EVALUATION OF GENE EXPRESSION PROGRAMMING TO PREDICT THE LOCAL SCOUR DEPTH AROUND A BRIDGE PIER

One of the main reasons for bridge failure is the local scouring around the piers. In this way, the precise prediction of a permissible depth of scouring is pivotal to ensure safety and to keep successful maintenance. The main objective of the current study is to predict local scour around the piers by developing new empirical equations using two effective approaches, i.e., gene expression programming (GEP) and artificial neural networks (ANN). Various important parameters were used to derive the empirical equations such as pier shape, flow depth, flow intensity, pier width, and flow direction angle (attack angle). All these parameters were determined from the dimensional analysis of the problem. The two relationships were developed based on 729 data points from the numerical models, which were divided into two sets, training, and validation (test). Moreover, three statistical indexes (i.e., RMSE, R-2 and MAE) were used to identify the performance of the two approaches and their new empirical equations. The results of the comparison indicated that the ANN model (RMSE=0.102, R-2 = 0.94 and MAE=0.076) is performed better than the GEP model (RMSE= 0.124, R-2 =0.90 and MAE=0.103) slightly. The latter is preferred on account of its ability to produce explicit and compressed arithmetic expressions. Furthermore, the sensitivity analysis results show that the index of flow depth/width ratio (y/b) has the significant influence on local scour depth predictions compared to other input variables.