Physics-informed hybrid model for scour evolution prediction around pile foundations under tidal currents

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(4)

Published: April 1, 2025

The local scour process around pile foundations under tidal currents exhibits complex nonlinear and nonstationary dynamic characteristics, primarily stemming from the intricate coupling relationship between levels, flow velocity, direction, evolution. In this paper, a novel hybrid machine learning (ML) framework (referred to as GVCBA) is proposed, which consists of grey wolf optimization (GWO), variational mode decomposition (VMD), convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), attention mechanism. By synergistically integrating physical mechanisms with deep learning, demonstrates significantly enhanced accuracy in predicting these spatiotemporal dynamics. Based on Buckingham Π theorem, feature input parameters (e.g., Froude number Fr, periodic parameter tsin) are constructed, explicitly embedding hydrodynamic periodicity into model space, effectively overcoming overfitting tendency traditional data-driven models. Verification using measured data sea-crossing bridge shows that GVCBA framework, through multi-scale decoupling, achieves collaborative modeling oscillations cumulative effects, root mean square error 0.001 60 coefficient determination (R2) 0.985 82 test set, reducing prediction errors by over 80% compared (support vector machine, extreme gradient boosting) benchmark architectures (recurrent its structure combined CNN). Additionally, sensitivity analysis reveals Fr tsin key factors influencing prediction. This provides new method for infrastructure environments, combining interpretability engineering applicability.

Language: Английский

Predicting scour depth in a meandering channel with spur dike: A comparative analysis of machine learning techniques

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(4)

Published: April 1, 2025

In this research, an assessment of scour depth prediction in meandering channels with spur dikes is made employing machine learning approaches. Efficient determination the therefore vital morphologic aspects and structural stability. The input parameters include sinuosity (S), dike locations (Ld), porosity (P) experimental data from sinusoidal flumes. Four models; Extreme Gradient Boosting (XGBoost) Particle Swarm Optimization (PSO) XGBoost-PSO, Random Forest (RF), k-Nearest Neighbors (k-NN), Decision Tree-Neural Network (DT-NN) were used compared. findings demonstrate R-value 0.995 case RF model while XGBoost-PSO gave second-best accuracy R = 0.988. results SHAP analysis illustrated that are significant factors affecting (Ds/Yn, Ds: depth, Yn: water depth) had moderate importance assigned to location. Kernel density plots further supported regarding error distribution consistency. Even though, both yielded better because hyperparameter tuning, k-NN DT-NN less precise outcomes specifically predicted for progressive hydraulic procedures. Taylor's diagram even revealed greater by RF. Hence, a proper selection appropriate models remains first step estimating sufficiently flood erosion control.

Language: Английский

Modeling the discharge coefficient of labyrinth sluice gates using hybrid support vector regression and metaheuristic algorithms

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(4)

Published: April 1, 2025

Gates and weirs are frequently used hydraulic structures employed for controlling water flow rates in irrigation drainage networks. Therefore, accurately estimating the discharge coefficient (Cd) is important precise measurement. The present study intelligent predictive models modeling Cd labyrinth sluice gates. For this purpose, key dimensionless parameters reliable experimental datasets were used. support vector regression (SVR) model was hybridized with particle swarm optimization (PSO) genetic algorithms (GA). statistical metrics graphical plots evaluated performance of generated models. Three commonly indicators, namely root mean square error (RMSE), absolute (MAE), determination (R2), quantitatively evaluating proposed SVR-PSO achieved lowest values RMSE (0.0287) MAE (0.0209) highest value R2 (0.9732), indicating that it more accurate than SVR-GA (RMSE = 0.0324, 0.0257, 0.9685) SVR 0.0575, 0.0468, 0.8958) on testing data. findings revealed hybrid methods standalone model. In addition, regarding objective function criterion (OBF), (OBF 0.0245) 0.0273) had lower OBF provided estimates compared to existing nonlinear regression-based formulas data-driven approaches. Finally, sensitivity SHapley Additive exPlanations (SHAP) analyses determined relative importance each input variable prediction Cd.

Language: Английский