Evaluation of GMDH networks for prediction of local scour depth at bridge abutments in coarse sediments with thinly armored beds

Abstract

Protection of the bridge abutment in waterways against scour phenomena is a very significant issue in hydraulic engineering fields. Several field and experimental investigations were carried out to produce a relationship between the abutment scour depth due to thinly armored bed and the governing variables. However, existing empirical equations do not always provide accurate scour prediction due to the complexity of the scour process. In the present study, group method of data handling (GMDH) networks are utilized to predict abutments scour depth in thinly armored beds. GMDH network is developed using evolutionary and iterative algorithms included those of gravitational search algorithm (GSA), particle swarm optimization (PSO), and back propagation (BP). The sediment size properties, bridge abutments geometry, and approaching flow are considered as effective parameters on the abutment scour depth. Training and testing stages of the models are carried out using experimental data sets. Performances results for alternative GMDH networks are compared with those obtained using traditional equations. A sensitivity analysis is also performed to determine the most important parameter in predicting the abutment scour depth in thinly armored beds.