Comparing the accuracy of individual and combined application of genetic algorithm and least squares of support vector machine in estimating scour depth of simple bridge piers

One of the landscape management approaches is the construction of bridges along the rivers. On the other hand, the bridge scouring is a serious damage to river engineering as the main source of water and sustaining planet life. Accordingly, in this research, using field data, the accuracy of empirical methods, genetic algorithm (GA), least squares support vector machine (LSSVM), and combined method were compared in estimating scour depth of simple bridge piers. In the GA method, a number of empirical relationships were modified and the results of these modified relationships were compared with the measured scour values. In the LSSVM method, through the input of different independent parameters, model training was performed, and scour depth was predicted. In the combined method, using the LSSVM model from combining the results of different individual relations, the scour depth of the bridge piers was estimated. The results showed that modified relationships by genetic algorithm and LSSVM model have higher accuracy than empirical methods. Also, if only the parameters used in the empirical relationships are included as input parameters to the LSSVM model, the modified relationships have less error than the LSSVM model. The statistical evaluation criteria of RMSE, E, R2, and NSE for the best state of the combined method were 0.4 m, 49%, 0.88, and 0.58 respectively in the training stage and 0.52 m, 50%, 0.7, and 0.38 respectively in the test stage. In general, the combined method estimates scouring depth with higher accuracy than other methods.