عنوان مقاله [English]
Identification of the scour pattern around bridge piers is one of the most important issues in optimized designing of such structures. In this study, for the first time, the scour pattern around twin bridge piers was predicted using an optimized hybrid algorithm. The hybrid algorithm (ANFIS-FA) was developed through combining the Adaptive Neuro-Fuzzy Inference System (ANFIS) and the Firefly algorithm (FA). The ANFIS system is often used with a Takagi-Sugeno-Kang (TSK) fuzzy system. The FA has the ability to find optimized values for non-linear problems (e.g. scour depth). This method has a more accurate search procedure than most optimization methods. Due to its high convergence rate and successful applications in various optimization problems, we decided to employ it for this research. Its high convergence rate causes to reduce computation volume as well as reaching to a convergent response in a small number of iterations. After that, four ANFIS and ANFIS-FA models were introduced by means of parameters affecting the scour depth around twin piers. In order to evaluate the accuracy of soft computing models, the Monte Carlo simulations were employed. In addition, the validation of the numerical models was carried out by the k-fold cross validation approach with k=5. In this study, the experimental data obtained by Wang et al. (2016) were used for validating the results of the numerical models. Their experimental model consists of a rectangular channel with a length of 12m, the width of 0.42 and the height of 0.7m. They installed two abutments to report the scour amount around them. It is worth noting that the initial depth of the sediment layer in this experimental study is 15cm, its length is 6m, and the twin abutments were placed with the distance d from each other in the middle of the sediment layer. Based on the modeling results, the analysis of the results indicated that
ANFIS-FA models are more accurate than ANFIS models. Then, the superior model was introduced through conducting a sensitivity analysis. The superior model is a function of all input parameters. This model estimated scour values with
reasonable accuracy. For example, the values of R2, MAPE and RMSE were
calculated 0.991, 5.876 and 0.015, respectively. Furthermore, the error distribution results showed that about 66\% of the superior model results have an error less than 5\%. Next, the Froude number was detected as the most effective input parameter for estimating the scour hole around twin bridge piers. Finally, by conducting an uncertainty analysis, it was concluded that the superior model has an overestimated performance.