Predicting stable alluvial channel profiles using emotional artificial neural networks

Accurate prediction of the cross-sectional profile of stable alluvial channels is very important for hydraulic engineers in the design and construction of natural channels in urban watersheds. This paper presents the first-time application of Emotional Artificial Neural Network (EANN) modeling to predict the stable cross-sectional geometry of alluvial channel profiles. EANN applies emotional stimulus parameters to improve the network's training process. The key input parameters of the EANN model include modified Froude number (Fr-m), dimensionless mean sediment size (d*) and dimensionless lateral distance from the channel centerline (x*), and the output (or target) is the dimensionless vertical boundary level of the channel cross section (y*). Several observational datasets are used with different hydraulic and geometric conditions to verify and validate the model. Feed forward neural network (FFNN) and Gene expression programming (GEP) models are also designed to make a better evaluation and comparison of EANN performance. Furthermore, the prediction accuracy of the new EANN model is compared with several popular existing relationships related to previous researchers. The results declared that the EANN model with low error values of correlation coefficient (R) equal to 0.99 and root mean square error (RMSE) equal to 0.00058 is more accurate than FFNN model (R = 0.76 and RMSE = 0.231), which represents a significant improvement in EANN performance compared to classical FFNN. It is found that the EANN model has the lowest RMSE compared to all previous models (RMSE = 0.13 for the best previous model related to Diplas, 1990). Uncertainty analysis is performed to evaluate the quantitative performance of the new EANN and existing models. The results indicate that EANN has the lowest Width of the Confidence Bounds (WCB) and the lowest Mean Error of Predictions (MEP) of +/- 0.00004 and -0.00041 respectively compared to FFNN and GEP and also previous models. Therefore, the new EANN model is proposed as a superior alternative to existing models for river engineering applications in the design of stable natural alluvial channels. (C) 2019 Elsevier B.V. All rights reserved.