Energy Dissipation Prediction for Trapezoidal–Triangular Labyrinth Weirs Based on Soft Computing Techniques: A Comparison

Abstract

The present study aimed at determining the relative energy dissipation (ED) of trapezoidal-triangular labyrinth weirs (TTLWs) using soft computing methods such as neural networks, i.e., multilayer perceptron and radical basis function, support vector machine (SVM), and multivariate adaptive regression splines, using three different scenarios. Its performance was evaluated using a variety of performance indices. Observations indicate that TTLW typically expends the most energy because of the collisions of the nappes at the upstream apexes and the circulating flow in the pool formed behind the nappes. Furthermore, as the weir sidewall angle and height increase, the ED tends to decrease. The results of the models demonstrate that while all methods performed reasonably well in predicting the ED of TTLWs, the ANN-MLP and SVM models were more accurate. Specifically, the ANN-MLP model showed superior performance, with mean absolute percentage error, RMSE, DC, and R² values of 1.21, 0.009, 0.989, and 0.991, respectively, for the testing data set. The outcomes of the sensitivity analysis indicate that relative critical depth (y_c/E₀) and, after that, the angle of the LW wall (α) are the most effective factors in determining the TTLW relative ED in all methods. Overall, the comparison of model outcomes indicates that the ANN-MLP model is highly effective in predicting the ED of TTLWs.