Predicting River Suspended Load Using Artificial Neural Network and Non-Dominant Genetic Sorting Algorithm

Information on soil erosion and sediment production, meteorological features, hydrological features of rivers such as discharge, as well as human factors, are often very complex, indefinite, and nonlinear. Therefore, the use of machine intelligence algorithms (such as machine learning algorithms) is a good option in simulating and predicting river water quality variables such as suspended load. The aim of the present study is to present a proposed method based on Multilayer Perceptron Artificial Neural Network (MLP-ANN) and Non-Dominant Sorting Genetic Algorithm (NSGA) for predicting suspended river load. In the proposed method, the NSGA was used to train the MLP using the error propagation method and determining the optimal weight for the neurons. In this study, the suspended load of Tilabad station located in Gorganrood river during the 1982-2015 years was used as a case study. The results showed that the proposed method has a higher correlation coefficient compared to MLP and the value of R2 was 0.6728 and 0.4372, respectively. The value of Root-Mean-Square Error (RMSE) in the proposed method and MLP based on back-propagation (BP) algorithm is 4.7225 and 8.548, respectively. Therefore, in the proposed method, the NSGA has caused a good improvement of the MLP. The NSE value in the proposed method and MLP based on BP algorithm is 0.4321 and 0.2941, respectively. The results showed that the proposed method had good accuracy in predicting the suspended load. The proposed method with BP training algorithm has a better performance compared to the descending and Bayesian gradient training algorithm.