Enhancing Multivariate Time Series Forecasting: A Novel Approach with Mallows Model Averaging and Graph Neural Networks DOI Creative Commons
Haili Zhang, Jiawei Wang,

Zhaobo Liu

et al.

Journal of Systems Science and Complexity, Journal Year: 2024, Volume and Issue: unknown

Published: July 27, 2024

Multivariate time series forecasting holds substantial practical significance, facilitates precise predictions, and informs decision-making. The complexity of nonlinear relationships the presence higher-order features in multivariate data have sparked a burgeoning interest leveraging deep learning approaches for such tasks. Existing methods often use pre-scaled neural networks, whose reliability generalization can pose challenge. In this study, authors propose an instance-wise graph-based Mallows model averaging (IGMMA) framework prediction. incorporates module into network, where extracted are utilized as inputs candidate linear models. These models combined with weights to create new layer, forming novel graph network model. Moreover, loss function is modified based on criterion, penalties imposed parameters separately. proposed method predict multicommodity futures prices, empirical results show that IGMMA has superior predictive accuracy even when small networks used. This indicates significantly reduces required training, which enables training multiple alternative large

Language: Английский

Tropical Cyclone Wind Field Reconstruction for Hazard Estimation via Bayesian Hierarchical Modeling With Neural Network DOI Creative Commons
Chi Yang, Jing Xu

Earth and Space Science, Journal Year: 2024, Volume and Issue: 11(12)

Published: Dec. 1, 2024

Abstract Tropical cyclones (TCs) are one of the biggest threats to life and property around world. Accurate estimation TC wind hazard requires catastrophic TCs having a very long return period spanning up thousands years. Since reliable data available only for recently decades, stochastic modeling simulation turned out be an effective approach achieve more stable estimates. In common practice, hundreds synthetic generated first, then fields reconstructed along tracks estimation. A Bayesian hierarchical reconstruction field is proposed. modified Rankine vortex adopted as model, which four free parameters modeled simultaneously through multi‐output neural network latent process field. The finally represented, spatially temporally, by set weights, model averaging technique used parameter reconstruction, based on ensemble maximum posteriori estimates weights. Together with previously proposed algorithm simulation, two‐stage scheme has been formed, best‐track thus highly consistent. Application this offshore waters in western North Pacific basin shows inspiring performance great flexibility various purposes

Language: Английский

Citations

1
Prediction of Total Phosphorus Concentration in Canals by GAT-Informer Model Based on Spatiotemporal Correlations DOI Open Access
Juan Huan, Xincheng Li,

Jialong Yuan

et al.

Water, Journal Year: 2024, Volume and Issue: 17(1), P. 12 - 12

Published: Dec. 24, 2024

The accurate prediction of total phosphorus (TP) is crucial for the early detection water quality eutrophication. However, predicting TP concentrations among canal sites challenging due to their complex spatiotemporal dependencies. To address this issue, study proposes a GAT-Informer method based on correlations predict in Beijing–Hangzhou Grand Canal Basin Changzhou City. begins by creating feature sequences each site time lag relationship concentration between sites. It then constructs graph data combining real river distance and correlation sequences. Next, spatial features are extracted fusing node using attention (GAT) module. employs Informer network, which uses sparse mechanism extract temporal efficiently simulating model was evaluated R2, MAE, RMSE, with experimental results yielding values 0.9619, 0.1489%, 0.1999%, respectively. exhibits enhanced robustness superior predictive accuracy comparison traditional models.

Language: Английский

Citations

1
Prediction of water levels in large reservoirs base on optimization of deep learning algorithms DOI
Haoran Li, Lili Zhang,

Yunsheng Yao

et al.

Earth Science Informatics, Journal Year: 2024, Volume and Issue: 18(1)

Published: Dec. 28, 2024

Language: Английский

Citations

1
Flood Classification and Improved Loss Function by Combining Deep Learning Models to Improve Water Level Prediction in a Small Mountain Watershed DOI

Rukai Wang,

Ximin Yuan,

Fuchang Tian

et al.

Published: Jan. 1, 2024

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Language: Английский

Citations

0
Enhancing Multivariate Time Series Forecasting: A Novel Approach with Mallows Model Averaging and Graph Neural Networks DOI Creative Commons
Haili Zhang, Jiawei Wang,

Zhaobo Liu

et al.

Journal of Systems Science and Complexity, Journal Year: 2024, Volume and Issue: unknown

Published: July 27, 2024

Multivariate time series forecasting holds substantial practical significance, facilitates precise predictions, and informs decision-making. The complexity of nonlinear relationships the presence higher-order features in multivariate data have sparked a burgeoning interest leveraging deep learning approaches for such tasks. Existing methods often use pre-scaled neural networks, whose reliability generalization can pose challenge. In this study, authors propose an instance-wise graph-based Mallows model averaging (IGMMA) framework prediction. incorporates module into network, where extracted are utilized as inputs candidate linear models. These models combined with weights to create new layer, forming novel graph network model. Moreover, loss function is modified based on criterion, penalties imposed parameters separately. proposed method predict multicommodity futures prices, empirical results show that IGMMA has superior predictive accuracy even when small networks used. This indicates significantly reduces required training, which enables training multiple alternative large

Language: Английский

Citations

0