Spatio-temporal Correlation Research Articles

PI-STGnet: Physics-integrated spatiotemporal graph neural network with fundamental diagram learner for highway traffic flow prediction

At present, traffic state prediction primarily relies on purely data-driven methods, ignoring the incorporation of physical constraints within the field of traffic flow. Taking this as a starting point, this paper endeavors to embed the physical mechanism of the traffic flow fundamental graph into the deep learning model, and proposes a physics-integrated spatiotemporal graph neural network with fundamental diagram learner (PI-STGnet) for highway traffic flow prediction. The PI-STGnet mainly comprises a semantic enhancement module (SE), a spatiotemporal correlation extraction module (ST-Block), and a fundamental diagram learner module (FD-learner). These modules are strategically designed to sequentially enhance the contextual semantic relationships of the traffic flow and speed inputs, extract intricate spatiotemporal correlations of traffic states, and adaptively learn the dynamic evolution of the physical relationships between traffic flow and speed. The real-world dataset employed to assess the performance of the PI-STGnet is sourced from the monitoring data of highway gantry sensors located in Ningde City, Fujian Province, China. The experimental results demonstrate that the proposed PI-STGnet has the advantages to extract spatiotemporal features and achieve prediction, surpassing the accuracy of cutting-edge baseline models. In summary, as an exploratory work, this paper provides a novel approach that driven by the fusion of deep learning and physical mechanisms for traffic flow prediction to a certain extent.

3DTCN-CBAM-LSTM short-term power multi-step prediction model for offshore wind power based on data space and multi-field cluster spatio-temporal correlation

The accuracy of offshore wind power forecasting (OWPF) is the basis for guaranteeing the safe dispatch and economic operation of power systems, which can reduce the technical and economic risks faced by power market participants. Combining the theoretical approaches of data space and deep learning, this study proposes a hybrid short-term OWPF framework based on the integrated consideration of spatial and temporal correlations of regional field clusters in the data space and the 3DTCN-CBAM-LSTM model. Firstly, other offshore wind farms with spatio-temporal correlation with the target offshore wind farm are screened out using the K-means spatial clustering algorithm. The historical power sequences determined to be similar to the target offshore wind farm and feature data of target offshore wind farms are used as inputs to the prediction model to improve the efficiency of the framework's power prediction. Secondly, the short-term power prediction framework of offshore wind power based on the 3DTCN-CBAM-LSTM model is constructed, using the 3DTCN model's ability to capture the spatio-temporal dynamic characteristics of offshore wind farm data to improve the accuracy of the prediction model. The temporal characteristics of the data are strengthened using the LSTM model to improve the stability of the prediction model. Finally, the CBAM mechanism is introduced to combine channel attention and spatial attention to eliminate irrelevant information and capture the spatio-temporal characteristics of the data in a more targeted way, which further improves the accuracy of OWPF. To verify the practicality and reliability of the wind power prediction framework described in this study, the differences between the prediction results of the framework and those of the benchmark model are compared using single-step prediction and multi-step prediction (4, 6, 8, and 12 steps). Compared with the benchmark model (BPNN), the single-step prediction results show that the MAE, MAPE, and MSE metrics of the 3DTCN-CBAM-LSTM model are reduced by 20.004 MW, 50.13%, 55.61%, and 80.29%, respectively, and the R2 is improved by 3.39%, which is a substantial improvement in prediction performance. The multi-step prediction results show that, with the progression of the prediction steps, the 3DTCN-CBAM-LSTM model has the highest prediction accuracy and model fit at each step. Therefore, the proposed 3DTCN-CBAM-LSTM model presents obvious advantages and reliability in the prediction of offshore wind power multi-step power.

General Relations between Stress Fluctuations and Viscoelasticity in Amorphous Polymer and Glass-Forming Systems.

Mechanical stress governs the dynamics of viscoelastic polymer systems and supercooled glass-forming fluids. It was recently established that liquids with long terminal relaxation times are characterized by transiently frozen stress fields, which, moreover, exhibit long-range correlations contributing to the dynamically heterogeneous nature of such systems. Recent studies show that stress correlations and relaxation elastic moduli are intimately related in isotropic viscoelastic systems. However, the origin of these relations (involving spatially resolved material relaxation functions) is non-trivial: some relations are based on the fluctuation-dissipation theorem (FDT), while others involve approximations. Generalizing our recent results on 2D systems, we here rigorously derive three exact FDT relations (already established in our recent investigations and, partially, in classical studies) between spatio-temporal stress correlations and generalized relaxation moduli, and a couple of new exact relations. We also derive several new approximate relations valid in the hydrodynamic regime, taking into account the effects of thermal conductivity and composition fluctuations for arbitrary space dimension. One approximate relation was heuristically obtained in our previous studies and verified using our extended simulation data on two-dimensional (2D) glass-forming systems. As a result, we provide the means to obtain, in any spatial dimension, all stress-correlation functions in terms of relaxation moduli and vice versa. The new approximate relations are tested using simulation data on 2D systems of polydisperse Lennard-Jones particles.

Risk evaluation, spatiotemporal evolution, and driving factors of provincial food safety in China

The cross-regional transfer of food safety risks has become more prominent, bringing new challenges to the food ecosystem and food safety regulation. This study develops an evaluation index system of food safety risks based on the pressure-state-response (PSR) framework. The food safety risk indices of 30 provinces in China from 2011 to 2020 are calculated by determining the index weights using the real-coded accelerated genetic algorithm–projection pursuit model (RAGA-PP). Furthermore, this study elaborates the spatiotemporal differences and correlations of provincial food safety risks, characterizes the spatial aggregation and spatiotemporal changes of risks, and investigates driving factors affecting the spatiotemporal evolution of risks. The following findings are established. First, the overall risk value of food safety in China in 2020 decreased by approximately 5% compared with year 2010. In 2020, the national food safety environment was relatively favorable and tended toward stability. The overall trend of food safety risks in China is decreasing, although socioeconomic development has increased the risks, and the corresponding social management measures have also reduced the risks. Second, food safety risks were unevenly distributed among provinces, showing significant regional differences. Specifically, generally higher risks were observed in eastern provinces versus other regions and in coastal provinces versus inland provinces. Concentrated and superimposed risks were observed in the Yangtze River Delta, a large urban agglomeration centered on Shanghai. Lastly, population density and urbanization level were the main drivers of the evolution of regional food safety risks; whereas policy instruments used by the government to manage risks played a limited role. The results show that establishing a cross-regional collaborative governance mechanism to manage food safety risks should constitute an important component of the regional integration strategy of China.

A deep generative model for selecting representative periods in renewable energy-integrated power systems

The extensive integration of renewable energies into power systems has led to a challenging and computationally demanding scenario for the system planning. This is due to the increased number of time series involved and the greater complexity of time series data integrated into power systems. In this paper, a new deep learning-based time aggregation method is proposed for selecting representative periods for renewable energy-integrated power system datasets where multiple variable energy resources are present. Relying on the capabilities of deep generative models, especially Generative Adversarial Network (GAN), the proposed method selects a representative period, including one or more representative days, obtained from multiple time series with spatio-temporal correlations among them. The proposed approach contains the Long Short-Term Memory (LSTM) Network in both generator and discriminator parts. Also, it includes a loss term, specifically designed for clustering tasks, in the minimax objective of the vanilla GAN loss function to enhance the clustering performance in the latent space. Furthermore, the learning rate of the proposed model, as the most important parameter of the learning algorithm, is adaptively fine-tuned during the training process, based on the training error, to enhance its learning performance. Two real-world test cases with various datasets and time series are used for data-based and model-based evaluations. Results obtained on these two real-world test cases confirm the superiority of the proposed model with respect to the state-of-the-art conventional and deep learning-based models, obtaining the performance improvement in the range of [38.55 %, 46.14 %] in terms of Mean Absolute Percentage Error (MAPE), in the range of [45.75 %, 70.16 %] in terms of Root Mean Squared Error (RMSE), and in the range of [65.81 %, 74.58 %] in terms of Mean Absolute Error (MAE). The proposed model can significantly enhance the tractability and scalability of the planning problem of renewable energy-integrated power systems, facilitating renewable energy integration into power systems which is a key issue for net zero transitioning of communities.

Scalar-on-function regression: Estimation and inference under complex survey designs.

Increasingly, large, nationally representative health and behavioral surveys conducted under a multistage stratified sampling scheme collect high dimensional data with correlation structured along some domain (eg, wearable sensor data measured continuously and correlated over time, imaging data with spatiotemporal correlation) with the goal of associating these data with health outcomes. Analysis of this sort requires novel methodologic work at the intersection of survey statistics and functional data analysis. Here, we address this crucial gap in the literature by proposing an estimation and inferential framework for generalizable scalar-on-function regression models for data collected under a complex survey design. We propose to: (1) estimate functional regression coefficients using weighted score equations; and (2) perform inference using novel functional balanced repeated replication and survey-weighted bootstrap for multistage survey designs. This is the first frequentist study to discuss the estimation of scalar-on-function regression models in the context of complex survey studies and to assess the validity of various inferential techniques based on re-sampling methods via a comprehensive simulation study. We implement our methods to predict mortality using diurnal activity profiles measured via wearable accelerometers using the National Health and Nutrition Examination Survey 2003-2006 data. The proposed computationally efficient methods are implemented in R software package surveySoFR.

Prenatal Diagnosis of Persistent Left Superior Vena Cava Using High-Definition Flow Render Mode and Spatiotemporal Image Correlation.

This study aimed to assess the use of two-dimensional (2D) ultrasound combined with high-definition flow (HD-flow) render mode and spatiotemporal image correlation (STIC) in diagnosing and classifying fetal persistent left superior vena cava (PLSVC). Overall, 114 cases of fetal PLSVC were diagnosed using 2D ultrasound combined with STIC, and 114 normal fetuses of the same gestational week were selected. These cases were retrospectively analyzed to evaluate the effectiveness of the diagnostic approach. All 114 PLSVC cases were diagnosed using 2D ultrasound combined with STIC. Although the diagnostic coincidence rate of PLSVC in the HD-flow combined with STIC was similar to that in the 2D ultrasound combined with HD-flow (96.8 vs 96.2%), 2D ultrasound with STIC enabled dynamic visualization of the PLSVC, furthering prenatal diagnosis. These cases were classified as type I PLSVC: 80 cases of type Ia, 29 cases of type Ib, and 5 cases of type Ic. Seventy isolated PLSVC cases (61.4%) were noted, whereas 44 cases (35.6%) were associated with concomitant structural abnormalities. Intracardiac structural malformations accounted for the highest proportion (n = 53, 58.89%), followed by single umbilical artery and facial/bodily abnormalities (n = 10, 11.11%). Combining HD-flow and STIC complements 2D ultrasound in diagnosing and classifying fetal PLSVC, demonstrating significant clinical relevance.

DDGformer: Direction- and distance-aware graph transformer for traffic flow prediction

Being an indispensable core technology of intelligent transportation systems, accurate traffic flow prediction contributes to improving people’s travel efficiency and safety. How to accurately model the dynamic spatio-temporal correlation in traffic data is a key challenge in traffic flow prediction. The models based on self-attention and Graph Neural Networks (GNN) have shown great potential in addressing this challenge. However, existing methods ignore the relative distance and direction of traffic flow series, and cannot accurately capture the real dynamic spatio-temporal correlations in traffic systems. To overcome this limitation, we propose a novel framework called Direction- and Distance-aware Graph Transformer (DDGformer). Specifically, it utilizes a self-attention module that simultaneously perceives direction and distance, enabling it to identify the relative position and direction of the original traffic flow series, and models long-term temporal correlations. Additionally, a dynamically enhanced adaptive graph convolution network is designed to capture dynamic traffic patterns in traffic systems. Extensive experimental results on four real-world traffic datasets show that our approach is overall more competitive and exhibits outstanding computational efficiency.

Non-Bloch theory for spatiotemporal photonic crystals assisted by continuum effective medium

As one indispensable type of nonreciprocal mechanism, a system with temporal modulations is intrinsically open in the physical sense and inevitably non-Hermitian, but the space and time degrees of freedom are nonseparable in a large variety of circumstances, which restrains the application of the non-Bloch band theory. Here, we investigate the spatially photonic crystals (PhCs) composed of spatiotemporal modulation materials (STMs) and homogeneous media, dubbed as the STMPhC, wherein the spatial and temporal modulations are deliberately designed to be correlated. To bypass the difficulty of the spatiotemporal correlation, we first employ the effective medium theory to account for the dispersion of fundamental bands under the influence of Floquet sidebands. Based on the continuum generalized Brillouin zone condition, we then analytically give the criteria for the existence of the non-Hermitian skin effect in the STM. Assisted by developing a numerical method that embeds the plane wave expansion in the transfer matrix, we establish the non-Bloch band theory for the low-frequency Floquet bands in the STMPhCs, in which the identification of the generalized Brillouin zone is central. We finally delve into the topological properties, including non-Bloch Zak phases and non-Bloch bulk-boundary correspondence. Our work validates the idea that the effective medium assists the non-Bloch band theory applied to the STMPhCs, which delivers a prescription to broaden the horizons of non-Bloch theory. Published by the American Physical Society 2024

MGGTSP‐CAT: Integrating Temporal Convolution and LSTM for Multi‐Scale Greenhouse Gas Time Series Prediction via Cross‐Attention Mechanism

AbstractGreenhouse gas‐induced global warming represents one of the most pressing environmental challenges currently faced, with methane, a primary constituent of greenhouse gases, exerting a more potent greenhouse effect than carbon dioxide. Precise prediction of methane concentrations is therefore of paramount importance. Owing to the complex interplay of factors affecting methane levels, existing forecasting methodologies often fall short in terms of accuracy and long‐term applicability. In response, this article leverages a multiscale time series to propose an attention‐based deep learning model. This approach synthesizes data across varied time spans, enhancing the extraction of periodic patterns and inter‐scale feature correlations. The model employs a hybrid framework and attention mechanisms to adaptively discern both short‐ and long‐term dependencies, as well as spatiotemporal correlations, capturing the intricate coupling of change mechanisms in the environment. Focusing on the Yanqing district of Beijing, a dataset integrating multiple features is developed and conduct forecasts 48 h in advance for both surface methane concentrations and the average molar fraction of total methane columns, comparing our model against several benchmarks. The model demonstrates superior performance, achieving an RMSE of 5.6123, MAE of 3.4959, and an R2 of 0.9148, which significantly exceeded the accuracy and generalizability of all baseline models.

Construction of effective reproduction number of infectious disease individuals based on spatiotemporal discriminant search model: take hand-foot-mouth disease as an example

ObjectiveIn order to facilitate the tracing of infectious diseases in a small area and to effectively carry out disease control and epidemiological investigations, this research proposes a novel spatiotemporal model to estimate effective reproduction number(Re)for infectious diseases, based on the fundamental concept of contact tracing.MethodsThis study utilizes the incidence of hand, foot, and mouth disease (HFMD) among children in Bishan District, Chongqing, China from 2015 to 2019. The study incorporates the epidemiological characteristics of HFMD and aims to construct a Spatiotemporal Correlation Discrimination of HFMD. Utilizing ARC ENGINE and C# programming for the creation of a spatio-temporal database dedicated to HFMD to facilitate data collection and analysis. The scientific validity of the proposed method was verified by comparing the effective reproduction number obtained by the traditional SEIR model.ResultsWe have ascertained the optimal search radius for the spatiotemporal search model to be 1.5 km. Upon analyzing the resulting Re values, which range from 1.14 to 4.75, we observe a skewed distribution pattern from 2015 to 2019. The median and quartile Re value recorded is 2.42 (1.98, 2.72). Except for 2018, the similarity coefficient r of the years 2015, 2016, 2017, and 2019 were all close to 1, and p <0.05 in the comparison of the two models, indicating that the Re values obtained by using the search model and the traditional SEIR model are correlated and closely related. The results exhibited similarity between the Re curves of both models and the epidemiological characteristics of HFMD. Finally, we illustrated the regional distribution of Re values obtained by the search model at various time intervals on Geographic Information System (GIS) maps which highlighted variations in the incidence of diseases across different communities, neighborhoods, and even smaller areas.ConclusionThe model comprehensively considers both temporal variation and spatial heterogeneity in disease transmission and accounts for each individual's distinct time of onset and spatial location. This proposed method differs significantly from existing mathematical models used for estimating Re in that it is founded on reasonable scientific assumptions and computer algorithms programming that take into account real-world spatiotemporal factors. It is particularly well-suited for estimating the Re of infectious diseases in relatively stable mobile populations within small geographical areas.

3D Correlation Imaging for Localized Phase Disturbance Mitigation

Correlation plenoptic imaging is a procedure to perform light-field imaging without spatial resolution loss, by measuring the second-order spatiotemporal correlations of light. We investigate the possibility of using correlation plenoptic imaging to mitigate the effect of a phase disturbance in the propagation from the object to the main lens. We assume that this detrimental effect, which can be due to a turbulent medium, is localized at a specific distance from the lens, and is slowly varying in time. The mitigation of turbulence effects has already fostered the development of both light-field imaging and correlation imaging procedures. Here, we aim to merge these aspects, proposing a correlation light-field imaging method to overcome the effects of slowly varying turbulence, without the loss of lateral resolution, typical of traditional plenoptic imaging devices.

Engineering-Guided Deep Learning of Melt-Pool Dynamics for Additive Manufacturing Quality Monitoring

Abstract Additive manufacturing (AM) fabricates three-dimensional parts via layer-by-layer deposition and solidification of materials. Due to the complexity of this process, advanced sensing is increasingly employed to facilitate system visibility, leading to a large amount of high-dimensional and complex-structured data. While deep learning brings attractive characteristics for data-driven process monitoring and quality prediction, it is currently limited in the ability to assimilate engineering knowledge and offer model interpretability for understanding process–quality relationships. In addition, due to spatiotemporal correlations in AM, a melt-pool anomaly observed during fabrication is not always indicative of abnormal quality characteristics. There is a pressing need to go beyond pointwise analysis of melt pools and consider spatiotemporal effects for quality analysis. In this paper, we propose a novel feature learning framework guided by engineering knowledge for AM quality monitoring. First, engineering knowledge is integrated with deep learning to delineate various sources of process variations and extract melt-pool features that reflect quality-related relationships. Second, a 3D neighborhood model is designed to characterize spatiotemporal variations of melt pools based on their domain-informed features. The resulting 3D neighborhood profiles enable us to go beyond pointwise analysis of melt pools for capturing process–quality relationships. Finally, we built a regression model to predict internal density variations using 3D neighborhood profiles. Our experiments demonstrate that the proposed framework significantly outperforms traditional hand-crafted method and black-box learning in both the ability to provide quality-related features and predict internal density variations.

Capturing and interpreting wildfire spread dynamics: attention-based spatiotemporal models using ConvLSTM networks

Predicting the trajectory of geographical events, such as wildfire spread, presents a formidable task due to the dynamic associations among influential biophysical factors. Geo-events like wildfires frequently display short and long-range spatial and temporal correlations. Short-range effects are the direct contact and near-contact spread of the fire front. Long-range effects are represented by processes such as spotting, where firebrands carried by the wind ignite fires distant from the flaming front, altering the shape and speed of an advancing fire front. This study addresses these modeling challenges by clearly defining and analyzing the scale-dependent spatiotemporal dynamics that influence wildfire spread, focusing on the interplay between biophysical factors and fire behavior. We propose two unique attention-based spatiotemporal models using Convolutional Long Short-Term Memory (ConvLSTM) networks. These models are designed to learn and capture a range of local to global and short and long-range spatiotemporal correlations. The proposed models were tested on two datasets: a high-resolution wildfire spread dataset produced with a semi-empirical percolation model and a satellite observed wildfire spread data in California 2012–2021. Results indicate that attention-based models accurately predict fire front movements that are consistent with known wildfire spread-biophysical dynamics. Our research suggests there is considerable potential for attention mechanisms to capture the spatiotemporal behavior of wildfire spread, with model transferability, that can guide rapid deployment of wildfire management operations. We also highlight directions for future studies that focus on how the self-attention mechanism could enhance model performance for a range of geospatial applications.

Spatiotemporal correlation enhanced real-time 4D-CBCT imaging using convolutional LSTM networks.

To enhance the accuracy of real-time four-dimensional cone beam CT (4D-CBCT) imaging by incorporating spatiotemporal correlation from the sequential projection image into the single projection-based 4D-CBCT estimation process. We first derived 4D deformation vector fields (DVFs) from patient 4D-CT. Principal component analysis (PCA) was then employed to extract distinctive feature labels for each DVF, focusing on the first three PCA coefficients. To simulate a wide range of respiratory motion, we expanded the motion amplitude and used random sampling to generate approximately 900 sets of PCA labels. These labels were used to produce 900 simulated 4D-DVFs, which in turn deformed the 0% phase 4D-CT to obtain 900 CBCT volumes with continuous motion amplitudes. Following this, the forward projection was performed at one angle to get all of the digital reconstructed radiographs (DRRs). These DRRs and the PCA labels were used as the training data set. To capture the spatiotemporal correlation in the projections, we propose to use the convolutional LSTM (ConvLSTM) network for PCA coefficient estimation. For network testing, when several online CBCT projections (with different motion amplitudes that cover the full respiration range) are acquired and sent into the network, the corresponding 4D-PCA coefficients will be obtained and finally lead to a full online 4D-CBCT prediction. A phantom experiment is first performed with the XCAT phantom; then, a pilot clinical evaluation is further conducted. Results on the XCAT phantom and the patient data show that the proposed approach outperformed other networks in terms of visual inspection and quantitative metrics. For the XCAT phantom experiment, ConvLSTM achieves the highest quantification accuracy with MAPE(Mean Absolute Percentage Error), PSNR (Peak Signal-to-Noise Ratio), and RMSE(Root Mean Squared Error) of 0.0459, 64.6742, and 0.0011, respectively. For the patient pilot clinical experiment, ConvLSTM also achieves the best quantification accuracy with that of 0.0934, 63.7294, and 0.0019, respectively. The quantification evaluation labels that we used are 1) the Mean Absolute Error (MAE), 2) the Normalized Cross Correlation (NCC), 3)the Structural Similarity Index Measurement(SSIM), 4)the Peak Signal-to-Noise Ratio (PSNR), 5)the Root Mean Squared Error(RMSE), and 6) the Absolute Percentage Error (MAPE). The spatiotemporal correlation-based respiration motion modeling supplied a potential solution for accurate real-time 4D-CBCT reconstruction.

Influence of sea ice on ship routes and speed along the Arctic Northeast Passage

The accelerated melting of sea ice and the growing influx of ships in the Arctic pose significant challenges to navigational safety and the resilient development of the Arctic routes. This study aims to explore the influence of sea ice on ship routes and speed by analyzing the spatiotemporal correlation of Sea Ice Concentration (SIC), Sea Ice Thickness (SIT), Sea Ice Volume (SIV), and Automatic Identification System (AIS) data along the Northeast Passage (NEP) in 2022. Our findings indicate that during the navigation window period (July–October), 115 ships traversed the NEP. It was observed that ships preferred navigating at speeds ranging from 6kn to 14kn when the SIC was below 10% and the SIT was less than 0.1m. Compared with cargo ships and tankers, the trajectory and speed distributions of other ships show greater differences under the influence of sea ice. Furthermore, our results suggest that the SIV, compared to the SIC and SIT, could better reflect the influence of sea ice on ship navigation. These insights are crucial for developing management strategies in Arctic routes.

Integrating query data for enhanced traffic forecasting: A Spatio-Temporal Graph Attention Convolution Network approach with delay modeling

Accurate prediction of road traffic conditions is essential for the effectiveness of Intelligent Transportation Systems (ITS) and the advancement of smart cities. While existing methodologies mainly focus on the complex structures of road networks and consider factors like weather, holidays, and accidents, the impact of social events is often overlooked. Queries, representing user requests in navigation applications, offer valuable insights into the popularity of social events and their effects on the road network. In this paper, we introduce the Query-aware Spatio-Temporal Graph Attention Convolution Network (QA-STGACN), which utilizes query data to model the impacts of events and forecast traffic conditions. Initially, spatial and temporal blocks are developed to extract spatial dependencies among road segments and capture traffic dynamics. A dynamic query-aware graph constructor then adaptively learns and retains periodic interactions between road segments and queries. Furthermore, three novel query-aware blocks – convolution, attention, and pattern transformation – are designed to uncover hidden dependencies between traffic conditions and query counts during events, as well as to model the delayed impacts of these counts on traffic. Ultimately, a gated fusion mechanism integrates the spatio-temporal correlations in road networks with the representations derived from queries. Comprehensive experiments on six datasets validate the superiority of QA-STGACN over robust baselines.

Novel particulate matter (PM2.5) forecasting method based on deep learning with suitable spatiotemporal correlation analysis

Since air pollution caused by PM 2.5 (particulate matter with an aerodynamic diameter of ≤2.5 μm) is a serious threat to human health, the accurate forecasting of PM 2.5 concentration in metropolitan areas is one of the prior conditions to reduce and eliminate the harmful impacts on human beings produced by PM2.5. In this study, we analyzed the spatiotemporal correlations between target and observation parameters relevant to air pollution forecasting and proposed a convolutional neural network (CNN) and long short-term memory (LSTM) model (also called PM predictor) for next day's daily average PM 2.5 concentration forecasting in Beijing. The proposed spatiotemporal correlations were analyzed for efficient estimation of mutual information, not only if the degrees of variations between the two spaces under consideration are similar, but also if the degrees of variations are significantly different, thereby generating a spatiotemporal feature vector. CNN provided an efficient extraction of inherent features for latent air quality and meteorological input data relevant to PM 2.5, and LSTM delivered the historical information in the time series data, thus a novel PM predictor with remarkably improved performance was constructed, compared with multi-layer perceptron (MLP) and LSTM model in overall forecasting. The air quality and meteorological data from the monitoring stations in Beijing and four surrounding cities from January 1, 2015 to December 31, 2017 were adopted as dataset. The forecasting results suggest that the proposed PM predictor is superior to other models in overall forecasting, while LSTM is better than PM predictor with slight difference in seasonal forecasting.

Method for constructing typical operation scenarios in power systems considering spatiotemporal correlation of renewable energy generation

Abstract The proposed article focuses on developing a methodology for constructing typical operational scenarios in power systems considering the spatiotemporal characteristics of new energy output. The integration of new energy sources on a large scale into power systems can lead to issues such as line congestion and supply-demand imbalances. To support the planning and scheduling of future high-capacity power systems, this paper introduces a method that accounts for the spatiotemporal characteristics of new energy outputs. The method begins by proposing a spatiotemporal fusion-based multi-head attention model for generating new energy output scenarios. This model employs multi-head attention mechanisms to extract the spatiotemporal characteristics of new energy outputs, eliminating redundancy in the feature data. The processed spatiotemporal feature information is then transferred to a gated feature fusion model for integration, using an encoder structure based on Long Short-Term Memory (LSTM) to generate new energy output scenarios. The paper, by means of an advanced Fuzzy C-Means (FCM) clustering algorithm, identifies typical scenarios of net energy output. Subsequently, the proposed method’s efficacy and precision are confirmed through simulation analysis with real data from a certain area.

F-norm based low-power motion recognition on wearable devices in the presence of outlier motions

Motion recognition refers to the intelligent recognition of human motion using data collected from wearable sensors, which exceedingly has gained significant interest from both academic and industrial fields. However, temporary-sudden activities caused by accidental behavior pose a major challenge to motion recognition and have been largely overlooked in existing works. To address this problem, the multi-dimensional time series of motion data is modeled as a Time-Frequency (TF) tensor, and the original challenge is transformed into a problem of outlier-corrupted tensor pattern recognition, where transient sudden activity data are considered as outliers. Since the TF tensor can capture the latent spatio-temporal correlations of the motion data, the tensor MPCA is used to derive the principle spatio-temporal pattern of the motion data. However, traditional MPCA uses the squared F-norm as the projection distance measure, which makes it sensitive to the presence of outlier motion data. Therefore, in the proposed outlier-robust MPCA scheme, the F-norm with the desirable geometric properties is used as the distance measure to simultaneously mitigate the interference of outlier motion data while preserving rotational invariance. Moreover, to reduce the complexity of outlier-robust motion recognition, we impose the proposed outlier-robust MPCA scheme on the traditional MPCANet which is a low-complexity deep learning network. The experimental results show that our proposed outlier-robust MPCANet can simultaneously improve motion recognition performance and reduce the complexity, especially in practical scenarios where the real-time data is corrupted by temporary-sudden activities.