Xiaoyu He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Xiaoyu He has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
Xiulin Geng
7 shared publications- Xiulin Geng
School of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China
6 shared publications Suixiang Shi
5 shared publications- Mingfeng Jiang
School of Computer Science and Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China
4 shared publications Lingyu Xu
3 shared publications- Lingyu Xu
Department of Computer Engineering and Science, Shanghai University, Shanghai, China
3 shared publications
Latest Publications
- Open Access
- 1 October 2026
- Published by MDPI AG in Technologies
- Vol. 14  (10) , 619
ST-GODE-UQ: A Spatiotemporal Graph Neural Ordinary Differential Equation Framework with Uncertainty Quantification for Sudden Cardiac Death Prediction
Sudden cardiac death (SCD) remains difficult to anticipate from routine risk stratification, and the reliability of electrocardiographic (ECG) predictions remains underreported. In this study, SCD prediction is investigated retrospectively using ECG segments sampled at predefined lead times before ventricular fibrillation (VF) onset. We propose ST-GODE-UQ, which represents 5-s ECG segments as frequency-band spectral-similarity graphs, performs continuous-depth latent learning with a graph neural ordinary differential equation (ODE), and combines heteroscedastic logit variance with Monte Carlo (MC) dropout. Evaluation used subject-wise grouped five-fold outer cross-validation in two complementary settings: a cross-database comparison of ECG segments recorded 30–35 min before VF with healthy normal-sinus-rhythm recordings, and a within-cohort temporal-control comparison of 30–35 and 80–85 min pre-VF segments from the same SCD subjects. ST-GODE-UQ achieved mean accuracies of 94.14% and 93.00% and mean ROC-AUCs of 94.78% and 93.60% in the cross-database and temporal-control evaluations, respectively. Its pooled out-of-fold ROC-AUC was numerically higher than that of the graph neural network (GNN) + Transformer baseline, although the paired difference was not statistically significant. Misclassified predictions showed higher median uncertainty across all three reliability measures, and controlled Gaussian-noise and baseline-wander perturbations progressively increased the aleatoric estimate. These findings support the feasibility of uncertainty-aware SCD prediction from predefined pre-VF ECG windows under subject-independent evaluation and motivate further validation in prospective and clinically heterogeneous settings.
- 1 October 2026
- Published by MDPI AG in Technologies
- Vol. 14  (10) , 619
Open AccessST-GODE-UQ: A Spatiotemporal Graph Neural Ordinary Differential Equation Framework with Uncertainty Quantification for Sudden Cardiac Death Prediction
Sudden cardiac death (SCD) remains difficult to anticipate from routine risk stratification, and the reliability of electrocardiographic (ECG) predictions remains underreported. In this study, SCD prediction is investigated retrospectively using ECG segments sampled at predefined lead times before ventricular fibrillation (VF) onset. We propose ST-GODE-UQ, which represents 5-s ECG segments as frequency-band spectral-similarity graphs, performs continuous-depth latent learning with a graph neural ordinary differential equation (ODE), and combines heteroscedastic logit variance with Monte Carlo (MC) dropout. Evaluation used subject-wise grouped five-fold outer cross-validation in two complementary settings: a cross-database comparison of ECG segments recorded 30–35 min before VF with healthy normal-sinus-rhythm recordings, and a within-cohort temporal-control comparison of 30–35 and 80–85 min pre-VF segments from the same SCD subjects. ST-GODE-UQ achieved mean accuracies of 94.14% and 93.00% and mean ROC-AUCs of 94.78% and 93.60% in the cross-database and temporal-control evaluations, respectively. Its pooled out-of-fold ROC-AUC was numerically higher than that of the graph neural network (GNN) + Transformer baseline, although the paired difference was not statistically significant. Misclassified predictions showed higher median uncertainty across all three reliability measures, and controlled Gaussian-noise and baseline-wander perturbations progressively increased the aleatoric estimate. These findings support the feasibility of uncertainty-aware SCD prediction from predefined pre-VF ECG windows under subject-independent evaluation and motivate further validation in prospective and clinically heterogeneous settings.
- Open Access
- 20 August 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18  (16) , 2828
Multi-Scale Spatiotemporal Graph ODE Networks for Marine Chlorophyll-a Prediction
Chlorophyll-a concentration is a key indicator reflecting the growth status of phytoplankton, and its accurate prediction is of great significance for assessing the degree of water eutrophication. Although existing approaches have achieved good performance, they generally pay insufficient attention to multi-scale spatial information and show limitations in characterizing the continuous spatiotemporal dynamics. To address these issues, this paper proposes a multi-scale spatiotemporal graph ODE network (MGODE) for ocean chlorophyll-a prediction. The MGODE adopts a dual-layer structure, simultaneously processing chlorophyll-a concentration data at both the region level and node level to capture multi-scale spatial features, and it enables effective interaction of cross-scale features through dynamic transmission coefficients and a gated fusion mechanism. Meanwhile, the MGODE employs a dual-ODE architecture at both the node and region levels, utilizing spatiotemporal ODE blocks to continuously and deeply capture features, thereby simulating the continuous spatiotemporal dynamic evolution of chlorophyll-a. Experiments on real-world datasets from the Bohai Sea and South China Sea show that the proposed MGODE model achieves higher prediction accuracy than several current state-of-the-art models. Compared with the best baseline, the MGODE achieves reductions of 2.78% in MAE and 1.07% in RMSE on the Bohai Sea dataset and reductions of 1.19% in MAE and 1.38% in RMSE on the South China Sea dataset. These results demonstrate the potential of the MGODE to support marine chlorophyll-a forecasting and marine ecological monitoring.
- 20 August 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18  (16) , 2828
Open AccessMulti-Scale Spatiotemporal Graph ODE Networks for Marine Chlorophyll-a Prediction
Chlorophyll-a concentration is a key indicator reflecting the growth status of phytoplankton, and its accurate prediction is of great significance for assessing the degree of water eutrophication. Although existing approaches have achieved good performance, they generally pay insufficient attention to multi-scale spatial information and show limitations in characterizing the continuous spatiotemporal dynamics. To address these issues, this paper proposes a multi-scale spatiotemporal graph ODE network (MGODE) for ocean chlorophyll-a prediction. The MGODE adopts a dual-layer structure, simultaneously processing chlorophyll-a concentration data at both the region level and node level to capture multi-scale spatial features, and it enables effective interaction of cross-scale features through dynamic transmission coefficients and a gated fusion mechanism. Meanwhile, the MGODE employs a dual-ODE architecture at both the node and region levels, utilizing spatiotemporal ODE blocks to continuously and deeply capture features, thereby simulating the continuous spatiotemporal dynamic evolution of chlorophyll-a. Experiments on real-world datasets from the Bohai Sea and South China Sea show that the proposed MGODE model achieves higher prediction accuracy than several current state-of-the-art models. Compared with the best baseline, the MGODE achieves reductions of 2.78% in MAE and 1.07% in RMSE on the Bohai Sea dataset and reductions of 1.19% in MAE and 1.38% in RMSE on the South China Sea dataset. These results demonstrate the potential of the MGODE to support marine chlorophyll-a forecasting and marine ecological monitoring.
- Open Access
Heterogeneity-Aware Multi-Step Chlorophyll-a Forecasting for Marine Water Quality Monitoring Using a Multi-Scale Spatio-Temporal Mixture-of-Experts Network
Chlorophyll-a concentration (Chl-a) is a key indicator of marine water quality, phytoplankton biomass, and aquatic ecosystem status. Multi-step forecasting is challenging because Chl-a dynamics exhibit regional heterogeneity, multi-scale variability, and complex spatial dependence. Existing models often optimize domain-averaged errors, which can mask unstable node-level predictions at high-variability nodes. We propose MS-STMoE, a heterogeneity-aware multi-scale spatio-temporal mixture-of-experts framework. It uses a Haversine-distance-based K-nearest-neighbor graph, gated multi-scale temporal convolutions with seasonal encoding to model short-term and periodic variations, and node-level sparse top-k routing that assigns differentiated expert pathways to nodes with distinct dynamics based on recent-state, temporal-mean, temporal-change, and node-prior features. Using 30-day histories to forecast the next 15 days, experiments on 300 Bohai Sea and 265 South China Sea nodes show that MS-STMoE achieves the lowest average MAE and RMSE among six recent spatio-temporal baselines. Compared with the best baseline, MAE and RMSE decrease by 5.6% and 2.7%, respectively, in the Bohai Sea, and by 11.0% and 5.0%, respectively, in the South China Sea. Step-wise and node-wise analyses indicate improved medium-to-late-horizon accuracy and modest, region-dependent reductions in high-error node tails, supporting more reliable region-aware short- to medium-term water quality monitoring.
Open AccessHeterogeneity-Aware Multi-Step Chlorophyll-a Forecasting for Marine Water Quality Monitoring Using a Multi-Scale Spatio-Temporal Mixture-of-Experts Network
Chlorophyll-a concentration (Chl-a) is a key indicator of marine water quality, phytoplankton biomass, and aquatic ecosystem status. Multi-step forecasting is challenging because Chl-a dynamics exhibit regional heterogeneity, multi-scale variability, and complex spatial dependence. Existing models often optimize domain-averaged errors, which can mask unstable node-level predictions at high-variability nodes. We propose MS-STMoE, a heterogeneity-aware multi-scale spatio-temporal mixture-of-experts framework. It uses a Haversine-distance-based K-nearest-neighbor graph, gated multi-scale temporal convolutions with seasonal encoding to model short-term and periodic variations, and node-level sparse top-k routing that assigns differentiated expert pathways to nodes with distinct dynamics based on recent-state, temporal-mean, temporal-change, and node-prior features. Using 30-day histories to forecast the next 15 days, experiments on 300 Bohai Sea and 265 South China Sea nodes show that MS-STMoE achieves the lowest average MAE and RMSE among six recent spatio-temporal baselines. Compared with the best baseline, MAE and RMSE decrease by 5.6% and 2.7%, respectively, in the Bohai Sea, and by 11.0% and 5.0%, respectively, in the South China Sea. Step-wise and node-wise analyses indicate improved medium-to-late-horizon accuracy and modest, region-dependent reductions in high-error node tails, supporting more reliable region-aware short- to medium-term water quality monitoring.
- 31 July 2026
- Published by Elsevier in Marine Pollution Bulletin
- Vol. 233  (Pt 1) , 120196
Mask-aware biased graph learning for marine chlorophyll-a spatiotemporal forecasting under high missing rates
- 31 July 2026
- Published by Elsevier in Marine Pollution Bulletin
- Vol. 233  (Pt 1) , 120196
Mask-aware biased graph learning for marine chlorophyll-a spatiotemporal forecasting under high missing rates
- Open Access
- 15 July 2026
- Published by MDPI AG in Journal of Marine Science and Engineering
- Vol. 14  (14) , 1301
Uncertainty-Quantified Dynamic Graph Ordinary Differential Equation Network for Marine Chlorophyll-a Concentration Forecasting
Marine chlorophyll-a (Chl-a) concentration is a key proxy for phytoplankton biomass and an important indicator for eutrophication assessment, ecological monitoring, and harmful algal bloom early warning. Accurate forecasting of Chl-a concentration therefore has substantial scientific and practical value for marine environmental management. Nevertheless, existing deep models still have difficulty coupling relatively stable large-scale spatial structure with locally time-varying interactions, and many of them do not provide reliable uncertainty estimates, which limits their use in harmful algal bloom early warning and high-chlorophyll event assessment. To address these issues, this study proposes an uncertainty-quantified dynamic graph ordinary differential equation network, termed UQDGODE, for marine Chl-a concentration forecasting. The model combines a multivariate diffusion graph convolutional branch for stable spatial diffusion modeling with a dynamic graph ordinary differential equation branch for continuously evolving local interactions. A gating mechanism fuses the two branches, and a multivariate probabilistic prediction module outputs predictive means and covariance information for uncertainty quantification. Experiments on the Bohai Sea and South China Sea datasets show that the UQDGODE performs better in both point forecasting accuracy and probabilistic forecasting quality and produces more informative predictive distributions and more adaptive prediction intervals for marine ecological alerting.
- 15 July 2026
- Published by MDPI AG in Journal of Marine Science and Engineering
- Vol. 14  (14) , 1301
Open AccessUncertainty-Quantified Dynamic Graph Ordinary Differential Equation Network for Marine Chlorophyll-a Concentration Forecasting
Marine chlorophyll-a (Chl-a) concentration is a key proxy for phytoplankton biomass and an important indicator for eutrophication assessment, ecological monitoring, and harmful algal bloom early warning. Accurate forecasting of Chl-a concentration therefore has substantial scientific and practical value for marine environmental management. Nevertheless, existing deep models still have difficulty coupling relatively stable large-scale spatial structure with locally time-varying interactions, and many of them do not provide reliable uncertainty estimates, which limits their use in harmful algal bloom early warning and high-chlorophyll event assessment. To address these issues, this study proposes an uncertainty-quantified dynamic graph ordinary differential equation network, termed UQDGODE, for marine Chl-a concentration forecasting. The model combines a multivariate diffusion graph convolutional branch for stable spatial diffusion modeling with a dynamic graph ordinary differential equation branch for continuously evolving local interactions. A gating mechanism fuses the two branches, and a multivariate probabilistic prediction module outputs predictive means and covariance information for uncertainty quantification. Experiments on the Bohai Sea and South China Sea datasets show that the UQDGODE performs better in both point forecasting accuracy and probabilistic forecasting quality and produces more informative predictive distributions and more adaptive prediction intervals for marine ecological alerting.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23  (1545598X) , 1-5
Dynamic Graph Contrastive Learning Neural Network for Sea Surface Temperature Prediction
Accurate daily mean sea surface temperature (SST) forecasting not only contributes to a deeper understanding of natural processes but also enhances our knowledge of global climate change and marine ecosystems. However, existing methods are often susceptible to noise and overlook dynamic spatial correlations, making it difficult for them to capture complex spatiotemporal dependencies. To address these issues, we propose a Dynamic Graph Contrastive Learning Network (DGCLN) for daily mean SST spatiotemporal forecasting. Specifically, we improve robustness to noise through topological data augmentation and generate both local and global augmented graphs using predefined adjacency matrices and self-optimized matrices. In addition, a dual-branch shared network that integrates multi-head self-attention with dynamic graphs is employed to fully exploit the dynamic spatiotemporal dependencies in SST sequences. Furthermore, a spatiotemporal contrastive learning auxiliary task is introduced to alleviate the problem of sparse supervision and extract critical spatiotemporal information. Experiments conducted on the Bohai Sea and South China Sea datasets demonstrate that DGCLN significantly outperforms existing state-of-the-art methods. Specifically, on the South China Sea dataset, the model achieves an MAE of 0.1506 for 1-day prediction, while outperforming the best-performing baseline by an average of 3.3% across all evaluated horizons.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23  (1545598X) , 1-5
Dynamic Graph Contrastive Learning Neural Network for Sea Surface Temperature Prediction
Accurate daily mean sea surface temperature (SST) forecasting not only contributes to a deeper understanding of natural processes but also enhances our knowledge of global climate change and marine ecosystems. However, existing methods are often susceptible to noise and overlook dynamic spatial correlations, making it difficult for them to capture complex spatiotemporal dependencies. To address these issues, we propose a Dynamic Graph Contrastive Learning Network (DGCLN) for daily mean SST spatiotemporal forecasting. Specifically, we improve robustness to noise through topological data augmentation and generate both local and global augmented graphs using predefined adjacency matrices and self-optimized matrices. In addition, a dual-branch shared network that integrates multi-head self-attention with dynamic graphs is employed to fully exploit the dynamic spatiotemporal dependencies in SST sequences. Furthermore, a spatiotemporal contrastive learning auxiliary task is introduced to alleviate the problem of sparse supervision and extract critical spatiotemporal information. Experiments conducted on the Bohai Sea and South China Sea datasets demonstrate that DGCLN significantly outperforms existing state-of-the-art methods. Specifically, on the South China Sea dataset, the model achieves an MAE of 0.1506 for 1-day prediction, while outperforming the best-performing baseline by an average of 3.3% across all evaluated horizons.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23  (1545598X) , 1-5
Multiscale Dynamic Graph Network for Forecasting of Wind Speeds in Offshore Regions
Accurate prediction of wind speed in offshore areas is critical to maintaining the reliability and stability of offshore wind power. However, existing graph neural network methods focus only on a single time scale and ignore potential temporal patterns. Moreover, they rely on predefined graph structures and cannot capture dynamic spatial correlations between ocean locations, which affects the representation of offshore wind speed patterns. Therefore, we introduce the Multi-Scale Dynamic Graph Network (MSDGN) to enhance wind speed forecasting accuracy in offshore regions. MSDGN introduces a network with a multi-scale pyramid structure to capture potential time-based dependencies in raw data across various scales, and infers spatial relationships between oceanic locations at specific scales through a self-adapting learning graph network. Subsequently, the proposed temporal graph network with multi-scale concurrently captures both the time-based dependencies in sequence data and the spatial relationships among oceanic locations. Finally, the model integrates a scale-wise fusion module to facilitate cooperation among various scales, further enhancing prediction accuracy. Extensive experiments on Chinese sea data show our model outperforms state-of-the-art methods in multi-node wind speed prediction.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23  (1545598X) , 1-5
Multiscale Dynamic Graph Network for Forecasting of Wind Speeds in Offshore Regions
Accurate prediction of wind speed in offshore areas is critical to maintaining the reliability and stability of offshore wind power. However, existing graph neural network methods focus only on a single time scale and ignore potential temporal patterns. Moreover, they rely on predefined graph structures and cannot capture dynamic spatial correlations between ocean locations, which affects the representation of offshore wind speed patterns. Therefore, we introduce the Multi-Scale Dynamic Graph Network (MSDGN) to enhance wind speed forecasting accuracy in offshore regions. MSDGN introduces a network with a multi-scale pyramid structure to capture potential time-based dependencies in raw data across various scales, and infers spatial relationships between oceanic locations at specific scales through a self-adapting learning graph network. Subsequently, the proposed temporal graph network with multi-scale concurrently captures both the time-based dependencies in sequence data and the spatial relationships among oceanic locations. Finally, the model integrates a scale-wise fusion module to facilitate cooperation among various scales, further enhancing prediction accuracy. Extensive experiments on Chinese sea data show our model outperforms state-of-the-art methods in multi-node wind speed prediction.
- 20 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 187
A long- and short-term feature fusion network for wind speed forecasting
- 20 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 187
A long- and short-term feature fusion network for wind speed forecasting
- 13 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
FastTWE: Making TWE-based time-series similarity search faster via random convolutional kernels
- 13 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
FastTWE: Making TWE-based time-series similarity search faster via random convolutional kernels
- Open Access
Improving Early Prediction of Sudden Cardiac Death Risk via Hierarchical Feature Fusion
Sudden cardiac death (SCD) is a leading cause of mortality worldwide, with arrhythmia serving as a major precursor. Early and accurate prediction of SCD using non-invasive electrocardiogram (ECG) signals remains a critical clinical challenge, particularly due to the inherent asymmetric and non-stationary characteristics of ECG signals, which complicate feature extraction and model generalization. In this study, we propose a novel SCD prediction framework based on hierarchical feature fusion, designed to capture both non-stationary and asymmetrical patterns in ECG data across six distinct time intervals preceding the onset of ventricular fibrillation (VF). First, linear features are extracted from ECG signals using waveform detection methods; nonlinear features are derived from RR interval sequences via second-order detrended fluctuation analysis (DFA2); and multi-scale deep learning features are captured using a Temporal Convolutional Network-based sequence-to-vector (TCN-Seq2vec) model. These multi-scale deep learning features, along with linear and nonlinear features, are then hierarchically fused. Finally, two fully connected layers are employed as a classifier to estimate the probability of SCD occurrence. The proposed method is evaluated under an inter-patient paradigm using the Sudden Cardiac Death Holter (SCDH) Database and the Normal Sinus Rhythm (NSR) Database. This method achieves average prediction accuracies of 97.48% and 98.8% for the 60 and 30 min periods preceding SCD, respectively. The findings suggest that integrating traditional and deep learning features effectively enhances the discriminability of abnormal samples, thereby improving SCD prediction accuracy. Ablation studies confirm that multi-feature fusion significantly improves performance compared to single-modality models, and validation on the Creighton University Ventricular Tachyarrhythmia Database (CUDB) demonstrates strong generalization capability. This approach offers a reliable, long-horizon early warning tool for clinical SCD risk assessment.
Open AccessImproving Early Prediction of Sudden Cardiac Death Risk via Hierarchical Feature Fusion
Sudden cardiac death (SCD) is a leading cause of mortality worldwide, with arrhythmia serving as a major precursor. Early and accurate prediction of SCD using non-invasive electrocardiogram (ECG) signals remains a critical clinical challenge, particularly due to the inherent asymmetric and non-stationary characteristics of ECG signals, which complicate feature extraction and model generalization. In this study, we propose a novel SCD prediction framework based on hierarchical feature fusion, designed to capture both non-stationary and asymmetrical patterns in ECG data across six distinct time intervals preceding the onset of ventricular fibrillation (VF). First, linear features are extracted from ECG signals using waveform detection methods; nonlinear features are derived from RR interval sequences via second-order detrended fluctuation analysis (DFA2); and multi-scale deep learning features are captured using a Temporal Convolutional Network-based sequence-to-vector (TCN-Seq2vec) model. These multi-scale deep learning features, along with linear and nonlinear features, are then hierarchically fused. Finally, two fully connected layers are employed as a classifier to estimate the probability of SCD occurrence. The proposed method is evaluated under an inter-patient paradigm using the Sudden Cardiac Death Holter (SCDH) Database and the Normal Sinus Rhythm (NSR) Database. This method achieves average prediction accuracies of 97.48% and 98.8% for the 60 and 30 min periods preceding SCD, respectively. The findings suggest that integrating traditional and deep learning features effectively enhances the discriminability of abnormal samples, thereby improving SCD prediction accuracy. Ablation studies confirm that multi-feature fusion significantly improves performance compared to single-modality models, and validation on the Creighton University Ventricular Tachyarrhythmia Database (CUDB) demonstrates strong generalization capability. This approach offers a reliable, long-horizon early warning tool for clinical SCD risk assessment.