Prediction of local scour depth of deep-water foundations using AI-based neural network

Scour is one of the primary causes of deep-water structure damage, and scour depth prediction is critical in practical engineering design. However, accurate prediction remains challenging due to numerous influencing factors and lack of comprehensive calculation formulas. To solve this problem, this study proposes an artificial neural network (ANN) model for local deep-water foundation scour depth prediction based on raw data from classic laboratory tests, and verifies its prediction performance and practical feasibility. An ANN model with four hidden layers is constructed via forward propagation, then optimized by the BP algorithm combined with the Adam optimizer. Optimal parameters are selected through training and validation on the validation set. Partial experimental data are used as the test set for prediction, and the model's effectiveness is verified by comparing predicted and actual values. Feature importance analysis is also conducted for the influencing factors. Results show that the model fits training samples well with an R² value over 0.90, indicating high accuracy. Despite limited test samples, the R² of predictions on the test set is close to 0.80, proving the model's validity and reliability. Moreover, similar feature importance weights of different parameters corroborate the validity of the original tests.