ABSTRACT
Scouring, due to the interaction of fluid, pipeline and material of soil, is a complex phenomenon, numerous effective parameters are involved on the scouring phenomenon below the pipelines. In order to predict scour depth accurately, investigators have tried to predict scour depth by artificial intelligence models like support vector machines, radial basis function, genetic programming and group method of data handling. A prediction model of BP neural network is proposed for the pipeline scour depth in this paper, which weights and thresholds of the BP neural network were optimized by genetic algorithm. The scour depth was predicted by this model using nondimensional data sets that were collected from the existing research. The results of the prediction model were compared with empirical equations and BP neural network, which showed that GA-BP neural network predicted the scour depth with R=0.9437(Correlation coefficient), RMSE=0.1072(Root Mean Square Error, RMSE) and MAE=0.0788(Mean Square Error, MAE) compared with empirical equations (R=0.7605, RMSE=0.2662, MAE=0.1730 and R=0.7416, RMSE=0.2790, MAE=0.2015) and BP neural network (R=0.8281, RMSE=0.2323, MAE=0.1736) in live-bed conditions. Thus it can be seen GA-BP neural network has lower error and higher accuracy than the empirical equations and BP neural network. Results of the sensitivity analysis indicated that Fr is the most effective parameter for predicting the scouring depth below the pipeline. Therefore, the method proposed by this paper can be used to accurately predict the scour depth of submarine pipeline, and provide an important role to improve the stability of pipeline.
Pipelines are used to carry liquids, such as water, oil, and natural gas. These pipelines are basically embedded in the sea bed. When the flood occurs, the water flow oscillation caused by the shedding of wake vortex influence the stability of pipelines, which may cause local scour around the pipe. Therefore, the prediction of the scour depth around the pipeline is an important problem in submarine pipeline engineering. Aiming at the problem of pipeline scour prediction, the experimental research and numerical calculation are carried out(Brørs, 1999; Moncada-M et al, 1999; Chiew, 1990; Myrhaug et al, 2009; Çevik and Yüksel, 1999; Cheng et al, 2014). A large number of empirical equations have been obtained through previous studies. The main disadvantage of these traditional methods is that it is not accurate enough to predict the scour phenomenon. Therefore, artificial intelligence methods are propoesd for predicting this problems. For example, artificial neural networks (ANN), support vector machine (SVM), machine learning methods, adaptive neuro-fuzzy inference systems (ANFIS), genetic programming (GP), group method of data handling (GMDH) have been used to predict scour of underwater structures(Guven et al, 2008; Najafzadeh et al, 2014; Najafzadeh et al, 2015; Azamathulla et al, 2010; Azamathulla et al, 2011; Azamathulla et al, 2015; Etemad-Shahidi et al, 2011; TothE et al, 2011; Parsaie et al, 2019). Recently, the BP neural network optimized by genetic algorithm (GA-BP) are used to solve various problems in engineering field. In this paper, according to the scour influence factors, the BP neural network optimized by genetic algorithm (GA-BP) is used to establish the scour depth prediction model, which is introducing the genetic Algorithm to optimize the weights and thresholds of the network and taking advantage of the characteristics of GA global search, and the generalization performance of the neural network can be improved, so as to improve the prediction accuracy of the network model and provide another scientific method for the prediction of scour depth. The main purpose of this paper is to study the effectiveness of GA-BP model in predicting pipeline scour depth. Also, the results are compared with empirical equations and BP neural network.
