Advancing rapid urban flood prediction: a spatiotemporal deep learning approach with uneven rainfall and attention mechanism

Journal of Hydroinformatics, Journal Year: 2024, Volume and Issue: 26(6), P. 1409 - 1424

Published: May 28, 2024

ABSTRACT Urban floods pose a significant threat to human communities, making its prediction essential for comprehensive flood risk assessment and the formulation of effective resource allocation strategies. Data-driven deep learning approaches have gained traction in urban emergency prediction, addressing efficiency constraints physical models. However, spatial structure rainfall, which has profound influence on flooding, is often overlooked many investigations. In this study, we introduce novel model known as CRU-Net equipped with an attention mechanism predict inundation depths terrains based spatiotemporal rainfall patterns. This method utilizes eight topographic parameters related height waterlogging, combined data inputs model. Comparative evaluations between developed two other models, U-Net ResU-Net, reveal that adeptly interprets traits accurately estimates depths, emphasizing flood-vulnerable regions. The demonstrates exceptional accuracy, evidenced by root mean square error 0.054 m Nash–Sutcliffe 0.975. also predicts over 80% locations exceeding 0.3 m. Remarkably, delivers predictions 3 million grids 2.9 s, showcasing efficiency.

Language: Английский

---

Stability Assessment of Concrete Gravity Dams via Multifidelity Surrogate Models

Advances in Civil Engineering, Journal Year: 2025, Volume and Issue: 2025(1)

Published: Jan. 1, 2025

When many repetitions of an expensive or time‐consuming analysis are needed, simplified models usually adopted to reduce the cost. This is often case with gravity dams under seismic load, especially if geometry variation needs be considered. Deterministic important part preliminary analyses but generally leads overconservative designs. In recent years, researchers have studied potential machine learning techniques computational burden dam assessment. However, generating training dataset for a surrogate model based on high‐fidelity (HF) data can when large set uncertain parameters To address this issue, study proposes use multifidelity (MFS) models. method, datasets different levels fidelity combined generate highly accurate at lower illustrate this, behavior assessed by means HF nonlinear finite element that considers geometric, material, and uncertainties. addition, five (LF) samples The goodness fit time produce used identify combination optimizes MFS performance. results show including medium‐ low‐fidelity improves predictive performance reduces its burden. also generation selection best LF depend size dataset.

Language: Английский

---

Enhancing the accuracy and reliability of non-online metered demand allocation in water distribution systems through dynamic spatiotemporal population distribution

Water Research, Journal Year: 2025, Volume and Issue: 276, P. 123235 - 123235

Published: Feb. 3, 2025

Language: Английский

---

GATransformer: A Graph Attention Network-Based Transformer Model to Generate Explainable Attentions for Brain Tumor Detection

Algorithms, Journal Year: 2025, Volume and Issue: 18(2), P. 89 - 89

Published: Feb. 6, 2025

Brain tumors profoundly affect human health owing to their intricacy and the difficulties associated with early identification treatment. Precise diagnosis is essential for effective intervention; nevertheless, resemblance among tumor forms often complicates of brain types, particularly in stages. The latest deep learning systems offer very high classification accuracy but lack explainability help patients understand prediction process. GATransformer, a graph attention network (GAT)-based Transformer, uses mechanism, GAT, Transformer identify preserve key neural channels. channel module extracts deeper properties from weight-channel connections improve model representation. Integrating these elements results reduction size enhancement computing efficiency, while preserving adequate performance. proposed assessed using two publicly accessible datasets, FigShare Kaggle, cross-validated BraTS2019 BraTS2020 demonstrating explainability. Notably, GATransformer generates interpretable maps, visually highlighting regions aid clinical understanding medical imaging.

Language: Английский

---

Towards coordinated and robust real-time control: a decentralized approach for combined sewer overflow and urban flooding reduction based on multi-agent reinforcement learning

Water Research, Journal Year: 2022, Volume and Issue: 229, P. 119498 - 119498

Published: Dec. 15, 2022

Language: Английский

---

Shall we always use hydraulic models? A graph neural network metamodel for water system calibration and uncertainty assessment

Water Research, Journal Year: 2023, Volume and Issue: 242, P. 120264 - 120264

Published: June 24, 2023

Language: Английский

---

Research on surrogate models and optimization algorithms of compressor characteristic based on digital twins

Journal of Engineering Research, Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 1, 2024

For the digitization of turbocharger, prediction compressor working state is essential. How to build a model with accurate and less time-consuming premise studying turbochargers. As relationship between parameters obtained through experiments, it cannot be expressed by simple functional equations, so surrogate often used for fitting curve. Five models, Kriging model, Response Surface Methodology, Artificial Neural Networks, Radial Basis Function, Support vector machines, were fit regression characteristic curves. And four optimization algorithms, Particle Swarm Optimization, Genetic Algorithm, Gray Wolf algorithm, Firefly optimize model. A method construct hybrid proposed. The results show that influencing factors modeling pressure ratio efficiency at all speed groups confirmed; Different algorithms have different degrees five models; accuracy better than optimized single constructed can applied in digital twins system predict time achieve purpose rapid response. authors do not permission share data.

Language: Английский

---

A surrogate modeling method for distributed land surface hydrological models based on deep learning

Journal of Hydrology, Journal Year: 2023, Volume and Issue: 624, P. 129944 - 129944

Published: July 19, 2023

Language: Английский

---

A Temporal Fusion Transformer Model to Forecast Overflow from Sewer Manholes during Pluvial Flash Flood Events

Hydrology, Journal Year: 2024, Volume and Issue: 11(3), P. 41 - 41

Published: March 21, 2024

This study employs a temporal fusion transformer (TFT) for predicting overflow from sewer manholes during heavy rainfall events. The TFT utilised is capable of forecasting hydrographs at the manhole level and was tested on network with 975 manholes. As part investigations, compared to other deep learning architectures evaluate its predictive performance. In addition precipitation measurements forecasts, issue how additional consideration in as model inputs impacts forecast accuracy investigated. A varying number sensors different measurement signals were compared. results indicate high performance like long short-term memory (LSTM) or dual-stage attention-based recurrent neural (DA-RNN). Additionally, suggest that considering single measuring point outlet instead an entire yields better forecasts. One possible explanation correlation between measurements, which increases training complexity without adding much value.

Language: Английский

---

Towards transferable metamodels for water distribution systems with edge-based graph neural networks

Water Research, Journal Year: 2024, Volume and Issue: 261, P. 121933 - 121933

Published: June 20, 2024

Data-driven metamodels reproduce the input-output mapping of physics-based models while significantly reducing simulation times. Such techniques are widely used in design, control, and optimization water distribution systems. Recent research highlights potential based on Graph Neural Networks as they efficiently leverage graph-structured characteristics Furthermore, these possess inductive biases that facilitate generalization to unseen topologies. Transferable particularly advantageous for problems require an efficient evaluation many alternative layouts or when training data is scarce. However, transferability GNNs remains limited, due lack representation physical processes occur edge level, i.e. pipes. To address this limitation, our work introduces Edge-Based Networks, which extend set represent link-level more detail than traditional Networks. architecture theoretically related constraints mass conservation at junctions. verify approach, we test suitability edge-based network estimate pipe flowrates nodal pressures emulating steady-state EPANET simulations. We first compare effectiveness several benchmark systems against Then, explore by evaluating performance For each configuration, calculate model metrics, such coefficient determination speed-up with respect original numerical model. Our results show proposed method captures pipe-level accurately node-based models. When tested networks a similar demands, retains good up 0.98 0.95 predicted heads. Further developments could include simultaneous derivation flowrates.

Language: Английский

---