This study investigates sediment load prediction and generalization from laboratory scale to field scale using principle component analysis (PCA) in conjunction with data driven methods of artificial neural networks (ANNs) and genetic algorithms (GAs). Five main dimensionless parameters for total load are identified by using PCA. These parameters are used in the input vector of ANN for predicting total sediment loads. In addition, nonlinear equations are constructed, based upon the same identified dimensionless parameters. The optimal values of exponents and constants of the equations are obtained by the GA method. The performance of the so-developed ANN and GA based methods is evaluated using laboratory and field data. Results show that the expert methods (ANN and GA), calibrated with laboratory data, are capable of predicting total sediment load in field, thus showing their transferability. In addition, this study shows that the expert methods are not transferable for suspended load, perhaps due to insufficient laboratory data. Yet, these methods are able to predict suspended load in field, when trained with respective field data.
This study quantitatively investigates the generalization from laboratory scale to field scale using the soft computing (expert) and the empirical methods. Principal component analysis is utilized to form the input vector for the expert methods. Five main dimensionless parameters are used in the input vector of artificial neural networks (ANN), calibrated with laboratory data, to predict field total sediment loads. In addition, nonlinear equations are constructed based upon the same dimensionless parameters. The optimal values of the exponents and constants of the equations are obtained by the genetic algorithm (GA) method using the laboratory data. The performance of the sodeveloped ANN and GA based models are compared against the field data and those of the existing empirical methods, namely Bagnold, Ackers and White, and Van Rijn. The results show that ANN outperforms the empirical methods. The results also show that the expert models, calibrated with laboratory data, are capable of predicting field total loads and thus proving their transferability capability. The transferability is also investigated by a newly proposed equation which is based on the Bagnold approach. The optimal values of the coefficients of this equation are obtained by the GA.The performance of the proposed equation is found to be very efficient.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.