Gene expression programming to predict local scour using laboratory and field data

Abstract

ABSTRACT The accurate estimation of scour is essential for the safe design of bridge crossing river. Scour is a complex phenomenon; developing a generalized scour model is a big challenge to the hydraulic community. An artificial intelligence (AI) based Gene Expression Programming (GEP) technique was employed to develop a unique and universal (applicable to laboratory and field conditions) scour model. Model parameters were selected after model-independent sensitivity analysis. The model developed in the present study was compared with the five most widely used empirical scour models. Jain and Fischer model found to have good correlation (0.94 and 0.76 for field and laboratory data, respectively) under classical methods of model development, but an error in the prediction was most (RMSE = 0.06 and 2.08 m for field and laboratory data, respectively) amongst other scour models. The GEP model developed in the present study has good correlation (0.93 and 0.76 for laboratory and field data, respectively) and the least error (RMSE = 0.042 m & 0.78 m for laboratory and field data, respectively) amongst all the models considered in the present study. It is concluded that the AI-based GEP scour model performed better and recommended for its use in the laboratory and the field conditions.