proposed temporal differential excitation module (TDEM) and DLA-TSM backbone. Complementary visual features between multiple clips are then extracted by multi-feature fusion and fed into the parallel detection branches. Finally, a video dataset containing only actions with reduced data size (HOI-V) is constructed to motivate further research on end-to-end video HOI detection. Extensive experiments are also conducted to verify the validity of our proposed baseline.

Leveraging synthetic assimilation of remote sensing with the National Water Model (NWM) to improve above-normal flow prediction in ungauged basins

Abstract Effective flood prediction supports developing proactive risk management strategies, but its application in ungauged basins faces tremendous challenges due to limited/no streamflow record. This study investigates the potential for integrating streamflow derived from synthetic aperture radar (SAR) data and U.S. National Water Model (NWM) reanalysis estimates to develop improved predictions of above-normal flow (ANF) over the coterminous US. Leveraging the SAR data from the Global Flood Detection System to estimate the antecedent conditions using principal component regression, we apply the spatial-temporal hierarchical model (STHM) using NWM outputs for improving ANF prediction. Our evaluation shows promising results with the integrated model, STHM-SAR, significantly improving NWE, especially in 60% of the sites in the coastal region. Spatial and temporal validations underscore the model’s robustness, with SAR data contributing to explained variance by 24% on average. This approach not only improves NWM prediction, but also uniquely combines existing remote sensing data with national-scale predictions, showcasing its potential to improve hydrological modeling, particularly in regions with limited stream gauges.

Enhanced federated recognition mechanism based on spatial-temporal model with split learning for multi-view human activity classification in edge intelligent network

Enhanced federated recognition mechanism based on spatial-temporal model with split learning for multi-view human activity classification in edge intelligent network

Spatial–Temporal Analysis of Greenness and Its Relationship with Poverty in China

Ecological environmental protection and poverty alleviation are of great significance for the study of human–land relationship coordination and sustainable development, and they have also been a focus of attention in China in the past few decades. In this study, we chose 13 contiguous poverty-stricken areas in China as the study area. Using MODIS Leaf Area Index (LAI) data from 2000 to 2020, the spatial–temporal changes in greenness were obtained using the Bayesian spatial–temporal model (BYM). Spatial autocorrelation was used to identify the spatial distribution of poverty using socio-economic statistical data. Driving factors, including natural factors, poverty factors, and the Grain for Green Policy (GTGP), and their influence on greenness were analyzed by using the Geodetector model for detecting spatial differentiation and factors’ interactions. The results showed the following: (1) In 13 contiguous poverty-stricken areas (CPSAs) in China, 59% of the area presented an increasing trend of greenness. (2) In 2000, the high poverty levels with larger MPI values were widely distributed. After 20 years, the overall MPI value was lower, except in some northwest regions with increased MPI values. The spatial autocorrelation of poverty, which relates to the mutual influence of poverty in adjacent areas, also decreased. (3) In the study area, 65.24% of the regions showed strong synergistic effect between greening progress and poverty reduction in the interaction between poverty status and green development. With the improvement of greenness level, the positive correlation between poverty alleviation and ecological environment improvement has become increasingly close. (4) The impacts of interaction factors with the highest q values changed from temperature interacting with precision to regional division interacting with the Grain for Green Policy. The conclusions are that from 2000 to 2020, the impact of natural factors, geographical division, and poverty status on greenness has shown a decreasing trend; The effect of the Grain for Green Policy is gradually increasing; At the same time, the interaction and overlapping effects between the Grain for Green Policy and poverty were increasing. Taking into account the needs of ecological environment, poverty alleviation, and rural revitalization, this research provides valuable reference for formulating and implementing relevant policies based on the actual situation in different regions to promote harmonious coexistence between human-land relationship.

Spatial-temporal Data Model for Indoor Fire Response Considering Multi-Factors

Abstract. The complexity of high-rise building structures and functions significantly increases the challenges in emergency responses to indoor fire incidents. This complexity is due to multiple factors that must be evaluated for decision-making within the fire scenario, such as the entities within the building, the status of environment, and the actions of people. An effective and precise indoor fire emergency management model should fully take into consideration of multiple factors, aiming to ensure the safety and effectiveness of emergency response action.This paper proposes a spatial-temporal data model to integrate the entity-status-action, for designing the most suitable action strategies in various environmental status by considering all related entities in the indoor fire scenario. Firstly, a list of factors is designed to represent the scenario features including spatial location, semantic description, attribute characteristics, geometric form, evolutionary process, and the relationships among objects. Then, employs an ontology-based descriptive method to integrate multi-dimensions and effectively represents the interrelations among entities, actions, and environmental status within the context of an indoor fire scenario. These environmental status considered in the model focus on the supporting path-planning during execution of direct response actions. Finally, a set of path-planning rules has been developed to infer a series of environmental status supporting to formulate the optimal evacuation strategy to minimize the risk of evacuation failure. The case study shows that the proposed model is effective in unified expression for multitude of factors associated with emergency response in indoor fire scenario and supporting the decision-making of the emergency response actions.

A novel physics-guided spatial-temporal data mining method with external and internal causal attention for drilling risk evaluation

As drilling technology advances and operations extend into more complex geological environments, evaluating drilling risks has become increasingly complex, challenging the effectiveness of traditional methods. The novel physics-guided spatial-temporal data mining method that integrates external and internal causal attention mechanisms for drilling risk evaluation is proposed to address this issue. Firstly, a risk calibration method based on spatial-temporal sequence clustering is designed. This method dynamically calibrates drilling risks by mining subtle changes in sign data during drilling. Secondly, an expert experience extraction method based on the correlation of drilling risk signs and a fuzzy inference system is established. Kendall's tau is used to quantify the correlation between drilling risk signs. The fuzzy inference system is employed to convert fuzzy and difficult-to-quantify expert experience into computable and interpretable rules. In order to improve the flexibility and adaptability of the fuzzy inference system, an expert experience rules base is also constructed. Subsequently, a spatial-temporal data mining model integrating both external and internal causal attention mechanisms (STMIEICAM) is constructed. The external causal attention mechanism (ECAM) quantified the correlation between signs and risk. The internal causal attention mechanism (ICAM) improved the model's ability to capture and quantify the features of spatial-temporal sequences. Finally, the physical knowledge from the fuzzy inference system and well-site is embedded into the STMIEICAM model, forming a physics-guided spatial-temporal data mining model integrating both external and internal causal attention mechanisms (PG-STMIEICAM) that enables graded evaluation of drilling risks. The proposed method was applied to overflow risk evaluation in an oil field to validate its effectiveness. The results demonstrate that the method not only excels in uncovering hidden relationships within the data but also integrates expert knowledge, achieving accurate evaluation of drilling risks.

Neural dynamical operator: Continuous spatial-temporal model with gradient-based and derivative-free optimization methods

Data-driven modeling techniques have been explored in the spatial-temporal modeling of complex dynamical systems for many engineering applications. However, a systematic approach is still lacking to leverage the information from different types of data, e.g., with different spatial and temporal resolutions, and the combined use of short-term trajectories and long-term statistics. In this work, we build on the recent progress of neural operator and present a data-driven modeling framework called neural dynamical operator that is continuous in both space and time. A key feature of the neural dynamical operator is the resolution-invariance with respect to both spatial and temporal discretizations, without demanding abundant training data in different temporal resolutions. To improve the long-term performance of the calibrated model, we further propose a hybrid optimization scheme that leverages both gradient-based and derivative-free optimization methods and efficiently trains on both short-term time series and long-term statistics. We investigate the performance of the neural dynamical operator with three numerical examples, including the viscous Burgers' equation, the Navier–Stokes equations, and the Kuramoto–Sivashinsky equation. The results confirm the resolution-invariance of the proposed modeling framework and also demonstrate stable long-term simulations with only short-term time series data. In addition, we show that the proposed model can better predict long-term statistics via the hybrid optimization scheme with a combined use of short-term and long-term data.

Spatio-temporal grid-based storage method for aerospace operation and control data

Abstract The current aerospace operation and control data system has problems such as difficult integration organization and difficult data retrieval in terms of mission planning efficiency, which limits the development of satellite operation and control systems. This paper is oriented to the organization and management of aerospace operation and control data, according to the nature of the current big data indexing model. Based on the principle of earth dissection, we aim to design a grid that can be indexed. First of all, the Earth is dissected to form a hierarchical and uniformly distributed global geographic grid, and then the formed grid is encoded to store satellite attribute information and orbit data for mission planning. Finally, through the data retrieval experiment, the retrieval time is within 100 milliseconds under ten million pieces of data, which proves that the spatial-temporal grid model has the characteristics of efficient and fast retrieval. It can provide the basic information for satellite mission planning to further improve the efficiency of satellite mission planning and management and the efficiency and quality of satellite mission planning.

Collaborative Joint Perception and Prediction for Autonomous Driving.

Collaboration among road agents, such as connected autonomous vehicles and roadside units, enhances driving performance by enabling the exchange of valuable information. However, existing collaboration methods predominantly focus on perception tasks and rely on single-frame static information sharing, which limits the effective exchange of temporal data and hinders broader applications of collaboration. To address this challenge, we propose CoPnP, a novel collaborative joint perception and prediction system, whose core innovation is to realize multi-frame spatial-temporal information sharing. To achieve effective and communication-efficient information sharing, two novel designs are proposed: (1) a task-oriented spatial-temporal information-refinement model, which filters redundant and noisy multi-frame features into concise representations; (2) a spatial-temporal importance-aware feature-fusion model, which comprehensively fuses features from various agents. The proposed CoPnP expands the benefits of collaboration among road agents to the joint perception and prediction task. The experimental results demonstrate that CoPnP outperforms existing state-of-the-art collaboration methods, achieving a significant performance-communication trade-off and yielding up to 11.51%/10.34% Intersection over union and 12.31%/10.96% video panoptic quality gains over single-agent PnP on the OPV2V/V2XSet datasets.

A generic fusion framework integrating deep learning and Kalman filter for state of charge estimation of lithium-ion batteries: Analysis and comparison

Lithium-ion batteries (LIBs) are extensively utilized in power and energy storage systems. Accurately estimating state of charge (SOC) of LIBs using single methods is still challenging to fully address the problem associated with SOC. This paper proposes a novel generic fusion framework that integrates deep learning (DL) and Kalman filter (KF), characterized by an effective fusion of the results obtained by multiple-methods. Within the DL-based method, a spatial-temporal awareness model and an improved dung beetle optimizer are proposed to enhance the capture of spatial-temporal features and optimize hyperparameters. Additionally, three adaptive nonlinear KFs are employed for SOC estimation based on a simplified electrochemical model, with the adaptive cubature KF emerging as the most suitable following comparative analysis. Subsequently, the results from both approaches are integrated to derive the final result using various fusion methodologies. The effectiveness of the fusion framework is validated using experimental data, with kernel ridge regression emerging as the optimal fusion method. The best fusion results yield a mean absolute error of 0.2341% and a root mean square error of 0.3072%, demonstrating that proposed framework can effectively process and utilize multiple-source information, thereby enhancing accuracy and robustness compared to single methods and exhibiting adaptability to complex situations.

Extracting regional and temporal features to improve machine learning for hourly air pollutants in urban India

India is suffering from severe particulate matter (PM, including PM2.5 and PM10) pollution, while limited ground observations are insufficient to support a comprehensive understanding of its health risks. Machine learning (ML) has the potential to improve the estimation of PM distribution and exposure efficiently. Regional transport as well as accumulation and dispersion processes of PM and its components, which have significant impacts on PM concentrations, are crucial when building ML models, especially for sparsely observed regions like India. Here, geographic and temporal-rolling weighting methods were used to separately extract regional and temporal features for improving the performance of the ML model. The incorporation of temporal and regional features into the ML model significantly improved ML model performance, with root mean square error (RMSE) reduced by 21 % and 19% for PM2.5 and PM10 estimation, as well as an improvement in model underestimation for the heavy pollution scenarios. The spatial-temporal model shows out-of-sample test CV coefficients of determination (R2) of 0.87 and 0.88 for hourly PM2.5 and PM10. The ML model predicts an annual nationwide concentration of 68.3 μg/m3 for PM2.5 with a north (high, especially in Indo-Gangetic Plain) to south (low) distribution, which is consistent with high satellite aerosol optical depth (AOD) values. Boundary layer height is identified as the main meteorological factor influencing PM2.5 concentrations in winter. Characterizing the regional transport and cumulative dispersion processes of pollutants by extracting features can help in machine learning training, and this method can be further improved and applied to other studies.

A Bayesian spatial–temporal varying coefficients model for estimating excess deaths associated with respiratory infections

Abstract Disease surveillance data are used for monitoring and understanding disease burden, which provides valuable information in allocating health programme resources. Statistical methods play an important role in estimating disease burden since disease surveillance systems are prone to undercounting. This paper is motivated by the challenge of estimating mortality associated with respiratory infections (e.g. influenza and COVID-19) that are not ascertained from death certificates. We propose a Bayesian spatial–temporal model incorporating measures of infection activity to estimate excess deaths. Particularly, the inclusion of time-varying coefficients allows us to better characterize associations between infection activity and mortality counts time series. Software to implement this method is available in the R package NBRegAD. Applying our modelling framework to weekly state-wide COVID-19 data in the US from 8 March 2020 to 3 July 2022, we identified temporal and spatial differences in excess deaths between different age groups. We estimated the total number of COVID-19 deaths in the US to be 1,168,481 (95% CI: 1,148,953 1,187,187) compared to the 1,022,147 from using only death certificate information. The analysis also suggests that the most severe undercounting was in the 18–49 years age group with an estimated underascertainment rate of 0.21 (95% CI: 0.16, 0.25).