Comprehensive approach for scour modelling using artificial intelligence
Praveen Rathod, Vivek L. Manekar
Abstract
Literature review revealed that scour modeling using artificial intelligence (AI) lacks majorly in two aspects - one is the Input variable selection (IVS) and secondly, the relative ranking of scour models based on their performance. In this study, state-of-the-art AI algorithms, including artificial neural network (ANN), artificial neuro-fuzzy interface system (ANFIS), support vector machine (SVM), model tree (M5P), gene expression programming (GEP), and group method of data handling (GMDH) were employed to compute the Local Scour Depth (LSD) and to address identified lacunas. A total of 378 data set consisting of laboratory and field data was used for modeling. Evaluation criteria such as the index of agreement (IOA), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Skill Score (SS), Correlation (Correl), and Taylor diagram were employed to check the goodness-of-fit of the proposed models. IVS was performed using partial mutual information (PMI). The findings of the study showed that velocity (V), the diameter of the pier (b), Reynolds pier number (Rp), and particle densimetric Froude number (Frd) were highly sensitive parameters to the scour. Quantitative and qualitative results indicated that SVM performed better with RMSE (0.06), MAPE (0.04) and SS (0.9). Tayler diagram also confirmed the above findings.