عنوان مقاله [English]
The phenomenon of scouring around bridge piers is one of the important issues in river engineering science. Scouring creates a trench around the bridge piers and ultimately destroys them. So far, many studies have been done in this regard and numerous relationships have been proposed to estimate the scour depth parameter. However, due to the influence of various parameters and uncertainty in the scouring phenomenon, existing relationships do not lead to accurate and comprehensive results. In the present study, using a series of experimental data, the efficiency of the Gaussian Process Regression (GPR) method was evaluated to estimate the scour depth of bridge piers in cohesive and grainy beds. Various models were developed and the impacts of hydraulic parameters were evaluated. The results proved the high efficiency of the applied method in the research in estimating the scour depth compared to the semi-empirical equations. It was observed that the defined models for cohesive soils are more successful in estimating the scour depth than grainy and non- cohesive soil. The best result for test series was obtained in the state of soil with clay and sand with the values of CC=0.952, DC=0.801, RMSE=0.132 and MAPE=15.23%; in clay soil state with the values of CC=0.889, DC=0.883, RMSE=0.054 and MAPE=11.82% and in non-cohesive soil state with the values of CC=0.988, DC=0.796, RMSE=0.191 and MAPE=18.21%. The results showed that soil moisture and clay soil density were effective in predicting scour depth. Based on the results of sensitivity analysis, in the state of soil with clay and sand moisture with mean absolute percentage error of 24.42%, in the state of clay soil shear stresses and the percentage of soils density with errors of 43.42% and 47.04%, respectively, and in non-cohesive soil state pier Froude number with error of 28.65% were respectively the most effective parameters in estimating the scour depth in cohesive and non-cohesive soils.